JP2006100993A - Image coding device and coding parameter calculation program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve a compression ratio without deteriorating subjective image quality greatly. <P>SOLUTION: An image coding device comprises an orthogonal transform 4 for orthogonally transforming image data 101 for outputting an orthogonal transform coefficient 102; an orthogonal transform coefficient statistical value calculator 5 for calculating statistical information 103 for each frequency in the orthogonal transform coefficient; and a coding parameter control unit 6 for calculating coding parameters 104 used for quantization and coding based on the statistical information 103 for each frequency in the calculated orthogonal transform coefficient. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an image encoding apparatus and an encoding parameter calculation program for calculating appropriate encoding parameters used when image data is quantized and encoded.

  In the conventional image encoding device described in Non-Patent Document 1, the input image data is orthogonally transformed, the orthogonal transformation coefficient after the orthogonal transformation is quantized with a predetermined quantization step size, and the quantization after the quantization is performed. The coefficient is subjected to variable length coding, and the variable length coded image data is stored in the transmission buffer as a coded bit string.

  The encoded bit string stored in the transmission buffer is output to the outside of the image encoding device at a predetermined transmission rate. At this time, it is necessary to control the code amount which is the data amount of the encoded bit string stored in the transmission buffer so that the transmission buffer does not underflow or overflow. As a method of controlling the code amount, a method of changing the quantization step size at the time of quantization according to the code amount accumulated in the transmission buffer is widely used. That is, when the code amount accumulated in the transmission buffer is large, the code amount generated by decreasing the quantization step size is decreased, and when the code amount accumulated in the transmission buffer is small, the quantization step size is decreased. By increasing the amount of code generated in this way, it is possible to generate a code amount according to the transmission rate.

In the case of MPEG (Moving Picture Experts Group) -2, the quantization step size is determined by two parameters: a quantization scale (quantizer_scale) and a quantization matrix (quantizer_matrix). At the time of quantization, for example, the orthogonal transformation coefficient is calculated as shown in the following equation (1) using a quantization matrix (quantizer_matrix) and a quantization scale (quantizer_scale).
Quantization coefficient (u, v)
= (Orthogonal transform coefficient (u, v) × 8 ÷ quantizer_matrix (u, v)
+ R) // quantizer_scale (1)
Here, u and v are integers satisfying 0 ≦ u and v ≦ 7, r is an integer, and // indicates a process of rounding off after the decimal point after division.

  To change the quantization step size, a method of changing the quantization scale by fixing the value of the quantization matrix is widely used, but a method of changing the quantization matrix according to the spatial frequency distribution of the image data is also reported. Has been. For example, in Patent Document 1, a high frequency component of image data is detected, and the quantization matrix is changed using the detection result.

Yasuyuki Nakajima, Toshinori Otaka, Katsumi Tahara, “3-2 Video Compression”, Journal of the Television Society, 1995, Vol. 49, NO. 4, p435-466 Japanese Patent Laid-Open No. 11-234669 (paragraphs 0026 and 0027)

  Since the conventional image coding apparatus is configured as described above, the method for controlling the code amount by changing the quantization scale does not always provide good image quality. For example, when the quantization scale is increased However, there is a problem that the subjective image quality may be remarkably deteriorated because the quantization accuracy of low-frequency components that are highly related to the image quality is also lowered.

  Further, in Patent Document 1, by changing the quantization matrix according to the spatial frequency distribution of the image, it is possible to suppress deterioration in subjective image quality due to a decrease in quantization accuracy of low frequency components. The problem is that the compression ratio cannot be improved sufficiently because changes in the coding characteristics due to the change in the values, that is, changes in the appropriate scan order of the quantization coefficients and changes in the appropriate values of the VLC table are not considered. was there.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to obtain an image encoding device and an encoding parameter calculation program capable of improving the compression rate without significantly degrading the subjective image quality. And

  An image encoding apparatus according to the present invention includes an orthogonal transform unit that orthogonally transforms image data and outputs orthogonal transform coefficients, and orthogonal transform coefficient statistical information that calculates statistical information for each frequency of the output orthogonal transform coefficients. A calculation unit and an encoding parameter control unit that calculates an encoding parameter to be used for quantization and encoding based on the calculated statistical information for each frequency of the orthogonal transform coefficient are provided.

  According to the present invention, it is possible to improve the compression rate without significantly degrading the subjective image quality.

An embodiment of the present invention will be described below.
Embodiment 1 FIG.
FIG. 1 is a block diagram showing a configuration of an image coding apparatus according to Embodiment 1 of the present invention. This image encoding apparatus includes a frame memory 1, a subtracting unit 2, a switch 3, an orthogonal transform unit 4, an orthogonal transform coefficient statistical information calculation unit 5, an encoding parameter control unit 6, a quantization unit 7, and a VLC (Variable Length Coding). 8, a transmission buffer 9, an inverse quantization unit 10, an inverse orthogonal transform unit 11, an addition unit 12, a switch 13, a frame memory 14, and a motion compensation unit 15.

  In FIG. 1, the frame memory 1 temporarily stores input image data 101, and the orthogonal transform unit 4 performs motion compensation with the image data 101 stored in the frame memory 1 or the image data 101 generated by the subtraction unit 2. The difference data 116 that is the difference from the predicted reference image data 115 output from the unit 15 is orthogonally transformed to output the orthogonal transformation coefficient 102.

  The orthogonal transform coefficient statistical information calculation unit 5 receives the orthogonal transform coefficient 102 output from the orthogonal transform unit 4 and calculates statistical information 103 for each frequency of the orthogonal transform coefficient, and the encoding parameter control unit 6 performs orthogonal transform coefficient statistics. Based on the statistical information 103 for each frequency of the orthogonal transform coefficient calculated by the information calculation unit 5, an appropriate encoding parameter 104 used for quantization and variable length encoding is calculated.

  The quantization unit 7 quantizes the orthogonal transform coefficient quantized by quantizing the orthogonal transform coefficient 102 output from the orthogonal transform unit 4 based on the coding parameter 104 calculated by the coding parameter control unit 6. The coefficient 105 is output. Based on the encoding parameter 104 calculated by the encoding parameter control unit 6, the VLC unit 8 performs variable length encoding on the quantization coefficient 105 and encoding related information, and outputs encoded data 106. The transmission buffer 9 temporarily stores the encoded bit string 106 output from the VLC unit 8 and outputs it to the outside.

  The inverse quantization unit 10 inversely quantizes the quantization coefficient 105 output from the quantization unit 7 based on the encoding parameter 104 calculated by the encoding parameter control unit 6, and outputs an orthogonal transform coefficient 111. The orthogonal transform unit 11 performs inverse orthogonal transform on the inversely quantized orthogonal transform coefficient 111 to restore the image data 112. The frame memory 14 includes the image data 112 output from the inverse orthogonal transform unit 11 or the image data 112 from the inverse orthogonal transform unit 11 generated by the addition unit 12 and the predicted reference image data 115 output from the motion compensation unit 15. The added addition data 113 is temporarily stored, and the motion compensation unit 15 uses the reference image data 114 stored in the frame memory 14 to generate predicted reference image data 115 for the image data 101 to be encoded later. .

  FIG. 2 is a block diagram showing an internal configuration of the encoding parameter control unit 6. The encoding parameter control unit 6 includes a quantization matrix calculation unit 21, an expected quantization coefficient calculation unit 22, a scan order calculation unit 23, a VLC table calculation unit 24, a quantization scale calculation unit 25, and an encoding parameter setting unit 26. It has.

  In FIG. 2, the quantization matrix calculation unit 21 calculates an appropriate quantization matrix 121 based on the statistical information 103 for each frequency of the orthogonal transform coefficient calculated by the orthogonal transform coefficient statistical information calculation unit 5. The quantized coefficient expected value calculation unit 22 is a quantized orthogonal transform coefficient after quantization based on the statistical information 103 for each frequency of the orthogonal transform coefficient and the quantization matrix 121 calculated by the quantization matrix calculation unit 21. The expected value 122 of the coefficient is calculated. The scan order calculation unit 23 calculates an appropriate scan order 123 based on the quantization coefficient expected value 122 calculated by the quantization coefficient expected value calculation unit 22 or the statistical information 103 for each frequency of the orthogonal transform coefficient, and An expected value 124 of the quantized coefficients rearranged according to the calculated scan order 123 is output.

  The VLC table calculation unit 24 calculates an appropriate VLC table 125 based on the expected value 124 of the quantized coefficients rearranged in the scan order or the statistical information 103 for each frequency of the orthogonal transform coefficient, and calculates the calculated VLC table. The expected value 126 of the code amount is calculated by performing variable length encoding using 125. The quantization scale calculation unit 25 calculates an appropriate quantization scale 127 based on the expected value 126 of the code amount or the statistical information 103 for each frequency of the orthogonal transform coefficient. The encoding parameter setting unit 26 sets the calculated quantization matrix 121, scan order 123, VLC table 125, and quantization scale 127 as appropriate encoding parameters 104 at appropriate timings, respectively. Output to the unit 8 and the inverse quantization unit 10.

Next, the operation will be described.
In FIG. 1, input image data 101 is stored in the frame memory 1. One frame corresponding to the encoding order is selected from the image data 101 stored in the frame memory 1 and output to the orthogonal transform unit 4. The input of the orthogonal transformation unit 4 is the image data 101 output from the frame memory 1 or the difference between the image data 101 generated by the subtraction unit 2 and the predicted reference image data 115 generated by the motion compensation unit 15. It becomes one of the difference data 116. The orthogonal transform unit 4 performs orthogonal transform on the input image data 101 or difference data 116 and outputs an orthogonal transform coefficient 102.

  The quantization unit 7 quantizes the orthogonal transform coefficient 102 output from the orthogonal transform unit 4 based on the encoding parameter 104 output from the encoding parameter control unit 6, that is, the quantization matrix 121 and the quantization scale 127. A quantization coefficient 105 that is an orthogonal transform coefficient after quantization is output. The VLC unit 4 receives the encoding parameters 105 output from the encoding parameter control unit 6, that is, the quantization matrix 121, the scan order 123, the VLC table 125, and the quantization scale 127, and the input scan order 123 and the VLC table. 125, the quantization coefficient 105 from the quantization unit 7 and the encoding related information such as the quantization matrix 121 added as a header, the scan order 123, the VLC table 125, and the quantization scale 127 are variable-length encoded. . The transmission buffer 9 stores a coded bit string 106 that has been subjected to variable length coding.

  The inverse quantization unit 10 is based on the same step size as the quantization unit 7, that is, the quantization matrix 121 and the quantization scale 127 that are the encoding parameters 104 output from the encoding parameter control unit 6. The quantized coefficient 105 output from is dequantized and an orthogonal transform coefficient 111 is output. The inverse orthogonal transform unit 11 performs inverse orthogonal transform on the inversely quantized orthogonal transform coefficient 111 to generate image data 112. This image data 112 is accumulated in the frame memory 14, or is added to the prediction reference image data 115 generated by the motion compensation unit 15 by the addition unit 12 and accumulated in the frame memory 14 as addition data 113. The motion compensation unit 15 generates predicted reference image data 115 from the image data 101 to be encoded later and the reference image data 114 stored in the frame memory 14.

  The orthogonal transform coefficient statistical information calculation unit 5 receives the orthogonal transform coefficient 102 output from the orthogonal transform unit 4, and performs statistics for each frequency of orthogonal transform coefficients in an N picture (N: arbitrary real number) period or an N slice period. Information 103 is calculated and output to the encoding parameter control unit 6. The types of statistical information 103 for each frequency of the orthogonal transform coefficient to be calculated include, for example, an average value, a maximum value, a minimum value, a variance value, and a deviation value for each frequency of the orthogonal transform coefficient in the N picture period or the N slice period. , Intermediate value, total value, etc.

  In FIG. 2, the quantization matrix calculation unit 21 of the encoding parameter control unit 6 outputs statistical information 103 for each frequency of orthogonal transform coefficients in the N picture period or N slice period output from the orthogonal transform coefficient statistical information calculation unit 5. Based on the above, a quantization matrix 121 suitable for the input image data 101 is calculated.

Next, an example of a calculation formula of the quantization matrix 121 by the quantization matrix calculation unit 21 is shown in the following formula (2).
quantiser_matrix (u, v)
= Normalization [f (u, v) + | K (u, v) | × g (u, v)] (2)
Here, u and v are integers satisfying 0 ≦ u and v ≦ 7, and K (u, v) is for each frequency of orthogonal transform coefficients aggregated in the past N picture (N: arbitrary real number) period or N slice period. F (u, v) and g (u, v) are coefficient sets. Normalization [*] represents a conversion function for normalizing numerical values so that the quantization matrix 121 is in a predetermined format. For example, in MPEG-2, the components of the quantization matrix 121 are integer values in the range of 1 to 255, and in the case of intra coding, quantizer_matrix (0, 0) = 8 is defined. Normalize the number to match That is, when the calculated quantizer_matrix (0, 0) value of the quantization matrix 121 is not 8, the entire quantizer_matrix (u, v) value is multiplied by a predetermined real number so that it becomes 8, or the quantization matrix Clip 121 to be in the range of 1 to 255.

In the above equation (2), f (u, v) and g (u, v) are coefficient sets arbitrarily set according to human visual frequency characteristics, image coding characteristics, user judgment, and the like.
FIG. 3 is a diagram showing a specific example of the coefficient sets f (u, v) and g (u, v). F (u, v) shown in FIG. 3A is a standard matrix (default matrix) in MPEG-2, and is determined in consideration of human visual frequency characteristics and the like. G (u, v) shown in FIG. 3B is derived from experimental data, and is a gradient type coefficient set in which the low frequency component has a small value and the high frequency component has a large value. In this way, “f (u, v) + | K (u, v) | × g (u, v)” shown in Expression (2) represents the standard matrix f (u, v) in the past N pictures. The coefficient set g (u) derived from the average value K (u, v) as the statistical information 103 for each frequency of the orthogonal transform coefficients aggregated in the (N: arbitrary real number) period or N slice period and the experimental data , V) is corrected by a value obtained by multiplication.

  When the quantization matrix 121 calculated from the above equation (2) is used, the value of | K (u, v) | of the frequency component where power is concentrated becomes large, and the quantization accuracy is rough in the frequency component where power is concentrated. Therefore, it is possible to prevent an enormous amount of codes from being generated at a specific frequency component. Further, by using a gradient type coefficient set as shown in FIG. 3B for g (u, v), the low frequency component in the above equation (2) has a value of | K (u, v) | The higher the frequency component, the less susceptible to changes, the more susceptible to changes in the value of | K (u, v) |, and high quantization accuracy can be maintained for low frequency components that greatly affect subjective image quality. The quantization accuracy can be lowered in order from the high-frequency component with the lower priority.

  In the above equation (2), the average value of each orthogonal transform coefficient calculated in the past N picture period or N slice period is used as K (u, v), but other statistical values may be used instead. You can also. For example, when orthogonal transform coefficient information of a picture to be encoded is obtained in advance, such information may be used. Further, instead of the average value, any one of a maximum value, a minimum value, a variance value, a deviation value, an intermediate value, a total value, and the like may be used.

  Further, in this example, the frequency of the orthogonal transform coefficient 102 is divided into 64 (8 × 8) areas as a unit of aggregation of the orthogonal transform coefficient 102, but may be an arbitrary area division according to the encoding method and the device configuration. Absent.

Next, another example of the quantization matrix calculation formula is shown in the following formula (3).
quantiser_matrix (u, v)
= Normalization [f (u, v) + k × g (u, v)] (3)
Here, k is a real number calculated using the statistical information 103 for each frequency of the orthogonal transform coefficient. For example, a value proportional to a specific frequency component in the statistical information 103 calculated for each frequency or a value proportional to the slope of the statistical information 103 from the low frequency component to the high frequency component is set to k. When the above equation (3) is used, since only the statistical information 103 of some frequencies is used, there is an advantage that the statistical information can be aggregated and the amount of calculation can be suppressed.

In FIG. 2, the expected quantization coefficient value calculation unit 22 is based on the statistical information 103 for each frequency of the orthogonal transform coefficient and the quantization matrix 121 calculated by the quantization matrix calculation unit 21, and the expected value 122 of the quantization coefficient. Is calculated.
Next, an example of a calculation formula for the expected value 122 of the quantization coefficient is shown in the following formula (4).
Expected value of quantization coefficient (u, v)
= (K (u, v) × 8 ÷ quantizer_matrix (u, v) + r)
// Target quantizer_scale (4)
Here, u and v are integers satisfying 0 ≦ u and v ≦ 7, r is an integer, and K (u, v) is an orthogonal transform aggregated in, for example, the past N picture (N: any real number) period or N slice period. An average value of the coefficient 102 as the statistical information 103 for each frequency is shown. As described above, this average value can be substituted by other statistical information. The target quantizer_scale is set with the target quantizer_scale in the encoding of each picture, but is corrected with an appropriate quantization scale 127 calculated by the quantization scale calculation unit 25 described later.

In FIG. 2, the scan order calculation unit 23 calculates an appropriate scan order 123 based on the quantization coefficient expected value 122 calculated by the quantization coefficient expected value calculation unit 22.
FIG. 4 is a diagram showing the scan order of the quantization coefficient 105 defined by MPEG-2. FIG. 4A shows the zigzag scan order, and FIG. 4B shows the vertical priority scan order. The quantized coefficients 105 are encoded in the numerical order shown in FIG. 4A or FIG. 4B and multiplexed on the encoded bit string 106. Which order of FIG. 4 (a) and FIG. 4 (b) is adopted can generally be arbitrarily set by the image coding apparatus. In MPEG-2, a shorter variable length code can be assigned as the run length (length of consecutive zero coefficients when the quantized coefficients 105 are arranged in the scan order) becomes shorter. By selecting an appropriate scan order 123, the amount of codes can be reduced.

FIG. 5 is a flowchart showing processing when the scan order calculation unit 23 selects an appropriate scan order 123.
FIG. 6 is a diagram illustrating an example of the division of the horizontal frequency area and the vertical frequency area of the expected value 122 of the quantization coefficient.
In step ST11 of FIG. 5, the absolute value sum Sum_H of the expected value 122 of the quantization coefficient in the horizontal frequency area and the absolute value sum Sum_V of the expected value 122 of the quantization coefficient in the vertical frequency area are calculated. Here, the absolute value sum of all or part of the expected value 122 of the quantization coefficient belonging to each area is calculated.

  In step ST12, the calculated absolute value sums Sum_H and Sum_V are compared. If the absolute value sum Sum_H is larger than the absolute value sum Sum_V, the zigzag scan is selected in step ST13. Otherwise, step ST14 is selected. The vertical direction priority scan is selected. In this way, an appropriate scan order 123 is calculated using the absolute value sum of all or part of the expected value 122 of the quantization coefficient belonging to each area as a scale.

  FIG. 7 is a flowchart showing another process when the scan order calculation unit 23 selects an appropriate scan order 123. In step ST21, the absolute value sum Sum_H of the expected value 122 of the quantization coefficient in the horizontal frequency area is calculated. In step ST22, the calculated absolute value sum Sum_H is compared with a predetermined real number TH. If the absolute value sum Sum_H is larger than the predetermined real number TH, a zigzag scan is selected in step ST23. In step ST24, the vertical direction priority scan is selected. The processing of the flowchart shown in FIG. 7 is advantageous in that it can be realized with a smaller amount of computation than the processing of the flowchart shown in FIG.

  In addition to the methods shown in FIGS. 5 and 7, it is also effective to select the smaller run-length total value when the quantized coefficient expected value 122 is aligned in each scan order.

  In an encoding scheme other than MPEG-2, when a scan order other than the two types shown in FIG. 4 can be used, an appropriate scan order 123 is determined so as to give priority to a component having a large quantized coefficient expected value 122. As a result, the amount of codes can be reduced.

  In the above example, the scan order calculation unit 23 calculates an appropriate scan order 123 based on the expected value 122 of the quantized coefficients calculated by the quantized coefficient expected value calculation unit 22, but the scan order calculation unit 23 Instead of the expected value 122 of the quantization coefficient, the proper scan order 123 may be calculated based on the statistical information 103 for each frequency of the orthogonal transform coefficient, for example, the average value or the minimum value. However, in this case, the orthogonal transform coefficient statistical information calculation unit 5 appropriately corrects the statistical information 103 for each frequency of the orthogonal transform coefficient so as to match the range of coding parameter values defined in advance. A value shall be output.

  In this way, the scan order calculation unit 23 performs an appropriate scan order 123 based on the quantization coefficient expected value 122 calculated by the quantization coefficient expected value calculation unit 22 or the statistical information 103 for each frequency of the orthogonal transform coefficient. And the expected quantized coefficient value 124 rearranged based on the calculated scan order 123 is output to the VLC table calculation unit 24.

In FIG. 2, the VLC table calculation unit 24 calculates an appropriate VLC table 125 based on the expected values 124 of the quantized coefficients rearranged in the scan order by the scan order calculation unit 23.
FIG. 8 is a flowchart showing processing when the VLC table calculation unit 24 calculates an appropriate VLC table 125. In the case of MPEG-2, two types of VLC tables can be arbitrarily selected, but by performing the processing shown in this flowchart, it is possible to select a VLC table 125 with good coding efficiency. Here, the two types of VLC tables defined in MPEG-2 are a VLC table that assigns variable length codes to pairs of run lengths and levels that occur relatively frequently, and an intra macroblock code at a high rate. This is a VLC table in which variable-length codes are assigned to a set of run lengths and levels having a large variance assuming that

  In step ST31 in FIG. 8, the run length and level are obtained from the expected value 124 of the quantized coefficients rearranged in the scan order. In step ST32, the sum of the code amount when the expected value 124 of the quantized coefficient rearranged using each VLC table is encoded is calculated. In step ST33, the VLC table having the smallest calculated code amount is selected and output as an appropriate VLC table 125.

  In the case of MPEG-2, an appropriate VLC table 125 is selected from two types of VLC tables as shown in FIG. 8, but a VLC table can be freely constructed in cases other than MPEG-2. In the case of the encoding method, the occurrence frequency of the run length and level pair is obtained from the expected value 124 of the quantized coefficients rearranged in the scan order, and the variable length shorter than the frequent run length and level pair is obtained. The VLC table may be updated so as to assign a code and output as an appropriate VLC table 125.

  In the above example, the VLC table calculation unit 24 calculates the appropriate VLC table 125 based on the expected values 124 of the quantized coefficients rearranged in the scan order by the scan order calculation unit 23. Calculates the appropriate VLC table 125 based on the statistical information 103 for each frequency of the orthogonal transform coefficient, for example, the average value or the minimum value, instead of the expected value 124 of the quantized coefficient rearranged in the scan order. May be. However, in this case, the orthogonal transform coefficient statistical information calculation unit 5 appropriately corrects the statistical information 103 for each frequency of the orthogonal transform coefficient so as to match the range of coding parameter values defined in advance. The VLC table calculation unit 24 determines an appropriate VLC table 125 after determining a scan order that reduces the code amount based on the frequency distribution of the statistical information 103 for each frequency of the orthogonal transform coefficient. It shall be calculated.

  In this way, the VLC table calculation unit 24 selects an appropriate VLC based on the expected value 124 of the quantization coefficient rearranged in the scan order by the scan order calculation unit 23 or the statistical information 103 for each frequency of the orthogonal transform coefficient. The table 125 is calculated, and the code amount expected value 126 is calculated and quantized by assigning a variable length code to the quantized coefficient expected value 124 rearranged in the scan order using the calculated VLC table 125. The data is output to the scale calculator 25.

  In FIG. 2, the quantization scale calculation unit 25 calculates an appropriate quantization scale 127 based on the expected value 126 of the code amount calculated by the VLC table calculation unit 24. That is, the quantization scale calculation unit 25 corrects the set quantization scale when the code amount expectation value 126 is large, and sets the code amount expectation value 126 when the code amount expectation value 126 is small. By correcting the quantization scale so as to be small, the amount of generated code can be stabilized and the image quality can be improved. By stabilizing the amount of code generated in this way, for example, when the amount of code is smaller than the bit rate, waste of the amount of code due to insertion of stuffing bits, or of the encoded bit string 106 stored in the transmission buffer 9 is achieved. It is possible to prevent non-uniform quantization accuracy due to code amount fluctuations.

  The calculated appropriate quantization scale 127 is output to the encoding parameter setting unit 26 and is also output to the quantization coefficient expected value calculation unit 22, and the calculated appropriate quantization scale 127 uses the quantization coefficient 127. The target quantizer_scale when the expected value calculation unit 22 calculates the expected value 122 of the quantization coefficient using the above equation (4) is corrected.

  In the above example, the quantization scale calculation unit 25 calculates the appropriate quantization scale 127 based on the expected value 126 of the code amount calculated by the VLC table calculation unit 24, but the quantization scale calculation unit 25 Instead of the expected value 126 of the code amount calculated by the VLC table calculation unit 24, an appropriate quantization scale 127 is calculated based on the statistical information 103 for each frequency of the orthogonal transform coefficient, for example, the average value or the minimum value. May be. However, in this case, the orthogonal transform coefficient statistical information calculation unit 5 appropriately corrects the statistical information 103 for each frequency of the orthogonal transform coefficient so as to match the range of coding parameter values defined in advance. The quantization scale calculation unit 25 calculates and calculates an appropriate scan order and a VLC table that reduce the code amount based on the frequency distribution of the statistical information 103 for each frequency of the orthogonal transform coefficient. It is assumed that an appropriate quantization scale 127 is calculated after calculating the expected value of the code amount using the VLC table 125.

  In FIG. 2, the encoding parameter setting unit 26 displays the calculated appropriate quantization matrix 121, the appropriate scan order 123, the appropriate VLC table 125, and the appropriate quantization scale 127 with appropriate codes at appropriate timings. Is set as the quantization parameter 104, and an appropriate quantization matrix 121 and an appropriate quantization scale 127 are output to the quantization unit 7 and the inverse quantization unit 10, and an appropriate quantization matrix 121, an appropriate scan order 123, and an appropriate VLC are output. The table 125 and the appropriate quantization scale 127 are output to the VLC unit 8.

  In FIG. 1, the quantization unit 7 outputs from the orthogonal transform unit 4 based on the appropriate encoding parameter 104 output from the encoding parameter control unit 6, that is, the appropriate quantization matrix 121 and the appropriate quantization scale 127. The quantized orthogonal transform coefficient 102 is quantized and a quantized coefficient 105 which is an orthogonal transform coefficient after quantization is output. The VLC unit 8 receives the proper coding parameters 104 output from the coding parameter control unit 6, that is, the proper quantization matrix 121, the proper scan order 123, the proper VLC table 125, and the proper quantization scale 127. Then, based on the input proper scan order 123 and the appropriate VLC table 125, the quantization coefficient 103 from the quantization unit 7, the quantization matrix 121 added as a header, the scan order 123, the VLC table 125, and the quantization Encoding related information such as the scale 127 is variable-length encoded. The transmission buffer 9 stores a coded bit string 106 that has been subjected to variable length coding.

  As described above, according to the first embodiment, the orthogonal transform coefficient statistical information calculation unit 5 calculates the statistical information 103 for each frequency of the orthogonal transform coefficient, and the encoding parameter control unit 6 calculates the orthogonal transform. Based on the statistical information 103 for each frequency of the coefficient, an appropriate encoding parameter 104 is calculated, and the quantization unit 7 and the VLC unit 8 perform quantization and encoding using the calculated appropriate encoding parameter 104. By doing so, the compression rate can be improved without significantly degrading the subjective image quality.

  Further, according to the first embodiment, the quantization matrix calculation unit 21 calculates an appropriate quantization matrix 121 based on the statistical information 103 for each frequency of the orthogonal transform coefficient, so that the frequency component where power is concentrated is obtained. Since the quantization accuracy becomes rough, it is possible to prevent an enormous amount of code from being generated at a specific frequency component, to improve the compression rate, and to reduce the quantization accuracy of low frequency components in the calculation. By using such an inclined coefficient set that is maintained, the compression ratio can be improved while suppressing deterioration of the subjective image quality.

  Further, according to the first embodiment, the quantization coefficient expected value 122 is calculated by the quantization matrix 121 calculated by the quantization coefficient expected value calculation unit 22, and the scan order calculation unit 23 calculates the expected quantization coefficient. By calculating the appropriate scan order 123 of the quantized coefficients based on the value 122 or the orthogonal transform coefficient statistical information 103, it is possible to suppress the generation of redundant code amounts and to improve the compression rate. .

  Furthermore, according to the first embodiment, the VLC table calculation unit 24 refers to the expected value 124 of the quantized coefficient rearranged in the scan order or the statistical information 103 for each frequency of the orthogonal transform coefficient to obtain an appropriate VLC. By calculating the table 125, it is possible to assign a variable length code suitable for the image data 101, and it is possible to obtain an effect that the compression rate can be dramatically improved.

  Furthermore, according to the first embodiment, the quantization scale calculation unit 25 performs the variable length coding on the expected value 124 of the quantized coefficient rearranged in the scan order, or the expected code value 126 or By calculating an appropriate quantization scale 127 based on the orthogonal transform coefficient statistical information 103, it is possible to stabilize the amount of generated code, and to obtain a stable image quality even when the compression rate is increased. The effect that it can be obtained.

Embodiment 2. FIG.
FIG. 9 is a block diagram showing a configuration of an image coding apparatus according to Embodiment 2 of the present invention. Compared with FIG. 1 of the first embodiment, this image coding apparatus includes an orthogonal transform coefficient statistical information calculation unit 17 instead of the frequency analysis unit 16 and the orthogonal transform coefficient statistical information calculation unit 5 shown in FIG. The other configurations are the same as those in FIG.
Further, the block diagram showing the internal configuration of the encoding parameter control unit 6 in the second embodiment is the same as FIG. 2 in the first embodiment.

Next, the operation will be described.
The frequency analysis unit 16 performs frequency analysis on the image data 101 stored in the frame memory 1 using means such as discrete Fourier transform, discrete cosine transform, or bandpass filter, and is quantized and encoded in the future. The orthogonal transformation coefficient frequency distribution information 131 to be calculated is calculated in advance and output to the orthogonal transformation coefficient statistical information calculation unit 17.

  The orthogonal transform coefficient statistical information calculation unit 17 inputs the orthogonal transform coefficient 102 output from the orthogonal transform unit 4 and the frequency distribution information 131 of the orthogonal transform coefficient calculated by the frequency analysis unit 16. Based on the orthogonal transform coefficient 102, statistical information 103 for each frequency of the orthogonal transform coefficients aggregated in the past N picture (N: arbitrary real number) period or N slice period is calculated in the same manner as in the first embodiment. Then, based on the frequency distribution information 131 of the orthogonal transform coefficient to be quantized and encoded in the future calculated by the input frequency analysis unit 16, the statistical information 103 for each frequency of the calculated orthogonal transform coefficient is corrected, and orthogonal transform is performed. The corrected statistical information 132 for each frequency of the coefficient is output.

  Based on the corrected statistical information 132 for each frequency of the orthogonal transform coefficient, the encoding parameter control unit 6 calculates an appropriate encoding parameter 104 in the same manner as in the first embodiment, and the quantization unit 7, VLC Output to the unit 8 and the inverse quantization unit 10. Other operations are the same as those in the first embodiment.

  As described above, according to the second embodiment, the orthogonal transform coefficient statistical information calculation unit 17 calculates the orthogonal transform coefficients for each frequency accumulated in the past N picture (N: arbitrary real number) period or N slice period. The statistical information 103 is corrected based on the frequency distribution information 131 of the future orthogonal transform coefficient calculated by the frequency analysis unit 16, and the encoding parameter control unit 6 is based on the corrected statistical information 132 for each frequency of the orthogonal transform coefficient. Compared to calculating the encoding parameter 104 using only the statistical information 103 for each frequency of the orthogonal transform coefficients aggregated in the past N picture period or N slice period by calculating the appropriate encoding parameter 104 Thus, it is possible to calculate a more appropriate encoding parameter 104 and to obtain a stable image quality even when the compression rate is increased. It is obtained.

  The image coding apparatus according to the first embodiment and the second embodiment includes a computer and a coding parameter calculation program for causing the computer to function as each unit in the first embodiment and the second embodiment. Can also be realized.

It is a block diagram which shows the structure of the image coding apparatus by Embodiment 1 of this invention. It is a block diagram which shows the internal structure of the encoding parameter control part of the image coding apparatus by Embodiment 1 of this invention. It is a figure which shows the specific example of the coefficient set which the encoding parameter control part of the image coding apparatus by Embodiment 1 of this invention uses for calculation of a quantization matrix. It is a figure which shows the scanning order of the quantization coefficient prescribed | regulated by MPEG-2. It is a flowchart which shows the process at the time of the encoding parameter control part of the image coding apparatus by Embodiment 1 of this invention selecting an appropriate scanning order. It is a figure which shows the example of the division of the horizontal frequency area of the expected value of a quantization coefficient when the encoding parameter control part of the image coding apparatus by Embodiment 1 of this invention selects an appropriate scanning order, and a vertical frequency area. is there. It is a flowchart which shows another process at the time of the encoding parameter control part of the image coding apparatus by Embodiment 1 of this invention selecting an appropriate scanning order. It is a flowchart which shows the process at the time of the encoding parameter control part of the image coding apparatus by Embodiment 1 of this invention calculating an appropriate VLC table. It is a block diagram which shows the structure of the image coding apparatus by Embodiment 2 of this invention.

Explanation of symbols

  1 frame memory, 2 subtraction unit, 3 switch, 4 orthogonal transform unit, 5 orthogonal transform coefficient statistical information calculation unit, 6 encoding parameter control unit, 7 quantization unit, 8 VLC unit, 9 transmission buffer, 10 inverse quantization unit , 11 Inverse orthogonal transform unit, 12 Adder unit, 13 Switch, 14 Frame memory, 15 Motion compensation unit, 16 Frequency analysis unit, 17 Orthogonal transform coefficient statistical information calculation unit, 21 Quantization matrix calculation unit, 22 Expected quantization coefficient Calculation unit, 23 scan order calculation unit, 24 VLC table calculation unit, 25 quantization scale calculation unit, 26 encoding parameter setting unit, 101 image data, 102 orthogonal transform coefficient, 103 statistical information for each frequency of orthogonal transform coefficient, 104 encoding parameters, 105 quantization coefficients, 106 encoded bit strings, 111 orthogonal transform coefficients, 112 Image data, 113 addition data, 114 reference image data, 115 prediction reference image data, 121 quantization matrix, 122 expected value of quantization coefficient, 123 scan order, expected value of quantization coefficient rearranged in 124 scan order, 125 VLC table, 126 expected value of code amount, 127 quantization scale, 131 frequency distribution information of orthogonal transform coefficient, 132 corrected statistical information for each frequency of orthogonal transform coefficient.

Claims (9)

An orthogonal transform unit that orthogonally transforms image data and outputs an orthogonal transform coefficient;
An orthogonal transform coefficient statistical information calculation unit that calculates statistical information for each frequency of the output orthogonal transform coefficient;
An image encoding apparatus comprising: an encoding parameter control unit that calculates an encoding parameter used in quantization and encoding based on the calculated statistical information for each frequency of the orthogonal transform coefficient.   The orthogonal transform coefficient statistical information calculation unit, as statistical information for each frequency of the orthogonal transform coefficient, an average value for each frequency which is an orthogonal transform coefficient in N picture periods or N slice periods (N is an arbitrary real number), 2. The image encoding apparatus according to claim 1, wherein a maximum value, a minimum value, a variance value, a deviation value, an intermediate value, or a total value is calculated.   The encoding parameter control unit includes a first coefficient set set in consideration of audiovisual characteristics, a second coefficient set set in consideration of statistical information for each frequency of the orthogonal transform coefficient, and experimental data The image coding apparatus according to claim 1, further comprising: a quantization matrix calculating unit that calculates a quantization matrix based on the above.   The coding parameter control unit is a quantization that is an orthogonal transform coefficient after quantization based on statistical information for each frequency of the orthogonal transform coefficient, the calculated quantization matrix, and a set quantization scale. A quantized coefficient expected value calculating unit that calculates an expected value of the coefficient, and a scan order calculating unit that calculates the scan order of the calculated expected value of the quantized coefficient so that the code amount at the time of encoding is reduced. The image encoding apparatus according to claim 3, further comprising:   The encoding parameter control unit includes a VLC table calculation unit that calculates a VLC table that minimizes a sum of code amounts when encoding the expected values of the quantized coefficients rearranged in the calculated scan order. The image coding apparatus according to claim 4, further comprising:   The encoding parameter control unit includes a quantization scale calculation unit that corrects a set quantization scale based on an expected value of a code amount encoded by the calculated VLC table. The image encoding device according to claim 5. A frequency analysis unit that performs frequency analysis of the image data and calculates frequency distribution information of the image data;
An orthogonal transform unit that orthogonally transforms the image data and outputs an orthogonal transform coefficient;
Statistical information for each frequency of the output orthogonal transform coefficient is calculated, and based on the calculated frequency distribution information, the calculated statistical information for each frequency of the orthogonal transform coefficient is corrected, and the orthogonal transform coefficient An orthogonal transform coefficient statistical information calculation unit that outputs corrected statistical information for each frequency;
An image encoding apparatus comprising: an encoding parameter control unit that calculates an encoding parameter used in quantization and encoding based on the statistical information corrected for each frequency of the output orthogonal transform coefficient. A computer to calculate encoding parameters for use in quantization and encoding;
Orthogonal transform means for orthogonally transforming image data and outputting orthogonal transform coefficients;
Orthogonal transform coefficient statistical information calculating means for calculating statistical information for each frequency of the output orthogonal transform coefficient;
Coding parameter control means for calculating a coding parameter to be used for quantization and coding based on the calculated statistical information for each frequency of the orthogonal transform coefficient,
An encoding parameter calculation program for functioning as A computer to calculate encoding parameters for use in quantization and encoding;
Frequency analysis means for calculating frequency distribution information of the image data by performing frequency analysis of the image data;
Orthogonal transform means for orthogonally transforming the image data and outputting orthogonal transform coefficients;
Statistical information for each frequency of the output orthogonal transform coefficient is calculated, and based on the calculated frequency distribution information, the calculated statistical information for each frequency of the orthogonal transform coefficient is corrected, and the orthogonal transform coefficient Orthogonal transform coefficient statistical information calculating means for outputting corrected statistical information for each frequency,
Coding parameter control means for calculating a coding parameter to be used for quantization and coding based on the corrected statistical information for each frequency of the output orthogonal transform coefficient,
An encoding parameter calculation program for functioning as

Images