JP4911052B2 - Image processing apparatus, image processing method, and image processing program - Google Patents

Description

  The present invention relates to an image processing device, an image processing method, and an image processing program, and more particularly, to an image processing device, an image processing method, and an image processing program for encoding image data.

  An example of an image processing method for encoding image data includes the following method.

  First, wavelet transform is performed on image data. Then, the coefficient (wavelet coefficient) obtained as a result of the wavelet transform is quantized to generate a frequency table indicating the frequency of occurrence of each value after quantization.

  Subsequently, a Huffman table is generated according to the frequency table, and each value after quantization is converted into a Huffman code according to the Huffman table.

  Image data can be encoded by the processing as described above.

  Patent Document 1 describes an image processing apparatus that suppresses the amount of code data to a target code amount or less. The image processing apparatus described in Patent Literature 1 performs quantization and data encoding, and counts the amount of encoded data while the encoding process proceeds. If the data amount of the encoded data exceeds the set upper limit value in the middle of the encoding process, the quantization process is changed and the encoding process is resumed. To do.

Japanese Patent Laying-Open No. 2004-320479 (paragraphs 0044 and 0045, FIG. 3)

  In image processing for encoding image data, the code amount of the code data needs to be equal to or less than the target code amount. Also in the image processing apparatus described in Patent Document 1, in order to keep the amount of code data below the target code amount, the quantization step is changed and the already encoded process is re-encoded. However, if encoded data that has already been encoded is re-executed from quantization while the encoding process is in progress, the processing load will increase.

  Further, although the image processing apparatus described in Patent Document 1 performs DCT conversion, even when wavelet conversion is performed, encoding is performed so that the code amount of encoded data is equal to or less than the target code amount with a small processing load. It is preferable to be able to perform.

  In addition, when encoding is repeated so as to satisfy the condition that the encoded data is equal to or less than the target code amount while changing the quantization step, the image quality is improved beyond the accuracy of the original orthogonal transform processing (wavelet or DCT). I can't.

  In general, in order to achieve a certain compression efficiency by encoding using wavelet transform, it is necessary to perform wavelet transform to a certain number of decomposition levels (generally 3 or more). However, if wavelet transform is repeated, it becomes difficult to maintain the accuracy of the coefficients, and no matter how fine the quantization step is after the wavelet transform, quantization will exceed the accuracy at the number of decomposition levels. The image quality cannot be restored. In order to solve this problem, it is conceivable to generate codes for a plurality of decomposition levels, and finally select the optimal number of decomposition levels. However, the processing load is increased to perform encoding with a plurality of decomposition levels. growing.

  Therefore, the present invention provides an encoding process for encoding so that the image quality of the encoded image data is improved as much as possible while making the encoding amount of the encoded image data equal to or less than the target code amount. An object is to provide an image processing apparatus, an image processing method, and an image processing program that can be realized with a load.

  The image processing apparatus of the present invention includes a wavelet transform unit that performs wavelet transform on image data up to a predetermined number of decomposition levels, a component other than the LL component equal to or less than the predetermined number of decomposition levels, and a plurality of types of decomposition levels Intermediate data for deriving intermediate data which is data quantized in a predetermined quantization step or a coefficient itself obtained by wavelet transformation from the LL components of the plurality of types of decomposition levels obtained by wavelet transformation in Generating frequency information indicating the frequency of generation of intermediate data of components other than the LL component equal to or less than the predetermined number of decomposition levels and the LL component of the plurality of types of decomposition levels; Among these, the minimum that satisfies the condition that the encoded code amount estimated based on the frequency information is not larger than the target code amount Characterized in that it comprises an encoding means for performing coding on decomposition level number of intermediate data.

  The image processing method of the present invention performs wavelet transform on image data up to a predetermined number of decomposition levels, and includes components other than the LL component equal to or less than the predetermined number of decomposition levels and a plurality of types of decomposition levels Deriving intermediate data that is data quantized in a predetermined quantization step or a coefficient itself obtained by wavelet transformation from the LL components of the plurality of types of decomposition levels obtained by wavelet transformation, Generate frequency information indicating the frequency of occurrence of intermediate data of components other than the LL component equal to or less than the number of decomposition levels and the LL component of the plurality of types of decomposition levels, and based on the frequency information of the plurality of types of decomposition levels Coding for intermediate data with the minimum number of decomposition levels that satisfies the condition that the estimated post-encoding code quantity is not larger than the target code quantity And performing.

  Further, the image processing program of the present invention allows a computer to perform wavelet transformation processing for performing wavelet transformation on image data up to a predetermined number of decomposition levels, components other than the LL component equal to or less than the predetermined number of decomposition levels, and a plurality of types Intermediate data that is data quantized in a predetermined quantization step from the LL components of the plurality of types of decomposition levels obtained by the wavelet transform with the number of decomposition levels of or a coefficient itself obtained by the wavelet transform Generating intermediate data deriving processing, and frequency information indicating the frequency of occurrence of intermediate data of components other than the LL component equal to or less than the predetermined number of decomposition levels and LL components of the plurality of types of decomposition levels, Of the number of decomposition levels, the encoded code amount estimated based on the frequency information is larger than the target code amount. Characterized in that to perform the encoding process for encoding to satisfy the minimum level of decomposition number of intermediate data called wards.

  According to the present invention, the encoding process for encoding the image data so that the image quality of the encoded image data is improved as much as possible while the encoding amount of the encoded image data is equal to or less than the target code amount is a low process. Can be realized with load.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing an example of an image processing apparatus of the present invention. In the following, a case where the image processing apparatus converts image data into a Huffman code will be described as an example. The image processing apparatus 1 of the present invention includes a wavelet transform unit 2, an intermediate code generation unit 3, a first determination unit 4, a second determination unit 5, a requantization unit 6, and an encoding unit 7. Is provided.

  The wavelet transform unit 2 receives image data to be encoded. The wavelet transform unit 2 performs wavelet transform on the image data up to a predetermined number of decomposition levels. This “predetermined number of decomposition levels” is denoted as “N”. In the drawings, the number of decomposition levels may be written as Rlevel. FIG. 2 is an explanatory diagram schematically showing the wavelet transform. The wavelet transform unit 2 performs wavelet transform on the input image data 100. The types of coefficients obtained by one wavelet transform include LL component, LH component, HL component, and HH component, and the wavelet transform unit 2 recursively performs wavelet transform on the LL component. Hereinafter, the LH component, the HL component, and the HH component are collectively referred to as xx. Further, the LL component and the xx component when the number of decomposition levels is k are referred to as “kLL component” and “kxx component”, respectively. FIG. 2 shows an xx component (1xx) having a decomposition level number 1, an LL component (2LL) and an xx component (2xx) having a decomposition level number 2. 2 illustrates wavelet transformation up to the number of decomposition levels “2”. However, the number of decomposition levels when the wavelet transform unit 2 performs wavelet transformation is not limited to 2, and the number of decomposition levels is 3 or more. Wavelet transformation may be performed up to. In the following description, an example will be described in which wavelet transformation is performed so that coefficients after wavelet transformation (wavelet coefficients) are expressed in YUV format and have Y, U, and V values.

  The intermediate code generation unit 3 is intermediate data (hereinafter referred to as intermediate code) that is a value obtained by quantizing a coefficient obtained by the wavelet transform in a predetermined quantization step or a coefficient itself obtained by the wavelet transform. .) Is generated. Here, a case where the result obtained by quantizing the coefficient obtained by the wavelet transform is an intermediate code will be described as an example. Therefore, in this example, the intermediate code generation unit 3 quantizes the values of Y, U, and V included in the wavelet coefficients in predetermined quantization steps, and sets the quantization result as an intermediate code. Hereinafter, the quantization step is referred to as Q step.

  The intermediate code generation processing by the intermediate code generation unit 3 will be described in more detail. Predetermined Q steps and dead zones at the time of intermediate code generation are determined for each intermediate code generation target component such as a 1xx component and a 2xx component. The intermediate code generation unit 3 quantizes the value of image data whose absolute value is equal to or less than a predetermined dead zone to 0. For example, if the absolute value of a certain Y included in the 1xx component is equal to or less than the dead zone, the Y is quantized to 0. Further, the intermediate code generation unit 3 divides a range in which the absolute value exceeds a predetermined dead zone every predetermined Q steps. Then, the values of the image data belonging to each of these ranges are quantized to the median value of the ranges. For example, it is assumed that the dead zone and the Q step are both 2. In this case, a range in which the absolute value exceeds 2 (that is, a range of less than −2 and a range of greater than 2) is divided into two Q steps, “−4 to −2”, “2 to 4”, “ Define a range such as 4-6 ″. If a certain Y included in the 1xx component belongs to a range such as “2 to 4”, the value of Y is quantized to the median value of the range. Here, Y included in the 1xx component is illustrated, but U and V quantization is performed in the same manner. The same applies to the quantization processing for components other than the 1xx component.

  The intermediate code generation unit 3 may include the result of the run length analysis in the intermediate code. That is, the result of the run length analysis may also be an intermediate code (intermediate data). The run length analysis is a process of counting the number of consecutive data when the same data is continuously arranged. As an aspect of the run length analysis, there is an aspect in which the number of consecutive pixels is counted when pixels whose Y, U, and V values after quantization are all 0 are consecutive. Alternatively, when pixels having the same combination of Y, U, and V after quantization are continuous, the number of consecutive pixels may be counted. FIG. 3 is an explanatory diagram schematically showing an intermediate code generated from wavelet coefficients. In FIG. 3, the continuous number counted by the run length analysis is expressed as “len”. For example, the intermediate code generation unit 3 quantizes Y, U, and V included in the 2LL component, the 2xx component, and the 1xx component as described above, and generates a 2LL intermediate code, a 2xx intermediate code, a 1xx intermediate code, and the like. Alternatively, the run length analysis may be performed on the quantized Y, U, and V, and the result (the counted number of consecutive “len”) may be included in the intermediate code. Hereinafter, a case where len is included in the intermediate code will be described as an example.

In addition, the number of decomposition levels (m _i ) that defines the highest image quality of an image obtained from encoded data is set in the image processing apparatus. Here, _mi is an integer of 1 or more and N or less. The intermediate code generation unit 3 generates an intermediate code for the LL component having the decomposition level number m _{i to} N and the xx component having the decomposition level number 1 to N. In other words, an intermediate code is generated for (m _i ) LL component, (m _i +1) LL component,..., NLL component, 1xx component, 2xx component,.

  The first determination unit 4 performs a process of determining whether or not to encode the xx component (that is, the component other than the LL component) in the intermediate data of the number of decomposition levels in a predetermined range.

The first determination unit 4, in order to make this determination, (xx components of the LL component of the decomposition level number m _i to N, and the division level number 1 to N) generates subject to the above respective components of the intermediate code for each In addition, the frequency (number of occurrences) of intermediate codes in each range used for quantization is counted. Specifically, the first determination unit 4 counts the occurrence frequency of data quantized in a range where the absolute value is equal to or less than a predetermined dead zone for each of Y, U, and V. Further, the frequency of occurrence of data quantized for each range obtained by dividing the range in which the absolute value exceeds the dead zone for each predetermined Q step is counted for each of Y, U, and V.

  FIG. 4 is an explanatory diagram showing an example of information (hereinafter referred to as a frequency table) indicating the result of counting the frequency of occurrence of quantized data in each range. FIG. 4 shows an example of the frequency table of Y and U of the 2LL intermediate code. The left column of each frequency table shown in FIG. 4 shows the quantized result, and the right column shows the occurrence frequency of data quantized to that value (Y or U in the example of FIG. 4). Yes. That is, in the example of Y of the 2LL intermediate code illustrated in FIG. 4, there are 230 pieces of Y data whose absolute value belongs to the range below the dead zone and is quantized to 0, and belongs to the range whose central value is 255. Indicates that there are 44 Y data quantized.

  The first determination unit 4 generates a Huffman table from each frequency table. FIG. 5 is an explanatory diagram illustrating an example of a Huffman table. FIG. 5 illustrates a Huffman table for Y of the 2LL intermediate code. The first determination unit 4 generates a Huffman table as illustrated in FIG. 5 from the Y frequency table of the 2LL intermediate code. Similarly, Huffman tables are also generated from other frequency tables. As illustrated in FIG. 5, the Huffman table includes each quantized value (0 to 255 in the example illustrated in FIG. 5) and a code converted from the value. The first determination unit 4 determines the code so that the quantized value and the code correspond one-to-one and the bit frequency of the code becomes shorter as the occurrence frequency increases, and the quantized value and the code To generate a Huffman table. By generating the Huffman code, the bit length of the Huffman code for each quantized data value (0 to 255 in the example shown in FIG. 4) is determined. For each Huffman code, the first determination unit 4 obtains the product of the bit length of the Huffman code and the corresponding occurrence frequency, and calculates the sum. If the len is included in the intermediate code, the bit length when each len is encoded (for example, the bit length when gamma-coded) is also added to the above sum. The first determination unit 4 calculates the sum of these xx components from the 1xx component to the (N-1) xx component and the (N-1) LL component. When the total is larger than the threshold value that is the target code amount of the code amount, the first determination unit 4 selects the number of decomposition levels in a predetermined range in descending order, and the intermediate data of the selected number of decomposition levels A data amount when the xx component is encoded is estimated, and whether or not the xx component of the intermediate data is encoded is determined according to the estimated data amount. The operation flow of this determination will be described later with reference to FIGS.

The second determination unit 5 determines whether or not to re-quantize the xx component (component other than the LL component) of the intermediate code of each decomposition level number smaller than the number of decomposition levels in the predetermined range. Judging. Let m _{q be} the maximum value of the number of decomposition levels for which it is determined whether or not to perform requantization. In this case, the number of decomposition levels in the above defined range is the number of decomposition levels in the range of m _q +1 to N.

When the second determination unit 5 determines that the requantization unit 6 performs the requantization, the requantization unit 6 determines each decomposition level number (1 to 1) smaller than the number of decomposition levels in a predetermined range (that is, m _q +1 to N) The xx component of the intermediate code of m _q ) is requantized.

  The encoding unit 7 encodes the value requantized from the intermediate code by the intermediate code and the requantization unit 6 into a Huffman code. Further, when len (continuous number counted by continuous length analysis) is included in the intermediate code, the len is also encoded. In this case, the encoding unit 7 may encode len into a gamma code, for example. If the bit length when len is expressed in binary number is l, the encoding unit 7 converts len to gamma code by adding “l−1” 0s to the head of the binary bit. To do. The encoding unit 7 converts the intermediate code or the re-quantization result into a Huffman code, and converts len into, for example, a gamma code, whereby the input image data is encoded.

  FIG. 6 is an explanatory diagram schematically illustrating an example of the result of Huffman coding. FIG. 6 illustrates a case where encoding is performed on the 2LL component, the 2xx component, and the 1xx component. When the data before encoding includes len as well as Y, U, and V, the encoded data of Y, U, V, and len are included in the encoding result of the 2LL component or the like.

  The wavelet transform unit 2, the intermediate code generation unit 3, the first determination unit 4, the second determination unit 5, the requantization unit 6, and the encoding unit 7 are realized by, for example, a CPU that operates according to a program.

  Next, the operation will be described. 7 and 8 are flowcharts showing an example of the processing progress of the present invention. Note that “output” described in the flowchart represents “encoding processing”.

  First, the wavelet transform unit 2 performs wavelet transform on the input image data up to the decomposition level number N.

Then, the intermediate code generation unit 3 generates an intermediate code with respect to the LL component having the decomposition level number m _{i to} N and the xx component having the decomposition level number 1 to N. That is, the intermediate code generation unit 3 quantizes Y, U, and V included in these components according to a dead zone and a Q step that are determined according to the type of component. As already described, the intermediate code generation unit 3 quantizes Y, U, and V whose absolute values are equal to or less than a predetermined dead zone to zero. Further, a range where the absolute value exceeds a predetermined dead zone is divided every Q steps, and Y, U, and V belonging to each range are quantized to a median value of the range. Further, for example, when pixels having the same combination of Y, U, and V are consecutive, the number of consecutive (len) is counted. The intermediate code generation unit 3 stores the Y, U, and V quantization results and len as intermediate codes in a storage device (not shown) included in the image processing apparatus.

In addition, the first determination unit 4 targets each of Y, U, and V for the intermediate code of the decomposition level number m _{i to} N that generated the intermediate code and the intermediate code of the decomposition level number 1 to N of the xx component. A frequency table is generated and stored in a storage device (not shown) included in the image processing apparatus. The first determination unit 4 may generate the frequency table by counting the occurrence frequency in each range determined by the dead zone and the Q step for each of Y, U, and V included in each component (the above step). S1).

Next, the first determination unit 4 substitutes m _i (the number of decomposition levels that defines the highest image quality of the image obtained from the encoded data) into the variable m (step S2). M set in step S2 is used for the control in steps S3 to S7.

  After step S2, the first determination unit 4 determines whether m <N is satisfied (step S3). If the variable m is less than N (Y in step S3), the first determination unit 4 determines the LL component (mLL component) with the decomposition level number m and the xx component (1xx to mxx component) with the decomposition level number 1 to m. The code amount when the intermediate code is encoded is estimated. At this time, the first determination unit 4 generates a Huffman table (see FIG. 5) from the frequency tables of Y, U, and V of these components, and for each Huffman code, the bit length of the Huffman code, Multiply the corresponding occurrence frequency of Y, U, or V, and calculate the sum of the multiplication results. Furthermore, in this example, since len is included in the intermediate code, the code amount when each len included in the intermediate code is converted into a code (for example, a gamma code) is also added to the sum. The 1st determination part 4 calculates | requires this sum for every component, and calculates those sum total (step S4). This total value is an estimated code amount when the intermediate code of the mLL component and the 1xx to mxx components is encoded. The estimated code amount is a value estimated as the data amount after encoding the intermediate code.

  Subsequently, the first determination unit 4 determines whether or not the estimated code amount is larger than the target code amount (step S5). The target code amount represents a maximum allowable value of the code amount when image data is encoded, and is used as a threshold value in steps S5, S13, and S19. Note that the target code amount may be appropriately set by the image processing apparatus 1 according to the situation such as communication of image data by the image processing apparatus itself. Alternatively, the target code amount may be input from the outside.

  When the estimated code amount is less than or equal to the target code amount (N in step S5), the process proceeds to step S9. In step S9, the encoding unit 7 converts the MLL intermediate code into a Huffman code. The encoding unit 7 may convert the YL, U, and V of the mLL intermediate code into corresponding Huffman codes using the Huffman table corresponding to Y, U, and V of the mLL component generated in step S4 (step S4). S9). Further, the encoding unit 7 also converts intermediate codes of components mxx, (m−1) xx,..., 1xx into Huffman codes (step S10). The intermediate codes Y, U, and V from mxx to 1xx may be converted into corresponding Huffman codes using the Huffman table generated in step S4. The encoding unit 7 encodes len in each intermediate code into, for example, a gamma code in steps S9 and S10.

  When the estimated code amount is larger than the target code amount (Y in Step S5), the first determination unit 4 increments the value of the variable m by 1 (Step S7), and repeats the processes after Step S3. When the variable m is incremented by 1 and the process proceeds from step S3 to step S4 again, it is not necessary to obtain the code amount again for the component for which the code amount has been obtained in the previous step S4, and the mLL component indicated by m after the increment And the mxx component are obtained, and the total code amount of the mLL component and the 1xx to mxx components may be estimated.

  If it is determined in step S3 that m = N (N in step S3), requantization processing (step S8) is performed. The process after step S11 shown in FIG. 8 represents the requantization process of step S8. That is, if m = N in step S3, the process proceeds to step S11 (see FIG. 8).

  The transition to step S11 means that the code estimation amount when the LL component of the intermediate code of the decomposition level number “N−1” and the xx component of the intermediate code from the decomposition level number 1 to “N−1” is encoded. Is larger than the target code amount. At this time, the first determination unit 4 causes the encoding unit 7 to encode the NLL component (step S11). The encoding unit 7 generates a Huffman table from the frequency tables of Y, U, and V of the NLL intermediate code, and converts the Y, U, and V into corresponding Huffman codes. The encoding unit 7 encodes each len included in the NLL intermediate code into, for example, a gamma code.

  Next, the 1st determination part 4 substitutes N to the variable m (step S12). M set in step S12 is used for the control in steps S13 to S21.

After step S12, the first determination unit 4 determines whether m ≦ m _q is _satisfied (step S13). If m ≦ m _q does not hold (N in step S13), the code amount of the xx component (that is, the mxx component) of the number of decomposition levels indicated by the variable m is estimated (step S14). In step S14, a Huffman table is generated from the frequency tables of Y, U, and V of the mxx component, and for each Huffman code, the bit length of the Huffman code is multiplied by the occurrence frequency of the corresponding Y, U, or V, The sum of the multiplication results is calculated. The bit length when each len is encoded may be added to the sum. If the estimated code amount of the mxx component has been calculated in step S4, the estimated code amount may be referred to.

  Subsequently, the first determination unit 4 determines whether or not the sum of the code estimation amount obtained in step S14 and the data amount of already encoded code data is larger than the target code amount ( Step S15). If the sum is larger than the target code amount (Y in step S15), the process is terminated. If the total is equal to or less than the target code amount (N in step S15), the encoding unit 7 encodes the mxx component (step S16). The encoding unit 7 converts Y, U, and V of mxx components into corresponding Huffman codes using the Huffman table corresponding to Y, U, and V of mxx components used for code amount estimation in step S14. The encoding unit 7 encodes len in the mxx component into, for example, a gamma code.

Subsequently, the first determination unit 4 subtracts 1 from the value of the variable m (step S17), and repeats the processing after step S13. By repeating the loop processing from step S13 to step S17, the first determination unit 4 selects the number of decomposition levels in a predetermined range (that is, m _q +1 to N) in descending order every time the process proceeds to step S14. Then, the code amount of the xx component is estimated.

If it is determined in step S17 that the value of the variable m is subtracted by 1 and m ≦ _mq is determined in step S13 (Y in step S13), the process proceeds to step S18. Therefore, after step S18, the value of m is _mq .

  In step S18, the second determination unit 8 encodes the intermediate code of each xx component from the xx component (that is, the mxx component) of the decomposition level number indicated by m at the time of transition to step S18 to the 1xx component. Estimate the amount. When the process first proceeds to step S18 (that is, when the requantization of step S20 has not been performed yet), the code amount when the intermediate code of each component of mxx to 1xx is encoded is estimated. When the process proceeds to step S18 after the requantization process in step S20, the code amount of the intermediate code of each component of mxx to 1xx after the requantization is estimated.

  When estimating the code amount of the intermediate code of each component, a Huffman table is generated from the frequency tables of Y, U, and V of each component, and for each Huffman code, the bit length of the Huffman code and the corresponding Y , U or V occurrence frequency, the sum of the multiplication results is calculated, and the bit length when each len is encoded may be added to the sum.

  In the requantization in step S20, a frequency table of Y, U, and V after requantization is also created. When the process proceeds from step S20 to step S18, a Huffman table may be generated using the frequency table.

  After step S18, the second determination unit 8 determines that the sum of the code amount obtained by encoding mxx to 1xx obtained in step S18 and the data amount of the already encoded code data is larger than the target code amount. It is determined whether or not it is larger (step S19).

  If the sum is larger than the target code amount, the second determination unit 8 determines to perform re-quantization, and proceeds to step S20 (Y in step S19).

  If the total is equal to or less than the target code amount, the second determination unit 8 determines not to re-quantize, and proceeds to step S21 (N in step S19).

  In step S21, the encoding unit 7 converts Yxx, Uxx, and 1xx components from mxx components into corresponding Huffman codes using the Huffman table used for code amount estimation in step S18. The encoding unit 7 encodes len in each component into, for example, a gamma code.

In step S20, the requantization unit 6 requantizes Y, U, and V from the 1xx intermediate code to the mxx intermediate code. Since m in step S18 and thereafter is m _q , Y, U, and V from the 1xx intermediate code to the m _q xx intermediate code are requantized.

In step S20, the requantization unit 6 determines a new dead zone for each component from 1xx to m _q xx. At this time, the re-quantization unit 6 sets dz + h · qs as a new dead, assuming that h is an integer equal to or larger than 0 and the dead zone and Q step used when the intermediate code is generated in step S1 are dz and qs, respectively. Set in the zone. Then, the requantization unit 6 requantizes the value of the intermediate code (Y, U, V) whose absolute value is equal to or less than the new dead zone to 0.

Further, the requantization unit 6 determines a new Q step for each component from 1xx to m _q xx. At this time, it is preferable that the requantization unit 6 determines qs · (2n + 1) as a new Q step using the above qs, where n is an integer of 0 or more. Then, the range in which the absolute value exceeds the new dead zone is divided for each new Q step (qs · (2n + 1)) to determine the range. Then, the intermediate codes (Y, U, V) of each component from 1xx to m _q xx belonging to each range are quantized to the median value of the range. The reason why it is preferable to set the new Q step to qs · (2n + 1) is that the Q value is determined in this way, so that the median value of the newly determined range is the center of the range used when quantizing in step S1. This is because the image quality of the encoded data can be prevented from deteriorating because it matches the value. The new range in which the Q step is defined as qs · (2n + 1) is a range obtained by combining the ranges used when the step S1 is quantized.

  However, if the Q step for deriving the intermediate code in step S1 is 1, an arbitrary value (however, an integer of 1 or more) may be set as a new Q step in step S20. If the Q step in step S1 is 1, the median value of the newly defined range matches the median value of the range used when quantizing in step S1, regardless of the value of the newly defined Q step. It is.

  In step S20, the second determination unit 5 generates new Y, U, and V frequency tables obtained by requantizing the intermediate code. The ranges used for requantization are individual ranges determined by dividing an absolute value of the dead zone or less by a new Q step. Each of these ranges is a combination of the ranges used when generating the intermediate code in step S1, and corresponds to the range used for quantization in step S1. Then, the frequency table in each range (for example, a range to be quantized to 0, a range to be quantized to 1, etc.) in step S1 has already been derived in step S1. Therefore, the second determination unit 5 identifies each range used in step S1 corresponding to each range defined for quantization in step S20, and newly calculates the sum of the occurrence frequencies in each range. The occurrence frequency may be within a certain range. By determining the occurrence frequency in this way, it is not necessary to recount the occurrence frequencies of Y, U, and V quantized in a new range.

  FIG. 9 is an explanatory diagram schematically showing frequency table derivation in step S20. The upper histogram shown in FIG. 9 indicates the frequency of occurrence of data quantized to the median value of each range (for example, Y) in the quantization for generating the intermediate code in step S1. Further, the lower histogram shown in FIG. 9 indicates the frequency of occurrence of data quantized to the median value of each range in the requantization in step S20. In FIG. 9, A to I and S to U are examples of median values after quantization. In the requantization process of step S20, as a result of the requantization unit 6 dividing the range from the new Q step and the dead zone, three ranges having A, B, and C as median values are combined, and S is It is assumed that one range as the median value is defined and data (Y) belonging to the range is quantized to the median value S. In this case, the second determination unit 5 determines the number of data quantized to the value A, the number of data quantized to the value B, and the quantum to the value C in the frequency table represented as the upper histogram in FIG. The sum of the number of converted data may be calculated, and the frequency of occurrence of data that belongs to a new range (a range with S as the median) corresponding to the three ranges and is quantized to the value S may be calculated. The occurrence frequency in other ranges may be calculated similarly. By calculating the occurrence frequency after requantization in this way, it is not necessary to recount the occurrence frequency of data in each range, so that the processing can be speeded up.

  The same applies to the occurrence frequency of data in other ranges. As described above, if the new dead zone is dz + h · qs and the new Q step is qs · (2n + 1), the median in the new range is the median in the quantization at the time of intermediate code generation. Will be equal. For example, in the example shown in FIG. 9, S = B, T = E, and U = H.

  After the above step S20, the processes after step S18 are repeated. The second determination unit 5 newly generates a frequency table in step S20. In the subsequent step S18, the second determination unit 5 uses the frequency table to generate a Huffman table indicating the sign of the value quantized in step S20. At this time, the second determination unit 5 uses each value after quantization in step S1 (that is, each intermediate code generated in step S1) as a new code in the Huffman table generated from each new frequency table. Should be associated with.

  FIG. 10 is an explanatory diagram showing the association between the value of the intermediate code and the code derived after requantization. For example, it is assumed that the intermediate code −1, 0, 1 before re-quantization illustrated in FIG. 10 is re-quantized to 0. In this case, the second determination unit 5 derives a Huffman table from the frequency table of each value such as 0 after requantization. In this Huffman table, it is assumed that the code corresponding to the intermediate code 0 is 0. Then, the second determination unit 5 associates a new code 0 with each of the original intermediate codes −1, 0, 1. For example, each code “100”, “0”, “101” corresponding to the original intermediate code-1, 0, 1 is replaced with a new code “0”. By processing in this way, the same code as 0 can be referred to from the original intermediate codes -1,1. Then, a Huffman code corresponding to the value after requantization can be generated without rewriting the intermediate code.

  By performing the requantization process, the number of bits of the Huffman code can be reduced. Therefore, the total code amount after encoding can be reduced by repeating the loop processing of steps S18 to S20. Then, when the estimated total code amount becomes equal to or less than the target code amount, the process proceeds to step S21 and the encoding process may be performed.

  In addition, as a result of increasing the dead zone value in the requantization process in step S20, all intermediate codes (Y, U, V) may be quantized to one value (0). In this case, the encoding unit 7 may prohibit the encoding process in step S21. That is, when all intermediate codes are quantized to 0 by requantization, the encoding unit 7 terminates the process without converting the Y, U, and V values after requantization into Huffman codes. It's okay. Also, the process of encoding len need not be executed. As described above, when all of Y, U, and V are quantized to 0, the processing amount can be reduced by not performing the encoding process.

According to the present invention, when the first determination unit 4 determines to encode the xx component for the number of decomposition levels (m _q +1 to N) in a predetermined range, the xx component is encoded. deep. Then, the intermediate code of the decomposition level number (1 to m _q ) is requantized according to the code amount when the decomposition level number (1 to m _q ) lower than the decomposition level number is encoded. Therefore, since re-quantization is performed for a part of the number of decomposition levels obtained by the wavelet transform, it is possible to control the code amount to be equal to or less than the target code amount while suppressing the processing load.

  In the requantization process, a range obtained by combining the dead zone used when deriving the intermediate code and the range determined from the Q step is determined, and the value in each new range is quantized to the median value of the range. In this way, since the new range is obtained as a range that combines the range at the time of deriving the intermediate code, when obtaining a new frequency table, the sum of the results counted at the time of deriving the intermediate code may be obtained (see FIG. 9). . Therefore, the processing load can be suppressed and the processing speed can be increased.

  In addition, by defining the Q step in the requantization process as qs · (2n + 1), the value after requantization can be made to match the value of the intermediate code, and deterioration of the image quality of the encoded data can be prevented. it can.

  Further, in the processing shown in FIGS. 7 and 8, the target code amount may be updated. Even in such a case, by performing processing using the updated target code amount as a threshold, it is possible to control the encoded data amount to be equal to or less than the updated target code amount.

In the above description, the number of decomposition levels at which the wavelet transform unit 2 performs wavelet transform is N, and the maximum value of the number of decomposition levels to be determined whether requantization is performed is described as m _q . Hereinafter, the operation of the present invention will be described using the case where N = 3 and m _q = 2 as an example. FIG. 11 and FIG. 12 are flowcharts showing an example of processing progress when N = 3 and m _q = 2. The example shown below corresponds to the case where m _i = 2, but corresponds to the loop process of steps 3 to S7 (see FIG. 7) when N = 3, m _q = 2 and m _i = 2. process of may be carried out once, in the example shown below, the process of substituting m _i to the variable m (process equivalent to step S2) such good without.

  First, the wavelet transform unit 2 performs wavelet transform on the input image data up to the decomposition level number 3.

  Then, the intermediate code generation unit 3 targets the LL components (2LL component, 3LL component) having the decomposition levels 2 and 3 and the xx components (1xx component, 2xx component, 3xx component) having 1 to 3 decomposition levels. An intermediate code is generated. That is, the intermediate code generation unit 3 quantizes Y, U, and V included in these components according to a dead zone and a Q step that are determined according to the type of component. FIG. 13 is an explanatory diagram illustrating an example of a dead zone and a Q step. Further, dead zone 1 and Q step 1 shown in FIG. 13 are examples of the dead zone and Q step for 2xx components. Dead zone 2 and Q step 2 are examples of the dead zone and Q step for 1xx components. The quantization granularity shown in FIG. 13 indicates the degree of quantization, and the larger the quantization granularity value, the larger the dead zone and Q step. In this example, it is assumed that the image processing apparatus holds the dead zone and the Q step of each quantization granularity exemplified in FIG. When an intermediate code is first generated, quantization is performed with a combination of a dead zone having a minimum quantization granularity “0” and a Q step. The operation of quantizing Y, U, and V according to the dead zone and the Q step is the same as that in step S1. For example, when pixels having the same combination of Y, U, and V are consecutive, the intermediate code generation unit 3 counts the number of consecutive (len) and includes len in the intermediate code.

  A dead zone having a quantization granularity of 1 or more and a Q step are used in the requantization process.

  In addition, the first determination unit 4 uses the frequency of Y, U, and V for the LL component with the decomposition level number 2 and 3 and the intermediate code with the xx component with the decomposition level number 1 to 3 for generating the intermediate code. A table is generated (step S31).

  With the decomposition level numbers 2 and 3, the decomposition level number 2 has less block distortion, and the gradation image quality is slightly better. Therefore, when the code amount of the decomposition level number 2 is equal to or less than the target code amount, in step S31, as described above, the decomposition level numbers 2, 3 are set so that the encoding can be performed only with the intermediate code of the decomposition level number 2. An intermediate code and a frequency table for each LL component are generated in advance. Although the image quality can be further improved with the decomposition level number 1 compared with the encoding with the decomposition level number 2, the compression efficiency is low, and the memory capacity for the frequency table and the intermediate code is also reduced. The code and its frequency table are not created.

  After step S31, the first determination unit 4 encodes the intermediate code of the decomposition level number 2 LL component (2LL component) and the decomposition level number 1 and 2 xx components (1xx component, 2xx component). The code amount is estimated (step S32). The first determination unit 4 generates a Huffman table from the frequency tables of Y, U, and V of these components, and for each Huffman code, the bit length of the Huffman code and the corresponding Y, U, or V Multiply the occurrence frequency and calculate the sum of the multiplication results. Further, the code amount when len included in the intermediate code is converted into a code (for example, a gamma code) is also added to the sum. The 1st determination part 4 should just calculate this sum for every component of 2LL, 2xx, and 1xx, and should just calculate those sum total. This total value is an estimated code amount when the intermediate code of the 2LL component, the 2xx component, and the 1xx component is encoded.

  Subsequently, the first determination unit 4 determines whether or not the estimated code amount is larger than the target code amount (step S33). The target code amount is used as a threshold value in steps S33, S43, and S46.

  When the estimated code amount is equal to or less than the target code amount (N in step S33), the process proceeds to step S34. In step S34, the encoding unit 7 converts the 2LL intermediate code into a Huffman code. That is, using the Huffman table corresponding to Y, U, and V of the 2LL component generated in step S32, the YLL, U, and V of the 2LL intermediate code may be converted into corresponding Huffman codes (step S34). Furthermore, the encoding unit 7 also converts the 2xx intermediate code and the 1xx intermediate code into Huffman codes (step S35). The 2xx intermediate code and the 1xx intermediate code may be converted into corresponding Huffman codes using the Huffman table generated in step S32. The encoding unit 7 encodes len in each intermediate code into, for example, a gamma code in steps S34 and S35.

  If the estimated code amount is larger than the target code amount (Y in step S33), requantization processing (step S36) is performed. The process after step S41 shown in FIG. 12 represents the requantization process of step S36. That is, if the estimated code amount is larger than the target code amount, the process proceeds to step S41 (see FIG. 12).

  In this example, N = 3, and the process proceeds to step S41. This means that the LL component (2LL) of the intermediate code with the decomposition level number “N−1” and the intermediate between the decomposition level number 1 and “N−1”. The code estimation amount when the xx component (1xx, 2xx) of the code is encoded is larger than the target code amount. At this time, the first estimation unit 4 causes the encoding unit 7 to encode the NLL component (that is, 3LL) component (step S41). The encoding unit 7 generates a Huffman table from the frequency tables of Y, U, and V of the 3LL intermediate code, and converts the Y, U, and V into corresponding Huffman codes. Further, each len included in the 3LL intermediate code is encoded into, for example, a gamma code.

  Next, the first determination unit 4 estimates a code amount when the 3xx intermediate code is encoded (step S42). In step S42, a Huffman table is generated from the frequency tables of Y, U, and V of the 3xx intermediate code, and for each Huffman code, the bit length of the Huffman code is multiplied by the corresponding Y, U, or V occurrence frequency. , The sum of the multiplication results is calculated. Then, the bit length when each len is encoded may be added to the sum, and the result may be used as the estimated code amount.

  The first determination unit 4 is configured such that the sum of the estimated code amount when the 3xx intermediate code obtained in step S42 is encoded and the data amount of the encoded data already encoded in step S41 is larger than the target code amount. It is determined whether or not (step S43). If the sum is larger than the target code amount (Y in step S43), the process is terminated. If the total is equal to or less than the target code amount (N in step S43), the encoding unit 7 is made to encode the 3xx component (step S44). The encoding unit 7 converts the Y, U, and V of the 3xx component into corresponding Huffman codes using the Huffman table generated in step S42. Also, len in the 3xx component is converted into a gamma code, for example.

Subsequently, the second determination unit 8 estimates a code amount when each xx intermediate code having a decomposition level number of 1 to m _q (2 in this example) is encoded (step S45). When the process proceeds to step S45 for the first time (that is, when the requantization in step S47 has not been performed yet), the code amount when the 2xx intermediate code and the 1xx intermediate code generated in step S1 are encoded is estimated. . If the process proceeds to step S45 after the requantization process in step S47, the code amount when the 2xx intermediate code and the 1xx intermediate code after requantization are encoded is estimated. The second determination unit 8 generates a Huffman table from the frequency tables of Y, U, and V of the 2xx intermediate code and the 1xx intermediate code, and for each Huffman code, the bit length of the Huffman code and the corresponding Y, U Alternatively, the frequency of V may be multiplied to calculate the sum of the multiplication results, and the bit length when each len is encoded may be added to the sum.

  After step 45, the second determining unit 8 determines whether the sum of the code amount estimated in step S45 and the data amount of the code data already encoded in steps S41 and S44 is larger than the target code amount. It is determined whether or not (step S46).

  If the sum is larger than the target code amount, the second determination unit 8 determines to perform re-quantization, and proceeds to step S47 (Y in step S46).

  If the total is equal to or less than the target code amount, the second determination unit 8 determines not to re-quantize, and proceeds to step S48 (N in step S46).

  In step S48, the encoding unit 7 converts Yxx, Uxx, and 1xx components from 2xx components to corresponding Huffman codes using the Huffman table generated at the time of code amount estimation in step S45. Also, len in each component is encoded into, for example, a gamma code.

  In step S47, the requantization unit 6 requantizes Y, U, and V from the 1xx intermediate code to the 2xx intermediate code. The requantization unit 6 determines a new dead zone and Q step for each component from 1xx to 2xx. At this time, the re-quantization unit 6 may select a dead zone and a Q step having a quantization granularity of 1 or more among the quantization granularities illustrated in FIG. That is, a combination of a dead zone and a Q step having a coarser grain size than the quantized grain size “0” used in step S31 may be selected.

  Each dead zone 1 and each dead zone 2 illustrated in FIG. 13 is determined so as to satisfy the condition of dz + h · qs. For example, regarding dead zone 1 and Q step 1 for 2xx components, dz = 0 and qs = 1 (see quantization granularity 0), and dead zone 1 of each granularity is 0 + h · 1.

  Further, Q step 2 shown in FIG. 13 is determined so as to satisfy the condition that qs · (2n + 1). Regarding Q step 2 for the 1xx component, qs = 2 (see quantized granularity 0), and the Q step for each granularity is 2 (2n + 1). Thus, it is preferable to set the Q step so as to be qs · (2n + 1) because the value after requantization does not deviate from the intermediate code before requantization.

  On the other hand, for Q step 1 shown in FIG. 13, qs = 1 (see quantization granularity 0). Therefore, regarding Q step 1, qs · (2n + 1) may not be satisfied. For example, the value of Q step 1 of the quantization granularities 1 and 2 is “2”, but this value is not an odd multiple of qs.

  The re-quantization unit 6 re-quantizes the value of the intermediate code (Y, U, V) whose absolute value is equal to or less than the dead zone to 0 using the newly selected dead zone. Further, a range in which the absolute value exceeds the value is divided for each new Q step, and the intermediate code (Y, U, V) belonging to the range is quantized to the median value of the range. This process is performed for each 2xx intermediate code and 1xx intermediate code.

  In step S47, the second determination unit 5 generates new Y, U, and V frequency tables obtained by requantizing the intermediate code. As described with reference to FIG. 9, the second determination unit 5 identifies the range used when generating the intermediate code corresponding to the range newly determined at the time of requantization. Then, the sum of the occurrence frequencies in the range may be set as the occurrence frequency in the new range. Since the occurrence frequency can be derived only by adding the already obtained values without counting the occurrence frequency again, the processing can be speeded up.

  After step S47, the processing after step S45 is repeated. The second determination unit 5 newly generates a frequency table in step S47. In the subsequent step S45, the second determination unit 5 uses the frequency table to generate a Huffman table indicating the sign of the value quantized in step S47. At this time, as described with reference to FIG. 10, the second determination unit 5 sets each value after quantization in step S31 (that is, each intermediate code generated in step S31) to a new frequency. What is necessary is just to match | combine with the new code | symbol in the Huffman table produced | generated from the table | surface. As a result, a Huffman code corresponding to the value after quantization can be generated without rewriting the intermediate code.

  In step S47, the quantization granularity is determined. For example, the quantization granularity may be determined by binary search. For example, when the process proceeds to step S47 for the first time, the quantization granularity 7 is selected and requantization is performed. When the estimated code amount is larger than the target code amount in the subsequent step S46, the quantization granularity 11 is selected. You may make it do. In this case, even if it is determined in step S46 after the quantization granularity 7 is selected that the estimated code amount is equal to or less than the target code amount, the process proceeds to step S47 and the quantization granularity 3 is selected. As described above, the binary search may select the dead zone and Q step having the smallest granularity (that is, the highest image quality) within the range in which the estimated code amount falls within the target code amount, and the process may proceed to step S48. Alternatively, every time the process proceeds to step S47, the quantization granularity may be selected in ascending order. In the operations shown in FIGS. 7 and 8, the Q step and the dead zone may be determined by binary search in the requantization process. Alternatively, when performing the re-quantization process, a Q step equal to or higher than the already selected Q step or a dead zone equal to or higher than the already selected dead zone may be newly determined.

  If the quantization granularity is increased, all intermediate codes (Y, U, V) may be quantized to one value (0). In this case, the encoding unit 7 may prohibit the encoding process in step S48. That is, the conversion to the Huffman code may not be performed.

  For example, assume that the absolute values of Y, U, and V are 128 or less. When the quantization granularity is set to 14, dead zone 2 and Q step 2 are both 128, so Y, U, and V of the 1xx intermediate code are all quantized to 0. At this time, the encoding unit 7 may not perform Y, U, and V encoding of the 1xx intermediate code in step S48. Also, the encoding of len of the 1xx intermediate code need not be performed.

  Similarly, when the quantization granularity is set to 15, dead zone 1, Q step 1, dead zone 2 and Q step 2 are all 128. Therefore, Y, U, and V of the 2xx intermediate code and the 1xx intermediate code are all quantized to 0. At this time, the encoding unit 7 may not perform Y, U, and V encoding of the 2xx intermediate code and the 1xx intermediate code in step S48. Also, the encoding of len does not have to be performed.

  In the operations shown in FIGS. 11 and 12, steps S42 and S43 may not be performed, and the process may proceed to step S44 after step S41. In this case, determining that the estimated code amount is larger than the target code amount in step S33 means determining to encode the 3xx code.

Next, another example of operation when N = 3 and m _q = 2 is described. FIG. 14 is a flowchart illustrating another example of processing progress when N = 3 and m _q = 2. Hereinafter, the creation of the intermediate code of each component such as the 1xx component, the creation of the frequency table, the estimation of the amount of code when encoded, and the encoding are the same as the operations already described, and detailed description thereof will be omitted.

  First, the wavelet transform unit 2 performs wavelet transform on the input image data up to the decomposition level number 3.

  Then, the intermediate code generation unit 3 generates an intermediate code of 1xx component. The first determination unit 41 generates a frequency table of 1xx intermediate codes (step S41).

  Next, the intermediate code generation unit 3 generates an intermediate code of 3LL components. In addition, the first determination unit 41 generates a frequency table of 3LL intermediate codes (step S42).

  Next, the encoding unit 7 encodes the 3LL intermediate code (step S43). The encoding unit 7 may derive the Huffman table for encoding the 3LL intermediate code from the frequency table of the 3LL intermediate code and perform the encoding. Alternatively, in step S42, the first determination unit 41 may generate a Huffman table for 3LL intermediate code from the frequency table.

  Next, the intermediate code generation unit 3 generates an intermediate code of 3xx components. The first determination unit 41 generates a frequency table of 3xx intermediate codes (step S44). Furthermore, the first determination unit 41 estimates a code amount when the 3xx intermediate code is encoded (step S45).

  Subsequently, the first determination unit 41 determines whether or not the sum of the code amount estimated in step S45 and the data amount of the code data already obtained by the encoding in step S43 is larger than the target code amount. If it is larger than the target code amount, the process is terminated. In this case, only the 3LL component is encoded and the process is terminated.

  On the other hand, when the sum of the estimated code amount and the data amount of the code data obtained in step S43 is equal to or less than the target code amount (N in step S46), the first determination unit 41 sends 3xx to the encoding unit 7. The intermediate code is encoded (step S47).

  Next, the intermediate code generation unit 3 generates an intermediate code of 2xx components. The second determination unit 5 generates a frequency table of 2xx intermediate codes (step S48).

  Then, the requantization unit 6 requantizes the 2xx intermediate code, and the second determination unit 5 generates a frequency table of the 2xx intermediate code after requantization (step S49). This frequency table may be created using the frequency table created in step S48. That is, if each range used in step S48 corresponding to each range defined for quantization in step 49 is specified, the sum of the occurrence frequencies in each range is set as the occurrence frequency in the new range. Good.

  The requantization unit 6 requantizes the 1xx intermediate code, and the second determination unit 5 generates a frequency table of the 1xx intermediate code after requantization (step S50). Similarly to the case of step S49, the frequency table of the intermediate code after requantization is generated using the frequency table that has already been created in step S41.

  Next, the second determination unit 5 estimates a code amount when the 2xx intermediate code after requantization is encoded into a Huffman code (step S51). Similarly, the code amount when the 1xx intermediate code after requantization is encoded into a Huffman code is estimated (step S52). Then, the second determination unit 5 determines whether the sum of the already encoded 3LL component and 3xx component code data and the estimated code amount obtained in steps S51 and S52 is larger than the target code amount. Is determined (step S53).

  If the sum is larger than the target code amount (Y in step S53), the processes in and after step S49 are repeated. If the total is equal to or less than the target code amount (N in step S53), the encoding unit 7 encodes the re-quantized 2xx intermediate code and 1xx intermediate code into Huffman codes (steps S54 and S55). ).

  After step S48, the process proceeds to step S51. When the process first proceeds to steps S51 and S52, the data amount of the code data when the intermediate code before re-quantization is encoded may be estimated. . Thereafter, the processing after step S49 may be repeated according to the determination result of step S53.

  In the above embodiment, when estimating the code amount when the intermediate code is encoded, the code amount when len is encoded into the gamma code is also calculated. The code amount when len is encoded into a gamma code may be obtained by converting len into a gamma code. Alternatively, the total amount of codes (total length of run length codes) L when len is encoded may be obtained approximately by the following equation (1).

  In equation (1), b is the longest number of bits that the encoded len can take. For example, b is predetermined as 31 bits or the like. P is the number of sets of coefficients (Y, U, V) and len values. Expression (1) is an approximate expression assuming that the run length occurs according to the exponential distribution.

  In the above embodiment, the case where the frequency table is generated when the intermediate code is generated or the intermediate code is requantized and the Huffman table is generated from the frequency table has been described. However, the Huffman table is prepared in advance. You may do it. That is, the Huffman table used when encoding the intermediate code and the Huffman table used when encoding the re-quantized intermediate code are stored in advance, and the code amount is estimated and encoded using the Huffman table. May also be performed. Since the Huffman table used when encoding the re-quantized intermediate code differs depending on the number of re-quantization processes, it may be held individually for each re-quantization. When a Huffman table is held in advance, it is highly possible that the Huffman table does not match the actual data (Y, U, V) occurrence frequency, but it can be used for code amount estimation and encoding processing. .

In the above embodiment, the case where the encoding method is Huffman encoding has been described as an example. However, encoding may be performed by other methods. For example, the intermediate code (Y, U, V) may be arithmetically encoded. When Y, U, and V are arithmetically encoded, the data amount when encoded may be estimated as follows. That is, assuming that the probability of occurrence of certain data is P, the code amount when the data is arithmetically encoded may be calculated as -Log ₂ P bits. Thus, even when data is arithmetically encoded, the amount of code after encoding can be estimated, and arithmetic encoding may be applied to the present invention.

  In each of the above embodiments, when intermediate data is generated (for example, in step S1 or step S31), the result obtained by quantizing the coefficient obtained by the wavelet transform is an intermediate code. Although described as an example, a coefficient (wavelet coefficient) obtained by wavelet transform itself may be used as an intermediate code. Specifically, the wavelet coefficient itself may be used as intermediate data for a component having a predetermined number of decomposition levels or more. For example, when the wavelet transform is performed up to the decomposition level number N, the wavelet coefficients having the decomposition level number “N−T” or more and N or less are used as intermediate codes, and the wavelet coefficients less than the decomposition level number “N−T” are processed in step S1. As described in S31, the result of quantizing the wavelet coefficients may be used as the intermediate code. The number of decomposition levels “N−T” to “T” using wavelet coefficients as intermediate codes as they are may be determined in advance according to the image quality and compression rate required for the encoded data. The wavelet coefficient may be used as an intermediate code for the LL component, and the data obtained by quantizing the wavelet coefficient as described in steps S1 and S31 may be used as the intermediate code for the xx component.

  Next, the outline of the present invention will be described. FIG. 15 is a block diagram showing an outline of the present invention. As shown in FIG. 15, the image processing apparatus of the present invention includes a wavelet transform unit 81, an intermediate data derivation unit 82, and an encoding unit 83.

  The wavelet transform unit 81 (for example, the wavelet transform unit 2) performs wavelet transform on the image data up to a predetermined number of decomposition levels (for example, N).

The intermediate data deriving means 82 (for example, the intermediate code generation unit 3) has components other than the LL component equal to or less than the predetermined decomposition level (for example, each xx component from the decomposition level number 1 to N), and a plurality of types of decomposition. wavelet transform obtained in the plurality of decomposition levels the number of LL component at level number (e.g., LL component from m _i to N) from or by a wavelet transform which is a value obtained by quantizing at a predetermined quantization step Intermediate data which is the obtained coefficient itself is derived.

  The encoding means 83 (for example, is realized by the first determination unit 4 that generates a frequency table in step S1 and performs steps S3, S4, S5, and S7 and the encoding unit 7 that performs step S9). Frequency information (for example, a frequency table) indicating the frequency of occurrence of intermediate data of components other than the LL component equal to or less than the predetermined number of decomposition levels and LL components of the plurality of types of decomposition levels is generated, and the number of decomposition levels of the plurality of types Encoding the intermediate data with the minimum number of decomposition levels that satisfies the condition that the encoded code amount estimated based on the frequency information is not larger than the target code amount (for example, N in step S5). I do.

  Since the intermediate data having the minimum number of decomposition levels satisfying the above conditions is encoded, a high-quality and large-capacity code can be obtained if the number of decomposition levels is small, and a low image quality is small if the number of decomposition levels is large. A capacity sign is obtained. Since the coding is performed for the number of decomposition levels that satisfies the conditions as described above, the coding can be performed so that the image quality of the encoded image data is improved as much as possible while the code amount is equal to or less than the target code amount. In addition, since it is not necessary to perform wavelet transform or encoding processing again, the processing load can be reduced.

  If it is determined that the encoded code amount is larger than the target code amount even at a decomposition level number 1 smaller than the predetermined decomposition level number (for example, N in step S3), the LL component of the predetermined decomposition level number The maximum level LL component encoding means for encoding the intermediate data (for example, the part that executes step S11), and the encoded code amount is smaller than the target code amount even in the number of decomposition levels smaller than the predetermined number of decomposition levels If it is determined to be large, the intermediate data of the components other than the LL component is quantized so as to satisfy the condition that the encoded code amount is not larger than the target code amount, and the quantized intermediate data is encoded. The post-quantization data encoding means (for example, the part which performs step S18-S21) to convert may be sufficient.

  According to such a configuration, when encoding is performed with the maximum number of decomposition levels, code amount control can be performed so as to further reduce the code amount.

  Further, when the quantized data encoding means quantizes the intermediate data so as to satisfy the condition that the encoded code amount is not larger than the target code amount, all the quantized data is 0. If it becomes, the structure which prohibits the encoding of the intermediate data after the quantization may be sufficient.

  According to such a configuration, it is possible to reduce the code amount and the processing load by omitting the encoding process itself when the quantization granularity becomes coarser than a certain level.

The quantized data encoding means is configured to perform quantization on intermediate data of components other than the LL component having a number of decomposition levels of 1 or more and a fixed value (for example, 1 or more and _mq or less). Also good.

  According to such a configuration, quantization processing of intermediate data is limited to high-resolution components with a large percentage of code amount (in other words, high-resolution components that have a large influence on the code amount due to quantization of intermediate data). To reduce the processing load.

  Further, the number of decomposition levels exceeding the predetermined value is selected in descending order from the predetermined number of decomposition levels, and the code amount when the intermediate data of components other than the LL component of the selected number of decomposition levels is estimated is estimated. If the sum of the encoded amount and the already encoded code amount is not larger than the target code amount, the intermediate data is encoded, and if the sum is larger than the target code amount, stop determination means for stopping the encoding process (for example, step S12). -The part which performs S17) may be sufficient.

  According to such a configuration, for the component for which the intermediate data is not quantized, the encoding is terminated when it is determined that the estimated code amount becomes larger than the target code amount. The amount can be suppressed below.

  Alternatively, the predetermined number of decomposition levels may be 3, and the intermediate data deriving unit may derive intermediate data of LL components having decomposition levels of 2 and 3.

  When wavelet transform is performed with the number of decomposition levels set to 3, the balance between encoding efficiency and image quality is good, so it is preferable to set the predetermined number of decomposition levels to 3. On the other hand, in the high image quality region where the quantization granularity becomes fine, when the number of decomposition levels is 3, image quality degradation is easily perceived (particularly when the wavelet basis function is HAAR). In this case, if the encoding is performed based on the intermediate data having the number of decomposition levels of 2, the effect of suppressing deterioration in image quality is great. Therefore, with the configuration as described above, it is possible to reduce the required memory and processing load and realize effective code amount control.

  Further, the constant value may be 2, and the post-quantization data encoding unit may perform quantization on intermediate data of components other than the LL components having 1 and 2 decomposition levels. .

  In general, in the wavelet transform, at a standard compression ratio of 1/10, the code amount of the 1xx component occupies about 1/2 of the whole, and the code amount of the 2xx component shows about 1/3 of the whole. Therefore, according to the above configuration, it is possible to achieve efficient code amount control by quantizing the intermediate data by limiting to the components other than the LL components having the number of decomposition levels of 1 and 2.

  The present invention is suitably applied to an image processing apparatus that encodes image data.

It is a block diagram which shows the example of the image processing apparatus of this invention. It is explanatory drawing which shows a wavelet transformation typically. It is explanatory drawing which shows typically the intermediate code produced | generated from the wavelet coefficient. It is explanatory drawing which shows the example of a frequency table. It is explanatory drawing which shows the example of a Huffman table. It is explanatory drawing which shows the example of the result of a Huffman encoding typically. It is a flowchart which shows an example of the process progress of this invention. It is a flowchart which shows an example of the process progress of this invention. It is explanatory drawing which shows frequency table derivation typically. It is explanatory drawing which shows matching with the code | cord | chord derived after the value of the intermediate code and requantization. It is a flowchart which shows an example of a process progress in case N = 3 and _mq = 2. It is a flowchart which shows an example of a process progress in case N = 3 and _mq = 2. It is explanatory drawing which shows the example of a dead zone and Q step. It is a flowchart which shows the example of other process progress when N = 3 and _mq = 2. It is a block diagram which shows the outline | summary of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Image processing apparatus 2 Wavelet transformation part 3 Intermediate code production | generation part 4 1st determination part 5 2nd determination part 6 Requantization part 7 Encoding part 81 Wavelet transformation means 82 Intermediate data generation means 83 Coding means

Claims (11)

Wavelet transform means for performing wavelet transform on image data up to a predetermined number of decomposition levels;
Quantization was performed in a predetermined quantization step from components other than the LL component equal to or less than the predetermined number of decomposition levels and the LL components of the plurality of types of decomposition levels obtained by wavelet transform with a plurality of types of decomposition levels. Intermediate data deriving means for deriving intermediate data which is a value or a coefficient itself obtained by wavelet transform;
Frequency information indicating the frequency of occurrence of intermediate data of components other than the LL component equal to or less than the predetermined number of decomposition levels and LL components of the plurality of types of decomposition levels is generated, and the frequency information among the plurality of types of decomposition levels Encoding means for encoding intermediate data having a minimum number of decomposition levels that satisfies the condition that the encoded code amount estimated based on the above is not larger than the target code amount Image processing device. If it is determined that the encoded code amount is larger than the target code amount even at a decomposition level number that is one smaller than the predetermined decomposition level number, intermediate data of the LL component of the predetermined decomposition level number is encoded. Maximum level LL component encoding means;
If it is determined that the encoded code amount is larger than the target code amount even at a decomposition level number that is one smaller than the predetermined decomposition level number, intermediate data of components other than the LL component is converted into the encoded code amount. The image processing apparatus according to claim 1, further comprising: a quantized data encoding unit that performs quantization so as to satisfy a condition that is not larger than a target code amount, and encodes the intermediate data after the quantization. If the quantized data encoding means quantizes the intermediate data so that the condition that the encoded code amount is not larger than the target code amount is satisfied, all the quantized data becomes 0 The image processing device according to claim 2, wherein encoding of the intermediate data after the quantization is prohibited. The image processing apparatus according to claim 2, wherein the quantized data encoding unit performs quantization on intermediate data of components other than the LL component having a decomposition level number of 1 or more and a constant value or less. The number of decomposition levels exceeding the predetermined value is selected in descending order from the predetermined number of decomposition levels, the code amount when the intermediate data of components other than the LL component of the selected number of decomposition levels is encoded, and the estimated code 5. The apparatus according to claim 4, further comprising: a stop determination unit that encodes the intermediate data if the sum of the amount and the already encoded code amount is not greater than the target code amount, and stops the encoding process if the sum is greater than the target code amount. The image processing apparatus described. The predetermined number of decomposition levels is 3,
The image processing apparatus according to claim 1, wherein the intermediate data deriving unit derives intermediate data of LL components having two and three decomposition levels. The constant value is 2,
7. The post-quantization data encoding unit performs quantization on intermediate data of components other than LL components having a decomposition level number of 1 and 2, according to any one of claims 4 to 6. Image processing device. Perform wavelet transform on the image data up to a predetermined number of decomposition levels,
Quantization was performed in a predetermined quantization step from components other than the LL component equal to or less than the predetermined number of decomposition levels and the LL components of the plurality of types of decomposition levels obtained by wavelet transform with a plurality of types of decomposition levels. Deriving intermediate data that is data or the coefficients themselves obtained by wavelet transform,
Frequency information indicating the frequency of occurrence of intermediate data of components other than the LL component equal to or less than the predetermined number of decomposition levels and LL components of the plurality of types of decomposition levels is generated, and the frequency information among the plurality of types of decomposition levels An image processing method characterized in that encoding is performed on intermediate data having a minimum number of decomposition levels that satisfies a condition that a code amount after encoding estimated based on the above is not larger than a target code amount. If it is determined that the encoded code amount is larger than the target code amount even at a decomposition level number 1 smaller than the predetermined decomposition level number, intermediate data of the LL component of the predetermined decomposition level number is encoded,
If it is determined that the encoded code amount is larger than the target code amount even at a decomposition level number that is one smaller than the predetermined decomposition level number, intermediate data of components other than the LL component is converted into the encoded code amount. The image processing method according to claim 8, wherein quantization is performed so as to satisfy a condition that is not larger than a target code amount, and intermediate data after the quantization is encoded. On the computer,
Wavelet transform processing for performing wavelet transform on image data up to a predetermined number of decomposition levels,
Quantization was performed in a predetermined quantization step from components other than the LL component equal to or less than the predetermined number of decomposition levels and the LL components of the plurality of types of decomposition levels obtained by wavelet transform with a plurality of types of decomposition levels. Intermediate data derivation process for deriving intermediate data that is data or a coefficient itself obtained by wavelet transform, components other than the LL component equal to or less than the predetermined number of decomposition levels, and LL components of the plurality of types of decomposition levels The frequency information indicating the frequency of occurrence of the intermediate data is generated, and among the plurality of types of decomposition levels, the condition that the encoded code amount estimated based on the frequency information is not larger than the target code amount is satisfied An image processing program for executing an encoding process for encoding intermediate data having a minimum number of decomposition levels. In the computer,
If it is determined that the encoded code amount is larger than the target code amount even at a decomposition level number that is one smaller than the predetermined decomposition level number, intermediate data of the LL component of the predetermined decomposition level number is encoded. Maximum level LL component encoding process, and
If it is determined that the encoded code amount is larger than the target code amount even at a decomposition level number that is one smaller than the predetermined decomposition level number, intermediate data of components other than the LL component is converted into the encoded code amount. The image processing program according to claim 10, wherein quantization is performed so as to satisfy a condition that is not larger than a target code amount, and post-quantization data encoding processing for encoding intermediate data after the quantization is performed.

Images