JPH09284570A - Image signal coding method and image signal decoding method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce a data amount and a bit amount and to output data close to an original image by estimating a value of a grating point based on characteristic point information so as to decode coded data having no correction information of the grating point. SOLUTION: Coded data S7 are given to a buffer memory 21 and a variable length decoding circuit 22, from which a characteristic point signal S10 is generated and the signal is outputted to a shunt output circuit 23. After the S10 is divided into position information and quantization coefficient S11, the signal S11 is given to an inverse quantization device 24, in which inverse quantization is conducted to generate a decoding coefficient S12. A grating picture element estimate circuit 71 estimates a grating point value and gives it to an interpolation device 25 as a sample coefficient D1. Then the decoded coefficient S12 and the sample coefficient D1 are used to interpolate other points than the characteristic point and interpolation coefficient D2 is given to an inverse conversion circuit 26, from which an output signal D3 is outputted by using the interpolation coefficient D2 as it is or using it after inverse weblet transformation is applied thereto. Thus, the coded data amount is reduced to reduce the bit amount and the data close to the original image are outputted.

Description

Detailed Description of the Invention

[0001]

[Table of Contents] The present invention will be described in the following order. TECHNICAL FIELD OF THE INVENTION Conventional Technology Problems to be Solved by the Invention (FIGS. 7 to 14) Means for Solving the Problems Embodiments of the Invention (1) First Example (FIGS. 1 to 3) (2) ) Second embodiment (Figs. 4 to 6) (3) Other embodiments Effect of the invention

[0002]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image signal coding method and an image signal decoding method, and more particularly to a high efficiency coding of an image signal by detecting a characteristic point of an image and coding the information of the characteristic point. It can be applied when converting.

[0003]

2. Description of the Related Art Conventionally, a DCT (Discrete Cosine Transform) is applied to an input image as a method for highly efficient encoding of an image signal.
The orthogonal transform is performed according to the method, and the adaptive quantization is performed for each frequency band according to the human visual characteristics, or the image is sub-band divided by the wavelet base, and each band is weighted and encoded. Method is used. According to these methods, the distortion is visually inconspicuous and a high compression rate can be obtained.

However, when the compression rate is further increased, there is a drawback that block distortion and other unfavorable visual effects become remarkable. Therefore, as a coding method that does not produce visually undesired distortion even at high compression rates, so-called structure extraction coding by detecting feature points of an image structure and efficiently coding it The scheme is used.

[0005]

For example, as shown in FIG. 7, a structure extraction coding apparatus 1 for detecting a feature point of an image and coding the image signal is a smoothing filter, a wavelet or the like for an input image signal S1. The filter coefficient S2 is generated by passing through the filter 2 which is a band division filter of No. 1 and is sent to the quantizer 3 and the two-dimensional feature point detection circuit 4. The quantizer 3 quantizes the filter coefficient S2 to generate a quantized coefficient S3, which is sent to the select multiplexing circuit 5. Further, the two-dimensional feature point detection circuit 4 detects the feature point from the filter coefficient S2, and as a result, if the current signal is detected as the feature point, flag 1 is selected. It is sent to the circuit 5.

The select multiplex circuit 5 has a select signal S.
Quantization coefficient S3 for feature points when 4 is flag 1
And the chain-coded feature point coordinates are multiplexed and sent to the variable length coding circuit 6 as a feature point signal S5. The variable length coding circuit 6 performs entropy coding on the feature point signal S5, and sends the resulting variable length coded feature point signal S6 to the buffer memory 7. The buffer memory 7 smoothes the information amount of the variable length coded feature point signal S6 and outputs it as the coded data S7.

The image data coded by the structure extraction coding device 1 is decoded by the structure extraction decoding device 20 as shown in FIG. That is, the structure extraction / decoding apparatus 20 smoothes the information amount of the encoded data S7 output from the structure extraction / encoding apparatus 1 by the buffer memory 21, and then performs the variable length decoding by the variable length decoding circuit 22 that follows. Then, the characteristic point signal S10 is obtained and is sent to the shunt output circuit 23.

The shunt output circuit 23 divides the feature point signal S10 into position information and quantized coefficient information, and then sends the quantized coefficient S11 to the subsequent dequantizer 24 in the line scan order. The inverse quantizer 24 inversely quantizes the quantized coefficient S11 to generate a restoration coefficient S12, which is interpolated by the interpolator 25.
To send to. The interpolator 25 interpolates the values of points other than the feature points based on the restoration coefficient S12, and sends the interpolation coefficient S13 to the inverse conversion circuit 26. The inverse conversion circuit 26 uses the interpolation coefficient S13.
The output signal S14 is output to, for example, a television monitor as it is or after being subjected to conversion processing such as inverse wavelet conversion based on the above.

Here, the interpolator 25 is constructed as shown in FIG. That is, the interpolator 25 temporarily stores the input restoration coefficient S12 in the previous frame buffer 30, and then interpolates the points other than the feature points by filtering with a diffusion filter 31 having a low-pass filter coefficient as shown in FIG. To do. The post-interpolation diffusion restoration coefficient S15 is stored in the post-frame buffer 32, transferred to the previous frame buffer 30 at the time when the processing for one frame is completed, and thereafter subjected to the same diffusion filtering processing. The interpolator 25 sends the interpolation coefficient S13 to the inverse conversion circuit 26 after repeating the same processing several tens of times in order to obtain a sufficient restored image quality.

A structure extraction coding device 40 as shown in FIG. 11 has been proposed. That is, in FIG. 11 in which parts corresponding to those in FIG. 7 are assigned the same reference numerals, the structure extraction encoding device 40 uses the subsampler 41 to filter the filter coefficient S.
2 is sampled at a predetermined interval, and the information of the lattice points obtained as a result is used as a sample coefficient S20 to select multiplexer 4.
Send to 2. The select multiplexing circuit 42 multiplexes the quantized coefficient S3 for the feature point when the select signal S4 from the two-dimensional feature point detection circuit 4 is flag 1 and the coordinates of the feature point, and also multiplexes the sample coefficient S20. Turned into
The characteristic point signal S21 is sent to the variable length coding circuit 6. The variable-length coding circuit 6 entropy-codes the feature point signal S21, and sends the variable-length coded feature point signal S22 obtained as a result to the buffer memory 7. The buffer memory 7 smoothes the information amount of the variable length coded feature point signal S22 and outputs it as the coded data S23.

As a decoding device for decoding this encoded data S23, a structure extraction decoding device 5 having the configuration shown in FIG.
0 has been proposed. The structure extraction / decoding device 50 smoothes the information amount of the coded data S23 by the buffer memory 51 and then performs variable length decoding by the subsequent variable length decoding circuit 52 to obtain the feature point signal S24. To the shunt output circuit 53. The shunt output circuit 53 divides the feature point signal S24 into a sample coefficient S26 and position information and quantized coefficient information, sends the sample coefficient S26 to the interpolator 55, and sets the quantized coefficient information as a quantized coefficient S25 in the line scan order. It is sent to the inverse quantizer 54.

The dequantizer 54 dequantizes the quantized coefficient S25 to generate a restoration coefficient S27, and sends this to the interpolator 55. The interpolator 55 generates an interpolation coefficient S28 by interpolating the values of points other than the feature points based on the restoration coefficient S27 and the sample coefficient S26, and sends this to the inverse conversion circuit 56. The inverse conversion circuit 56 uses the interpolation coefficient S2.
8 is output as it is or after being subjected to conversion such as inverse wavelet conversion, as an output signal S29.

Here, the interpolator 55 is constructed as shown in FIG. The interpolator 55 temporarily accumulates the sample coefficient S26 and the restoration coefficient S27 in the frame buffer 60, and then, by the horizontal grid interpolator 61, the same intervals as the sampling intervals for the pixels sampled and the pixels other than the feature points. The pixels on the grid are interpolated using the sampled pixels and the feature points. Horizontal grid interpolator 61
Using the sample pixels S _{00 to} S _{m * n} in the frame buffer 60 as shown in FIG. 14 and the restoration coefficients A _{0 to} A _i of the feature points, the horizontal sections H _{1 to} H _m of the sample pixels. _{At * n} , interpolation is performed for points other than the sample pixel and the restoration coefficient of the feature point.

Here, the horizontal grid interpolator 61 uses the section H ₁
The coordinates of an arbitrary point other than the sample pixel of H _{m * n} and the restoration coefficient are (X _a , Y _a ), and the sample pixel or restoration coefficient having the same X coordinate and the Y coordinate smaller than Y _a is the closest. The coordinates of the object are (X _a , Y _b ), and these coordinates (X _a , Y _b ).
Let A _{b be} the coefficient value of X, and the coordinates of the nearest sample pixel or restoration coefficient having the same X coordinate and Y coordinate larger than Y _a are (X _a , Y _c ), and this coordinate (X _a ,
When the coefficient value of Y _c ) is A _c , the coordinates (X _a , Y _a )
The coefficient value A _a at the point of

[Equation 1] Or

[Equation 2] The interpolated value in the horizontal direction is written in the frame buffer 60 by performing the interpolation as described above and sending this as the horizontal grid interpolator output S30 to the frame buffer 60.

The vertical grid interpolator 62 includes sample pixels S ₀₀ -S in the frame buffer 60 as shown in FIG.
_{Using m * n} and the restoration coefficients A _{0 to} A _i of the feature points, interpolation is performed for points other than the restoration coefficients of the sample pixels and the feature points in the vertical sections V _{1 to} V _{m * n} of the sample pixels. Here, the vertical grid interpolator 62 sets the coordinates of any point other than the sample pixels and the restoration coefficient in the sections V _{1 to} V _{m * n} to (X _a , Y
_a), X-coordinate at the same Y coordinate of the X _a smaller sample pixel or coefficient of restitution, the coordinates of which the nearest (X _b, Y _a), the coordinates (X _b, the coefficient values of Y _a) A
and is _b, among the X coordinate at the same Y coordinate of X _a larger sample pixel or coefficient of restitution, the coordinates of which the nearest (X _a, Y _c), the coordinates (X _a, Y _c) the coefficient values of Assuming A _c , the coefficient value A _a of the point of coordinates (X _a , Y _a )
Is given by

(Equation 3) Or

(Equation 4) The vertical interpolation value is written in the frame buffer 60 by interpolating as described above and sending it to the frame buffer 60 as the vertical grid interpolator output S31.

The two-dimensional interpolator 63 performs interpolation on points other than the sample pixel, the feature point restoration coefficient, and the interpolation end pixel after the interpolation on the vertical and horizontal grids is completed. That is, the two-dimensional interpolator 63 sets the coordinates of any point other than the sample pixel, the restoration coefficient of the feature point, and the interpolation end pixel to (X _a ,
Y _a ), the coordinates of the sample pixel, the restoration coefficient or the interpolation end pixel, which is the closest in the same X coordinate and whose Y coordinate is smaller than Y _a, are (X _a , Y _b ), and these coordinates (X _a , Y
The coefficient values of _b) A _yb, Y coordinate Y _a larger sample pixels at the same X-coordinate, of the coefficient of restitution or interpolation end pixel, the coordinates of which the nearest (X _a, Y _c), the coordinates ( X _a , Y _c ) has a coefficient value of A _yc , _a sample pixel having the same Y coordinate and an X coordinate smaller than X _a , the restoration coefficient or the interpolation end pixel, whichever is the closest to the coordinate is (X _b , Y _c) .
_a ), the coefficient value of this coordinate (X _b , Y _a ) is A _xb , the same Y
X coordinate coordinate X _a larger sample pixel, among the restored coefficients or interpolation end pixel, the coordinates of which the nearest (X _c, Y _a), the coordinates (X _c, Y _a) the coefficient values of A
_{When xc} is set, the coefficient value A of the point of coordinates (X _a , Y _a )
_a is

(Equation 5) Or

(Equation 6) The two-dimensional interpolation value is written in the frame buffer 60 by interpolating as described above and sending it to the frame buffer 60 as the two-dimensional interpolator output S32. The interpolator 55 finally outputs the contents of the frame buffer 60 after the two-dimensional interpolation is completed as the interpolator output S28 in the line scan order. Thus, the interpolator 55 can obtain the interpolated pixel without performing the iterative process by performing the interpolating process using the sample pixel and the feature point, and as a result, a restored image with good image quality can be obtained in a short time.

In the structure extraction / decoding device 20 (FIG. 8) as described above, the interpolator 25 repeats the diffusion filtering process many times to obtain a restored image close to the original image. There is. However,
There is a problem that real-time processing becomes difficult due to the time required for repeated processing.

Further, although the structure extraction / decoding device 50 (FIG. 12) has an advantage that a restored image can be obtained by one-time interpolation processing, it is necessary to code the information of the lattice points on the coding side. There was a problem that the amount of bits increased.

The present invention has been made in consideration of the above points, and is intended to propose an image signal coding method and an image signal decoding method which can obtain a sufficiently restored image by processing in a relatively short time. is there.

[0020]

In order to solve such a problem, in the present invention, an image signal coding method for detecting a feature point from an input image signal and coding the information of the feature point is converted into feature point information. Based on this, the pixel values of the grid points that are separated by a fixed interval on the image are estimated, the pixel value of the original image and the estimated pixel value are compared, and the absolute value of the difference between the pixel value of the original image and the estimated pixel value is Only for grid points that are greater than or equal to a predetermined threshold value, the coordinates of the grid point and the pixel value of the original image are multiplexed and encoded as feature point information as grid pixel correction information to reduce the amount of data to be coded and the bit. The amount can be reduced and image data closer to the original image can be output.

Furthermore, according to the present invention, a feature point is detected from an input image signal, and a coded signal formed by coding information of the feature point is decoded and decoded by an image signal decoding method. Estimate the pixel values of the grid points that are separated by a fixed interval on the image based on the feature point information,
A decoded image can be obtained by interpolating the values of points other than the feature points based on the pixel values of the grid points. As a result, 1 is obtained from the coded signal generated without the information on the grid points.
A restored image closer to the original image can be obtained in real time by the decoding process in a short time by performing the interpolation process once.

[0022]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below in detail with reference to the drawings.

(1) First Embodiment In FIG. 1 in which parts corresponding to those in FIG. 8 are assigned the same reference numerals, the structure extraction / decoding device 70 uses the above structure extraction / coding device 1 (FIG. 7) to generate a lattice. The restored image data is obtained by decoding the coded data S7 generated without coding the point correction information. Here, the grid point correction information is a difference value between the pixel value of the grid point estimated based on the feature point signal and the pixel value of the original image.

The structure extraction / decoding device 70 has a lattice pixel estimation circuit 71, and inputs the restoration coefficient S12 output from the inverse quantizer 24 to the lattice pixel estimation circuit 71. The lattice pixel estimation circuit 71 estimates the value of the lattice point based on the restoration coefficient S12 and sends it to the interpolator 25 as the sub-sample coefficient D1. The interpolator 25 interpolates the values of points other than the feature points by using the restoration coefficient S12 and the sub-sampling coefficient D1, and the interpolation coefficient D
2 is sent to the inverse conversion circuit 26. Inverse conversion circuit 26
Is designed to output the interpolation coefficient D2 as it is or after performing inverse wavelet conversion or the like to the television monitor or the like as an output signal D3.

Here, the grid pixel estimation circuit 7 described above in FIG.
1 is shown. In the lattice pixel estimation circuit 71, the restoration coefficient S12 output from the inverse quantizer 24 is temporarily stored in the frame buffer 72 for one frame and then sent to the modified lattice thinning circuit 73. The modified grid thinning circuit 73 is
For all grid points separated by a predetermined interval, search whether or not feature point pixels exist around the predetermined range of each grid point, and if no feature point pixel exists, "-1",
When the feature point pixel exists, the pixel value of the feature point is used as the value of the grid point and the corresponding grid frame buffer 74
To be stored.

Here, FIGS. 3A and 3B show input / output examples of the modified lattice thinning-out circuit 73 when the lattice spacing is 5 and the search range is 5, for example. As shown in FIG. 3A, as a result of searching and thinning out whether or not feature point pixels exist around a predetermined range of each grid point for all grid points separated by a predetermined interval, the result of FIG. An output as shown in (B) is obtained.

Subsequently, the pixel interpolating circuit 75 has a value "-1" for the contents stored in the lattice frame buffer 74.
The pixel of No. is rewritten not to "-1" but to the pixel value closest to the pixel of "-1". After that, the lattice frame buffer 74 outputs both the rewritten pixel value and the position information of the lattice point corresponding to the pixel value as the sub-sampling coefficient D1.

In the structure described above, the structure extraction / decoding device 70 searches the lattice point estimation circuit 71 for all the lattice points separated by a prescribed interval for characteristic point pixels existing around the prescribed range of the respective lattice points. Then, the value of the grid point is estimated. Then, the structure extraction / decoding device 70 interpolates the values of points other than the feature points based on the values of the grid points estimated by the grid pixel estimation circuit 71 to generate the correction information of the grid points without coding. A restored image can be obtained from the encoded data S7 thus obtained by one-time interpolation processing.

According to the above configuration, the structure extraction / decoding device 70 interpolates the values of points other than the feature points based on the values of the grid points estimated by the grid pixel estimation circuit 71.
A restored image can be obtained from the encoded data S7 generated without encoding the correction information of the lattice points by one-time interpolation processing, and thus the decoding processing can be performed in real time without repeating the diffusion filtering.

(2) Second Embodiment Next, FIG. 4 shows the structure of a structure extraction encoding device 80 by detecting feature points of an image when encoding correction information of lattice points. The structure extraction coding device 80 generates the filter coefficient S41 by passing the input image signal S40 through a filter 81 which is a band dividing filter such as a smoothing filter or a wavelet, and the quantizer 82, 2
It is sent to the dimensional feature point detection circuit 83 and the sub-sampler 84.

The quantizer 82 quantizes the filter coefficient S41 to generate a quantized coefficient S42, and sends it to the select multiplexing circuit 85. Two-dimensional feature point detection circuit 8
3 detects the feature point from the filter coefficient S41, and if the current signal is detected as the feature point as a result, the flag 1
Is not detected, flag 0 is set to select signal S4
3 to the select multiplex circuit 85. Further, the sub-sampler 84 samples the filter coefficient S41 at regular intervals and sends it to the lattice correction circuit 86 as a sample coefficient S44.

The select multiplexing circuit 85 multiplexes the quantized coefficient S42 for the feature point when the select signal S43 is flag 1 and the coordinates of the feature point, and the feature point signal S.
It is sent as 45 to the lattice correction circuit 86 and the lattice point correction information multiplexing circuit 87. The lattice correction circuit 86 estimates the values of the lattice points on the image, which are estimated from the feature point signal S45, and the sampling coefficient S44 from the subsampler 84.
The value of the grid point in the original image obtained based on the above is compared, and when the difference between the two is greater than or equal to a predetermined threshold value T, the grid pixel correction information including the coordinates of the grid point and the value of the grid point in the original image is calculated. The corrected grid point signal S46 is sent to the grid point correction information multiplexing circuit 87. As a result, the grid correction circuit 86
Is capable of obtaining correction information of grid points, and is capable of outputting information of grid points close to the original image.

The lattice point correction information multiplexing circuit 87 multiplexes the feature point signal S45 and the corrected lattice point signal S46 and sends the multiplexed feature point signal S47 to the variable length coding circuit 88. The variable length coding circuit 88 uses the multiplexed feature point signal S4.
7 is entropy coded and variable length coded feature point signal S
It is sent to the buffer memory 89 as 48. The buffer memory 89 is configured to smooth the information amount of the variable length coded feature point signal S48 and output it as the coded data S49. Thus, the structure extraction decoding device 80
Is designed to output image data closer to the original image.

FIG. 5 shows the structure of the lattice correction circuit 86. The lattice correction circuit 86 sends the feature point signal S45 from the select multiplexing circuit 85 to the lattice pixel estimation circuit 90. The lattice pixel estimation circuit 90 estimates the value of each lattice pixel based on the feature point signal S45, and estimates the estimated lattice pixel signal S50.
To the comparator 91.

The comparator 91 compares the value of the estimated grid pixel signal S50 with the value of the sample coefficient S44 from the subsampler 84. If the difference between the two is greater than a predetermined threshold value T, the coordinates of the grid pixel and the sample coefficient are compared. S44 and S44 are output as a corrected grid point signal S46. By the way, the grid pixel estimation circuit 90 used in FIG.
The grid pixel estimation circuit 71 shown in FIG. 2 has the same structure as the grid pixel estimation circuit 71. The modified grid thinning circuit provided in the grid pixel estimation circuit 90 thins out the data to be encoded and reduces the bit amount. It is done like this.

Subsequently, in FIG. 6, a structure extraction coding device 80 by detecting feature points of an image for coding correction information of lattice points is shown.
1 shows a configuration of an image signal decoding apparatus 100 that decodes coded data S49 coded according to the above. The image signal decoding apparatus 100 obtains the feature point signal S60 by smoothing the information amount of the coded data S49 by the buffer memory 101 and then by variable length decoding by the subsequent variable length decoding circuit 102, This is sent to the shunt output circuit 103. The shunt output circuit 103 divides the feature point signal S60 into position information, quantization coefficient information, and lattice pixel correction information,
The position information and the quantization coefficient information are sent to the inverse quantizer 104 as the quantization coefficient S61 in the line scan order, and the lattice pixel correction information is sent to the lattice correction circuit 105 as the lattice pixel correction information signal S62.

The dequantizer 104 dequantizes the quantized coefficient S61 to generate a restoration coefficient S63, and sends this to the interpolator 106 and the lattice pixel estimation circuit 107. The lattice pixel estimation circuit 107 estimates and obtains the values of lattice points separated by a constant interval on the image based on the restoration coefficient S63, and then corrects the value of the lattice pixel by the lattice correction circuit 105 to obtain the sub-sampling coefficient S64. Is output to the interpolator 106. Here, the grid correction circuit 105 replaces the value of the grid point of the coordinate designated by the grid pixel correction information signal S62 with the correction value and outputs it.

The interpolator 106 interpolates the values of points other than the feature points by the restoration coefficient S63 and the sub-sampling coefficient S64, and sends it to the inverse transform circuit 108 as the interpolation coefficient S65. The inverse conversion circuit 108 uses the interpolation coefficient S65 as it is,
Or, after performing conversion such as inverse wavelet conversion,
The output signal S66 is output to a television monitor or the like. Here, the grid pixel estimation circuit 107
2 has the same configuration as the grid pixel estimation circuit 71 shown in FIG. 2, and the interpolator 106 has the same configuration as the interpolator 55 shown in FIG.

In the above structure, the structure extraction coding device 80 is obtained based on the value of the grid point estimated from the feature point signal S45 by the grid pixel estimation circuit 90 of the grid correction circuit 86 and the sample coefficient S44. Compare with the value of the grid point in the original picture. At this time, when the difference between the two is greater than or equal to a predetermined threshold value T, the grid correction circuit 86 sends the coordinates of the grid point and the value of the grid point in the original image to the grid point correction information multiplexing circuit 87 as a correction grid point signal S46. Send out the characteristic point signal S
45 and the corrected grid point signal S46 are multiplexed and output as a multiplexed feature point signal S47.

As described above, the structure extraction coding device 80 outputs the image data closer to the original image by coding the correction information of the lattice points obtained by the lattice pixel estimation circuit 90 and outputting it as the encoded data S49. can do. Further, the structure extraction encoding device 80 includes a lattice correction circuit 86.
The data to be encoded can be reduced and the amount of bits can be reduced by thinning out the data by the modified lattice thinning-out circuit provided in the lattice pixel estimating circuit 90.

Further, the structure extraction / decoding apparatus 100 receives the image data closer to the original image generated by the structure extraction / encoding apparatus 80, that is, the encoded data S49, and the lattice pixel estimation circuit 107 based on the restoration coefficient S63. After the value of the grid point is estimated and obtained, the grid correction circuit 105 replaces the value of the grid pixel with the correction value and corrects it, and outputs it as the sub-sample coefficient S64 to the interpolator 106. Interpolator 10
6 can output the output signal S66 by interpolating the values of points other than the feature points by using the restoration coefficient S63 and the sub-sampling coefficient S64.

As described above, in the structure extraction / decoding apparatus 100, after the value of the grid point is estimated by the grid pixel estimation circuit 107, the value of the grid pixel is replaced by the correction value by the grid correction circuit 105 and the correction is performed, whereby the structure is obtained. A restored image closer to the original image can be obtained from the image data generated by the extraction encoding device 80 by one interpolation process. Further, the structure extraction / decoding apparatus 100 can perform the decoding process in a short time by decoding the image data having a small bit amount output by the structure extraction / coding apparatus 80.

According to the above configuration, the structure extraction coding device 80 reduces the data to be coded by thinning out the data by the modified lattice thinning circuit provided in the lattice pixel estimating circuit 90 of the lattice modifying circuit 86. By reducing the bit amount and encoding the correction information of the lattice points obtained by the lattice pixel estimation circuit 90, image data closer to the original image can be output.

Further, the structure extraction / decoding apparatus 100 estimates the value of the grid point in the grid pixel estimation circuit 107 and corrects the grid pixel value by the grid correction circuit 105 by replacing the grid pixel value with the correction value. It is possible to obtain a restored image closer to the original image by a decoding process in a short time from the image data closer to the original image generated by the conversion device 80 by one interpolation process.

(3) Other Embodiments In the first and second embodiments described above, the interpolator 25 is used.
And the case where the interpolator 106 sends out the interpolation coefficients D2 and S65 which have been subjected to the grid-like interpolation only once, and outputs the interpolation coefficients D2 and S65 through the inverse transform circuits 26 and 108 to the structure extraction decoding devices 70 and 100. Although described, the present invention is not limited to this, and the obtained interpolation coefficients D2 and S65 may be subjected to diffusion filtering processing several times more.
As a result, the structure extraction / decoding devices 70 and 100 can obtain a decoded image of higher quality.

[0046]

As described above, according to the present invention, after estimating the pixel values of the lattice points separated by a constant interval on the image based on the feature point information, the pixel value of the original image and the estimated pixel value And the pixel values of the original image and the pixel values of the original image are used as the lattice pixel correction information only for the lattice points whose absolute value of the difference between the pixel value of the original image and the estimated pixel value is equal to or greater than a predetermined threshold value. By multiplexing with information and encoding, the amount of data to be encoded can be reduced to reduce the bit amount, and an image signal encoding method capable of outputting image data closer to the original image can be realized.

Further, according to the present invention, the decoding is performed in the image signal decoding method for detecting the characteristic point from the input image signal and decoding the coded signal formed by coding the information of the characteristic point. After estimating the pixel values of grid points separated by a constant interval on the image based on the converted feature point information, interpolate the values of points other than the feature points based on the pixel values of the grid points. As a result, it is possible to realize an image signal decoding method in which a restored image closer to the original image can be obtained in real time by a single short-time decoding process from a coded signal generated without information on grid points by a single interpolation process.

[Brief description of drawings]

FIG. 1 is a block diagram showing a structure extraction / decoding apparatus I by detecting feature points of an image signal according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a lattice pixel estimation circuit according to an embodiment of the present invention.

FIG. 3 is a schematic diagram showing an input / output example of a modified lattice thinning circuit according to an embodiment of the present invention.

FIG. 4 is a block diagram showing a configuration of a structure extraction encoding device when using correction information of lattice points according to an embodiment of the present invention.

FIG. 5 is a block diagram showing a configuration of a lattice correction circuit according to an embodiment of the present invention.

FIG. 6 is a block diagram showing a configuration of a structure extraction / decoding apparatus II when using correction information of lattice points according to an embodiment of the present invention.

FIG. 7 is a block diagram showing a configuration of a structure extraction encoding device I by detecting feature points of an image signal.

FIG. 8 is a structure extraction / decoding device I by detecting feature points of an image.
It is a block diagram which shows a structure of.

FIG. 9 is a structure extraction / decoding device I based on image feature point detection.
3 is a block diagram showing the configuration of an interpolator I in FIG.

FIG. 10 is a schematic diagram showing an example of low-pass filter coefficients in the diffusion filter processing of the interpolator.

FIG. 11 is a block diagram showing the configuration of a structure extraction encoding device II by detecting feature points of an image signal.

FIG. 12 is a structure extraction / decoding device by detecting feature points of an image.
FIG. 3 is a block diagram showing the configuration of II.

FIG. 13 is a structure extraction / decoding device by detecting feature points of an image.
FIG. 3 is a block diagram showing a configuration of an interpolator II in II.

FIG. 14 is a schematic diagram showing the contents of a frame buffer of an interpolator.

[Explanation of symbols]

1, 40, 80 ... Structure extraction coding device, 2, 81 ...
Filter, 3, 82 ... Quantizer, 4, 83 ... Two-dimensional feature point detection circuit, 5, 42, 85 ... Select multiplexing circuit, 6, 88 ... Variable length coding circuit, 7, 21, 51,
89, 101 ... buffer memory, 20, 50, 70,
100 ... Structure extraction / decoding device, 22, 52, 102 ...
... variable length decoding circuit, 23, 53, 103 ... shunt output circuit, 24, 54, 104 ... inverse quantizer, 25, 5
5, 106 ... Interpolator, 26, 56, 108 ... Inverse conversion circuit, 30 ... Front frame buffer, 31 ... Diffusion filter, 32 ... Rear frame buffer, 41, 84 ... Subsampler, 60, 72 ... Frame buffer, 61 ...
... horizontal grid interpolator, 62 ... vertical grid interpolator, 63
Two-dimensional interpolator, 71, 90, 107 ... Lattice pixel estimation circuit, 73 ... Modified lattice thinning circuit, 74 ... Lattice frame buffer, 75 ... Pixel interpolation circuit, 86 ... Lattice correction circuit, 87 ... Grid point correction information multiplexing circuit, 91 ...
... comparator, 105 ... grating correction circuit, S20, S26,
S44, D1 ...... sample coefficients, S ₀₀ ~S _{m * n} ...... sample pixel, A ₀ ~A _i ...... feature points.

Claims (8)

[Claims] 1. An image signal coding method for detecting a feature point from an input image signal and coding the information of the feature point, wherein a grid point of a grid point separated by a constant interval on the image based on the feature point information. Estimate the pixel value, compare the pixel value of the original image with the estimated pixel value, and only for the grid points where the absolute value of the difference between the pixel value of the original image and the estimated pixel value is greater than or equal to a predetermined threshold value. An image signal coding method, wherein the coordinates of the grid point and the pixel value of the original image are multiplexed and coded on the feature point information as grid pixel correction information. 2. When estimating the pixel value of the grid point, first, for a grid point having a feature point within a predetermined range around the grid point, the pixel value of the feature point closest to the grid point is set to the grid point. The image signal encoding method according to claim 1, wherein, for a grid point having no feature point, the value of the grid point closest to the grid point is set as the value of the grid point. 3. When a feature point is detected from an input image signal and the information of the feature point is encoded, the pixel values of grid points separated by a constant interval on the image are estimated based on the feature point information. , Comparing the pixel value of the original image and the estimated pixel value, the absolute value of the difference between the pixel value of the original image and the estimated pixel value is greater than or equal to a predetermined threshold, An image signal decoding method for decoding a coded signal formed by multiplexing the coordinates and the pixel value of the original image as the grid pixel correction information on the feature point information and coding the feature point information, wherein the decoded feature Pixel values of grid points that are separated by a constant interval on the image are estimated based on the point information, and grid points of coordinates that require correction specified by the grid pixel correction information obtained from the feature point information. Replace the value of Picture signal decoding method characterized by obtaining a decoded image by interpolating the values of points other than the feature point based on the value of the recombination was the lattice points. 4. The image signal decoding method according to claim 3, wherein a final decoded image is obtained by subjecting the data of the decoded image to diffusion filtering processing a plurality of times. 5. When estimating the pixel value of the grid point, first, for a grid point having a feature point within a predetermined range around the grid point, the pixel value of the feature point closest to the grid point is set to the grid point. 4. The image signal decoding method according to claim 3, wherein, for a grid point having no feature point, the value of the grid point closest to the grid point is set as the value of the grid point. 6. An image signal decoding method for detecting a feature point from an input image signal and decoding a coded signal formed by encoding information of the feature point, wherein the decoded feature point It is possible to obtain a decoded image by estimating the pixel values of grid points that are separated by a fixed interval on the image based on the information and interpolating the values of points other than the feature points based on the pixel values of the grid points. A characteristic image signal decoding method. 7. When estimating a pixel value of the grid point, first, for a grid point having a feature point within a predetermined range around the grid point, the pixel value of the feature point closest to the grid point is set to the grid point. 7. The image signal decoding method according to claim 6, wherein, for a grid point having no feature point, the value of the grid point closest to the grid point is set as the value of the grid point. 8. The image signal decoding method according to claim 6, wherein a final decoded image is obtained by subjecting the data of the decoded image to diffusion filtering processing a plurality of times.