JP4078652B2 - Image signal decoding apparatus and image signal decoding method - Google Patents

Description

  The present invention relates to an image signal decoding apparatus and an image signal decoding method, and is particularly suitable for application when decoding an image signal that has been subjected to high-efficiency compression encoding using a block encoding method.

  Conventionally, in a so-called image signal transmission system that transmits an image signal to a remote place, such as a video conference system, or in an apparatus that digitizes an image signal and records and reproduces it on a video tape recorder or a video disk recorder, In order to efficiently use recording media, significant information is efficiently encoded using the correlation of digitized image signals to reduce the amount of transmitted information and recorded information, thereby increasing transmission efficiency and recording efficiency. It is made like that.

  In such a case, in general, the amount of data is greatly reduced by performing high-efficiency compression coding of an image signal. As a technique for this high-efficiency compression coding, an input image signal is divided into a plurality of blocks, for example, ADRC coding (Adaptive Dynamic Range Coding) or DCT coding (Discrete Cosine Transform: discrete). Block coding for performing cosine transform coding and the like has been proposed.

  Incidentally, ADRC coding uses visual dynamic range dependence, and in the region where the dynamic range of the pixel distribution is large, minute fluctuations in the pixel level are difficult to see, and conversely in the region where the dynamic range is small. This is an encoding method that compresses an image using a visual characteristic that even a minute fluctuation of a pixel level is easily visible. DCT encoding uses a plurality of cosine functions as the base waveform, and is an encoding method that decomposes and expresses an image signal into these base waveforms.

By the way, when the image is compressed to a very low rate by the above-described method, block distortion may occur in the vicinity of the block boundary during restoration, and image degradation may occur. This tendency is particularly noticeable in flat blocks where the number of allocated bits is small. For this reason, by applying a low-pass filter (so-called low-pass filter) locally to the block boundary, generally, such block distortion is removed to reduce image deterioration.
However, when a low-pass filter is applied, there is a problem that edges and detail portions containing high-frequency components are blurred, which is still insufficient as a method for removing block distortion.

  The present invention has been made in consideration of the above points, and an object of the present invention is to propose an image signal decoding apparatus and an image signal decoding method capable of removing block distortion without distorting edges and details.

In order to solve this problem, in the present invention, in an image signal decoding apparatus for decoding compressed image data obtained by compressing and encoding image data in units of blocks by predetermined compression encoding means, the compressed image data is decoded in units of blocks. Dynamic range detection means for detecting the dynamic range of each block of the decoded image data, and values obtained by compressing and encoding peripheral pixels centering on the target pixel in each block of the decoded image data. a class detection means for detecting a first class to which the decoded image data belongs by a classification code representing a class to be classified when pattern classification, the position indicating the relative position of the target pixel in the block of decoded image data position information detection means for detecting information, Dainamitsu to the classification code of the first class Every second class by adding the elements of the range elements and location information, a strong low-pass filter as when the correlation between the image data and the decoded image data is smaller dynamic range of the block with expressed by linear combination is The coefficient set to be applied is stored in advance, and the class classification code of the first class obtained by the class detecting means is obtained by the dynamic range detecting element obtained by the dynamic range detecting means and the position information detecting means. In accordance with the second class to which the position information element is added, a coefficient table for reading and outputting the coefficient combination, and a coefficient sum output from the coefficient table and the decoded image data are used to calculate the decoded image data. And product-sum operation means for correcting the above.

According to the present invention, in the image signal decoding method for decoding compressed image data obtained by compressing and encoding image data in units of blocks by predetermined compression encoding means, each of the decoded image data obtained by decoding the compressed image data in units of blocks When the dynamic range detection step for detecting the dynamic range of the block and the values obtained by compressing and encoding the peripheral pixels centering on the target pixel in each block of the decoded image data are classified into spatial patterns. A class detection step for detecting the first class to which the decoded image data belongs by using a class classification code representing the class to be classified, and a position for detecting position information indicating the relative position of the pixel of interest within the block of the decoded image data Information detection step and first class classification code dynamic range For each second class to which elements of elements and position information are added, the correlation between the image data and the decoded image data is expressed by a linear primary combination, and a stronger low-pass filter is applied as the block dynamic range is smaller. In accordance with the dynamic range element and position information detection step obtained from the dynamic range detection step, the class classification code of the first class obtained from the class detection step is stored in the coefficient table in which such coefficient sets are stored in advance. In accordance with the second class to which the position information element obtained is added, the coefficient read step for reading and outputting the coefficient set, and the coefficient set output from the coefficient table and the decoded image data are subjected to a product-sum operation, thereby the decoded image. A product-sum operation step for correcting data is provided.

A dynamic second range element and a position information element are added to the detected first class to generate a new second class, a coefficient is read according to the second class, and a product-sum operation is performed on the decoded image data. In addition to the fact that the decoded image data is modified, the image data can be optimally adaptively predicted optimally based on the correlation between the image data before and after encoding and the relative position in the block for each first class. Thus, block distortion generated near the block boundary can be adaptively removed without causing the edges and detail to be smoothed. This is because the block dynamic range is different between the block containing the edge and the detail portion and the block where the block distortion is likely to occur. Can be avoided in advance.

According to the present invention, the dynamic range element and the position information element are added to the detected first class to generate a new second class, the coefficients are read according to the second class, and the decoded image data is generated. Since the decoded image data is modified by performing a product-sum operation, the image data is optimally and adaptively predicted appropriately based on the correlation in the image data before and after encoding and the relative position in the block for each first class. In addition to being able to be processed, the block distortion generated near the block boundary can be adaptively removed without blurring the edges and detail portions.

  Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.

(1) First Embodiment In FIG. 1, reference numeral 1 denotes an image signal decoding apparatus according to the first embodiment as a whole, which converts transmission image data S1 that has been highly efficient compression-encoded by DCT encoding into an error correction circuit (FIG. 1). (Not shown) to be input to the frame decomposition circuit 2. The frame decomposition circuit 2 extracts the horizontal synchronization signal S2 and the vertical synchronization signal S3 from the transmission image data S1, outputs them to the control circuit 3, and supplies the transmission image data S1 to the block decoding circuit 4 at the subsequent stage.

  The block decoding circuit 4 restores the compression-encoded transmission image data S1 in units of blocks. After decoding the variable-length coding, inverse quantization is performed to calculate a coefficient, and inverse DCT conversion is performed to perform original DCT conversion. Restore level data. The restored block-unit image data S4 is supplied to the block delay circuit 5 and the dynamic range detection circuit 6.

In the dynamic range detection circuit 6, the image data S4 is supplied to the maximum value detection circuit 6A and the minimum value detection circuit 6B. The maximum value detection circuit 6A detects the maximum value in each block from the image data S4, and supplies the detected maximum value to the subtractor 6C. The minimum value detection circuit 6B detects the minimum value in each block and supplies the detected minimum value to the subtractor 6C. The subtractor 6C obtains the dynamic range of each block by subtracting the minimum value from the supplied maximum value, and supplies the obtained dynamic range to the dynamic range memory (hereinafter referred to as DR memory) 7. . As a result, the dynamic range of each block is sequentially stored in the DR memory 7.
On the other hand, the block delay circuit 5 is a circuit for performing temporal adjustment, and delays the image data S4 by the time required to detect the dynamic range as described above. The block decomposition memory 8 is supplied.

  The block decomposition memory 8 is composed of, for example, a frame memory, and blocks and stores the supplied image data S5, and reads the stored image data and supplies it to the subsequent product-sum operation circuit 9. At this time, the block decomposition memory 8 converts the read image data into a signal format suitable for the calculation form of the product-sum operation circuit 9 and supplies the signal format to the product-sum operation circuit 9. For example, when the product-sum operation circuit 9 performs one-dimensional arithmetic processing, the read image data is converted into one-dimensional time-series data and supplied. When the product-sum arithmetic circuit 9 performs two-dimensional arithmetic processing, The read image data is converted into parallel time-series data having a desired number of lines and supplied.

  In synchronism with the reading operation of the block decomposition memory 8, the dynamic range (DR) of the block corresponding to the target pixel is read from the DR memory 7. That is, the DR memory 7 reads the block dynamic range of the target pixel read by the block decomposition memory 8, and supplies the read dynamic range to the coefficient table circuit 10 as address information.

  The coefficient table circuit 10 is a memory, and coefficient combinations used for the product-sum operation of the product-sum operation circuit 9 are stored in advance, and the coefficient combination is read out using the supplied dynamic range (DR) as address information, The product-sum operation circuit 9 is supplied. This coefficient combination is a combination of numerical values such that a low-pass filter is realized by the arithmetic processing of the product-sum operation circuit 9. When the dynamic range is large, the cut-off frequency is high, and the dynamic range is small. Is set so that the cut-off frequency is low.

  This is because a block with a small dynamic range is regarded as a flat image, and since low bit allocation is applied, block distortion tends to occur, and in the case of a flat image, a low-pass filter is applied somewhat strongly. However, since it does not affect the blurring of the image, it is based on the principle that the block distortion can be effectively removed by applying a low-pass filter to a block having a small dynamic range. For blocks with a large dynamic range, edges and details are included, so it is possible to effectively prevent image blurring by applying a very weak low-pass filter or letting it pass as it is. It is.

  The coefficient table circuit 10 outputs different coefficient sets near the block boundary and near the block center even under the same block under the control of the control circuit 3. That is, the coefficient table circuit 10 outputs a coefficient set that realizes a low-pass filter near the block boundary, and outputs a coefficient set that passes the image data as it is near the center of the block. As a result, a low-pass filter can be applied only to pixels near the block boundary in the image data.

The product-sum operation circuit 9 uses the image data supplied from the block decomposition memory 8 and the coefficient set supplied from the coefficient table circuit 10 as described above to perform a product-sum operation (Σa _i) with a predetermined number of taps for each pixel of interest. x _i : where a _i is a coefficient and x _i is image data), a low-pass filter is applied to pixels near the block boundary in the image data to remove block distortion from the image data. The image data S6 from which the block distortion has been removed is output. At this time, since the coefficient set supplied from the coefficient table circuit 10 is switched depending on the size of the block dynamic range, the pass band characteristic of the low-pass filter can be appropriately switched, and the block distortion is adaptively removed. can do.

  Incidentally, the signals (S7, S8) supplied from the control circuit 3 to the block decomposition memory 8 and the DR memory 7 are control signals such as an address signal and an enable signal used during the write operation and the read operation. The signal (S9) supplied from the control circuit 3 to the coefficient table circuit 10 recognizes the relative position information of the pixels in the block obtained based on the horizontal synchronizing signal S2 and the vertical synchronizing signal S3, that is, the block boundary. This is identification information. Therefore, the coefficient table circuit 10 recognizes the block boundary based on the position information supplied from the control circuit 3, and switches the coefficient set to be output near the block boundary and near the block center as described above.

  Here, the coefficient combinations output by the coefficient table circuit 10 will be described. For example, when the product-sum operation circuit 9 performs a three-dimensional one-dimensional product-sum operation on the target pixel, as shown in FIG. 2, when the dynamic range is large and the target pixel is near the block boundary, [1/4, 2/4, 1/4] is output, and when the dynamic range is small and the target pixel is near the block boundary, [1/8, 6/8, 1/8] Coefficient combinations are output. When the target pixel is near the center of the block, a coefficient combination consisting of [0, 1, 0] is output regardless of the size of the dynamic range.

  Also, for example, when the product-sum operation circuit 9 performs a (3 × 3) -tap two-dimensional product-sum operation on the target pixel, as shown in FIG. 3, the dynamic range is large and the target pixel is blocked. When near the boundary, [0, 1/6, 0, 1/6, 1/3, 1/6, 0, 1/6, 0] is output, the dynamic range is small, and the pixel of interest Is near the block boundary, a coefficient set of [0, 1/8, 0, 1/8, 1/2, 1/8, 0, 1/8, 0] is output, and the pixel of interest is near the center of the block. Sometimes, a coefficient set consisting of [0, 0, 0, 0, 1, 0, 0, 0, 0] is output regardless of the size of the dynamic range.

  In the above configuration, the image signal decoding apparatus 1 first decodes the transmission image data S1 subjected to the high-efficiency compression coding by the DCT coding by the block decoding circuit 4. Then, using the maximum value detection circuit 6A, the minimum value detection circuit 6B, and the subtractor 6C, the maximum value and the minimum value in each block are detected from the restored image data S4, and the difference between the detected maximum value and the minimum value is obtained. To determine the dynamic range of each block. The obtained dynamic range is supplied as address information to the coefficient table circuit 10 via the DR memory 7. The control circuit 3 obtains the relative position information of the pixels in the block based on the horizontal synchronizing signal S2 and the vertical synchronizing signal S3 and supplies it to the coefficient table circuit 10.

  The coefficient table circuit 10 reads a set of coefficients for determining the pass band characteristics of the low-pass filter stored in advance as the supplied dynamic range as address information, and supplies it to the product-sum operation circuit 9. At that time, the coefficient table circuit 10 recognizes whether the pixel of interest is near the block boundary or near the center of the block based on the supplied position information, and switches the coefficient set to be output. Specifically, if the target pixel is near the block boundary, a coefficient set that realizes a low-pass filter is output, and if it is near the center of the block, a coefficient set that passes is output as it is.

  The product-sum operation circuit 9 performs a product-sum operation using the supplied coefficient set, thereby applying a low-pass filter only to pixels near the block boundary in the image data to remove block distortion. At this time, since the coefficient set output from the coefficient table circuit 10 changes according to the magnitude of the dynamic range, the dynamic range of the dynamic range which is easy to include blocks having a small dynamic range, which is likely to cause block distortion, and edges and detail portions. A large block can change the passband characteristics of the low-pass filter.

  In this way, the image signal decoding apparatus 1 detects the dynamic range which is the difference based on the maximum value and the minimum value in the block, and applies it to the pixels near the block boundary according to the size of the detected dynamic range. The pass band characteristic of the low pass filter to be applied is changed. As a result, the block distortion can be removed adaptively without blurring the edges or the detail portion.

  Incidentally, in the case of this embodiment, since only the processing in the image signal decoding apparatus 1 is required, there is an advantage that there is no burden on the transmission side apparatus and the transmission format need not be changed.

  According to the above configuration, the dynamic range in the block is detected, and the pass band characteristic of the low-pass filter applied to the pixels near the block boundary is changed according to the detected size of the dynamic range. Thus, the block distortion can be adaptively removed without blurring the edges and detail portions.

(2) Second Embodiment In FIG. 4, in which parts corresponding to those in FIG. 1 are assigned the same reference numerals, 20 denotes an image signal decoding apparatus according to the second embodiment as a whole. Block distortion is adaptively removed by performing adaptive prediction processing for each.

  First, in the image signal decoding apparatus 20, as in the first embodiment, the transmission image data S1 that has been compression-encoded with high efficiency by DCT encoding is input to the block decoding circuit 4 via the frame decomposition circuit 2, where To restore the original level data in blocks. The image data S4 restored in units of blocks is supplied to the maximum value detection circuit 6A and the minimum value detection circuit 6B of the dynamic range detection circuit 6, where the maximum value and the minimum value are detected for each block. The detected maximum value and minimum value are supplied to the subtractor 6C, and the dynamic range of each block is obtained by obtaining the difference. The obtained dynamic range is sequentially stored in the DR memory 7 for each block.

  The restored image data S4 is supplied to the block delay circuit 5, where it is delayed by the time required for detecting the dynamic range, and then supplied to the delay / prediction tap memory 21 and the class generation circuit 22. .

  The delay / prediction tap memory 21 stores the supplied image data S5 after block decomposition. The delay / prediction tap memory 21 delays the processing by the processing time of a class generation circuit 22, a DR memory 7 and a control circuit 23, which will be described later, and adjusts the time when reading the stored image data. By collecting the peripheral pixels, a prediction tap required for the product-sum operation is formed, and the stored image data is read out. The delay / prediction tap memory 21 converts the read image data into a signal format suitable for the calculation form of the subsequent product-sum operation circuit 24 and supplies the signal format to the product-sum operation circuit 24. For example, when the product-sum operation circuit 24 performs one-dimensional arithmetic processing, the read image data is converted into one-dimensional time-series data and supplied. When the product-sum arithmetic circuit 24 performs two-dimensional arithmetic processing, The read image data is converted into parallel time-series data having a desired number of lines and supplied.

  In synchronism with the read operation of the delay / prediction tap memory 21 as described above, the dynamic range (DR: for example, 8 bits) of the block corresponding to the target pixel is read from the DR memory 7. The DR memory 7 supplies the coefficient table circuit 25 with the read dynamic range as it is or with n bits shifted down to reduce the number of bits as part of the address information for accessing the coefficient table circuit 25. .

  The control circuit 23 obtains position information indicating the relative position in the block of the pixel of interest based on the horizontal synchronizing signal S2 and the vertical synchronizing signal S3. For example, as shown in FIG. 5, when the block is composed of (8 × 8) pixels, the relative position of the pixels in the block is represented by a total of 6 bits, 3 bits each in length and width. The position information S21 represented in this way is supplied to the coefficient table circuit 25 as a part of address information for accessing the coefficient table circuit 25.

  Further, the class generation circuit 22 sets a spatial cluster centered on the target pixel, and performs class detection by spatial pattern classification by, for example, ADRC encoding the decoded value in the spatial cluster (that is, the class to which the target pixel belongs). The class classification code S22 obtained as a result is supplied to the coefficient table circuit 25 as a part of address information for accessing the coefficient table circuit 25 in the same manner. Incidentally, when ADRC encoding is performed, the dynamic range for each block output from the dynamic range detection circuit 6 is used.

  The coefficient table circuit 25 is a memory, and a coefficient set for each class used for the product-sum operation of the product-sum operation circuit 24 is stored in advance by learning, and the supplied dynamic range DR, position information S21, and The combination of the class classification codes S22 is used as address information, and the coefficient combination for each class is read and supplied to the product-sum operation circuit 24. That is, the coefficient table circuit 25 reads out a coefficient set for each class by using the class classification code S22 plus the dynamic range DR and the position information S21 as a new class classification code.

  The product-sum operation circuit 24 performs a product-sum operation process on the image data supplied from the delay / prediction tap memory 21 and the coefficient set supplied from the coefficient table circuit 25, thereby modifying the image data for each class (so-called so-called). Adaptive prediction processing is performed for each class), and the resulting image data S23 is output. At this time, the coefficient set supplied from the coefficient table circuit 25 is a block dynamic range (DR) that represents characteristics such as block distortion or the presence of an edge or detail portion. Since the position information (S21) indicating the position and the position of the position is taken into consideration, the block distortion can be adaptively removed without blurring the edges and detail portions.

Here, the coefficient combinations stored in the coefficient table circuit 25 are obtained by learning as described below . First, already known first image data and second image data obtained by applying DCT encoding and decoding to the first image data are prepared. Similarly to the above-described class classification, a new class classification code is generated by adding the dynamic range and position information to the class classification code obtained by ADRC encoding. Then, for each of the obtained class classification codes, learning using the least square method is performed on the first and second image data to obtain an optimum coefficient set. The coefficient table circuit 25 is formed by storing the obtained coefficient group in the memory for each class classification code.

  That is, when the coefficient table circuit 25 is formed, a coefficient creating circuit 30 as shown in FIG. 6 is used. First, the already-known image data S30 is input to the coefficient creating circuit 30. The image data S30 is input to the coefficient selection circuit 31 for learning processing, and is input to the DCT encoding circuit 32 for encoding / decoding processing. In addition, position information S31 such as a vertical synchronization signal and a horizontal synchronization signal in the image data S30 is also input to the coefficient selection circuit 31 for performing class classification based on the position information.

  The DCT coding circuit 32 performs block-unit DCT coding on the image data S30, and supplies the compression-coded image data S32 obtained as a result to the decoder 33. The decoder 33 decodes the compressed and encoded image data S32 in units of blocks, and supplies the decoded image data S33 to the coefficient selection circuit 31 and also to the dynamic range detection circuit 34 and the class generation circuit 35.

  The dynamic range detection circuit 34 detects the dynamic range DR of each block based on the image data S33. This dynamic range DR is supplied to the coefficient selection circuit 31 as an element of class classification. The class generation circuit 35 generates a class classification code S34 by performing ADRC encoding on the image data S33. This class classification code S34 is supplied to the coefficient selection circuit 31 as an element of a new class classification.

The coefficient selection circuit 31 creates a new class classification code by adding the elements of the dynamic range DR and the position information S31 to the supplied class classification code S34, and linearly correlates the correlation between the image data S30 and S33 for each class. The coefficient combination used at this time is obtained by learning using the least square method. The obtained coefficient set is supplied to a memory forming the coefficient table circuit 25 together with the class classification code. As a result, a coefficient table capable of optimal adaptive prediction processing is created.
By the way, the dynamic range of the block that expresses when the block distortion is likely to occur or the edge or detail part is included easily, and the position information that indicates when and where the target pixel falls in the block are considered. When the coefficient group is obtained above, a coefficient group that adaptively reduces block distortion is obtained.

In the above configuration, the image signal decoding apparatus 20 first decodes the transmission image data S1 that has been subjected to high-efficiency compression coding by DCT coding, by the block decoding circuit 4. Then, the dynamic range detection circuit 6 obtains the dynamic range of each block and supplies the dynamic range to the coefficient table circuit 25 via the DR memory 7. Also, the restored block-unit image data S5 is supplied to the class generation circuit 22, where a class classification code S22 by ADRC encoding is generated, and the class classification code S22 is supplied to the coefficient table circuit 25.
Further, the control circuit 23 obtains position information S21 indicating the relative position in the block of the target pixel based on the horizontal synchronizing signal S2 and the vertical synchronizing signal S3, and supplies this position information S21 to the coefficient table circuit 25. .

  The coefficient table circuit 25 reads out a coefficient set stored in advance for each class by using a combination of the supplied class classification code S22, dynamic range DR and position information S21 as address information. That is, the coefficient combination is read out by using the class classification code S22 plus the dynamic range DR and the position information S21 as the class classification code.

  The product-sum operation circuit 24 performs the product-sum operation processing on the image data supplied from the delay / prediction tap memory 21 and the coefficient set supplied from the coefficient table circuit 25 to correct the image data for each class. At this time, the coefficient set supplied from the coefficient table circuit 25 is a block dynamic range (DR) that represents characteristics such as block distortion or the presence of an edge or detail portion. Since the position information (S21) indicating the position and the position in the position is taken into consideration, the block distortion generated near the block boundary is adaptively removed without blurring the edges and details. can do.

  In this way, the image signal decoding apparatus 20 detects the dynamic range (DR) in the block and also detects the relative position information (S21) of the pixels in the block. Then, new class classification is performed by adding the detected dynamic range and position information to the class classification code (S22) obtained by pattern classification of the restored image data (S5), and image data is corrected for each class. Do. As a result, the block distortion can be removed adaptively without causing the edges and detail to be blurred.

  Incidentally, in the case of this embodiment, since only the processing in the image signal decoding apparatus 20 is required, there is an advantage that there is no burden on the transmission side apparatus and the transmission format need not be changed. In the case of this embodiment, by performing so-called adaptive prediction processing that modifies image data for each class, not only block distortion can be removed adaptively, but also the resolution can be improved.

  According to the above configuration, a new class classification is performed by adding the detected dynamic range and position information to a class classification code obtained by pattern classification of the restored image data, and a coefficient is read out for each class. By correcting the data, it is possible to adaptively remove block distortion without blurring the edges and detail portions.

(3) Other Embodiments (3-1) In the first embodiment described above, the case where the image data S1 is compression-coded by DCT coding has been described. However, the present invention is not limited to this. Even when the image data is encoded by other compression encoding such as ADRC encoding, DPCM (Differential Pulse Code Modulation) encoding, BTC (Block Truncation Coding) encoding, etc., the same effect as described above Can be obtained.
For example, if the input image data is compressed and encoded by ADRC encoding, the minimum value is added after inverse quantization with the quantization step width adaptively determined by the dynamic range. Then, a block encoding circuit that performs decoding processing may be provided. In this case, since the dynamic range is originally present in the encoded image data, it is only necessary to extract the dynamic range, and the maximum value detection circuit 6A, the minimum value detection circuit 6B, and the subtractor 6C. There is no need to provide the dynamic range detection circuit 6.

(3-2) In the first embodiment described above, the case where only one dynamic range corresponding to the pixel of interest is read from the DR memory 7 has been described. If the dynamic range of the peripheral block to be read is also read out from the DR memory 7 and scaled down to an appropriate amount of information, it is possible to remove fine block distortion considering the state of the peripheral block.

(3-3) Furthermore, in the first embodiment described above, the coefficient table circuit 10 and the product-sum operation circuit 9 have been described separately. However, the present invention is not limited to this, and the coefficient table circuit 10 and the product-sum operation circuit 9 9 may be combined into a block distortion removing circuit.

(3-4) In the second embodiment described above, the case where the image data S1 is compression-coded by DCT coding has been described. However, the present invention is not limited to this, and the image data is ADRC-coded. In the case of encoding by other compression encoding such as DPCM encoding or BTC encoding, the same effect as described above can be obtained.

  For example, when input image data is compression-encoded by ADRC encoding, the configuration may be as shown in FIG. That is, the ADRC-encoded image data S40 is supplied to the ADRC decoder 41, the class generation circuit 42, and the DR memory 7 via the frame decomposition circuit 2. At that time, the horizontal synchronizing signal S2 and the vertical synchronizing signal S3 extracted by the frame decomposing circuit 2 are supplied to the control circuit 23.

  The ADRC decoder 41 performs inverse quantization with a quantization step width adaptively determined by the dynamic range, and then adds the minimum value to perform decoding processing to block the ADRC-encoded image data S40 in units of blocks. The image data S41 obtained as a result is supplied to the delay / prediction tap memory 21.

  The class generation circuit 42 extracts the requantization code Q included in the ADRC-encoded image data S40, and uses the extracted requantization code Q as it is or collects a predetermined number of taps to classify the classification code S42. And the class classification code S42 is supplied to the coefficient table circuit 43.

  The DR memory 7 extracts the dynamic range DR included in the ADRC-encoded image data S40, stores the extracted dynamic range sequentially, and reads it out and supplies it to the coefficient table circuit 43. That is, in this case, since the dynamic range is originally included in the image data S40, it is only necessary to extract the dynamic range.

The control circuit 23 obtains position information indicating the relative position in the block of the pixel of interest based on the horizontal synchronizing signal S2 and the vertical synchronizing signal S3 supplied as in the second embodiment, and the obtained position information. S21 is supplied to the coefficient table circuit 43.
The delay / prediction tap memory 21 is a memory that combines delay, block decomposition, and prediction tap formation as in the second embodiment, and sequentially reads the stored image data and supplies it to the product-sum operation circuit 24.

  The coefficient table circuit 43 is a memory in which coefficient pairs for each class used for the product-sum operation of the product-sum operation circuit 24 are stored in advance by learning. The supplied dynamic range DR, position information S21, and class classification The stored coefficient group is read out using the combination of the codes S42 as address information and supplied to the product-sum operation circuit 24.

In this way, the product-sum operation circuit 24 performs a product-sum operation process on the image data supplied from the delay / prediction tap memory 21 and the coefficient set supplied from the coefficient table circuit 43 to perform an adaptive prediction process for each class. Image data S43 from which distortion has been removed is obtained.
When the image data S40 input in this way is compressed and encoded by ADRC encoding, the dynamic range detection circuit is not required, and the configuration can be further simplified.

Incidentally, in this case, the coefficient table circuit 43 is formed by using a coefficient creating circuit 50 as shown in FIG. In this coefficient creation circuit 50, first, for learning processing, already known image data S50 and image data obtained by coding and decoding the image data S50 by the ADRC coding circuit 52 and the ADRC decoder 53 are used. S51 is supplied to the coefficient selection circuit 51.
In addition, in the coefficient generation circuit 50, for class classification, position information S52 such as a vertical synchronization signal and a horizontal synchronization signal in the image data S50, a dynamic range DR included in the ADRC-encoded image data S53, and an image The class selection code S54 generated by the class generation circuit 54 based on the requantization code Q included in the data S53 is supplied to the coefficient selection circuit 51.

  The coefficient selection circuit 51 creates a new class classification code by adding the elements of the dynamic range DR and the position information S52 to the supplied class classification code S54, and sets the correlation between the image data S50 and S51 to the minimum two for each class. Obtained by learning using multiplication. Then, the coefficient group representing the obtained correlation is supplied to the memory forming the coefficient table circuit 43 together with the class classification code. As a result, a coefficient table capable of optimal adaptive prediction processing is created.

(3-5) Further, in the second embodiment described above, the case where ADRC encoding is used when class classification is performed by the class generation circuit 22 has been described. However, the present invention is not limited to this, and a class classification technique is used. Even if other compression coding such as DCT coding, DPCM coding, BTC coding or the like is used, the same effect as in the above case can be obtained.

(3-6) Further, in the above-described embodiment, the case where the present invention is applied to the image signal decoding apparatus that decodes the transmission image data S1 transmitted after being compressed and encoded has been described. The present invention is not limited to this, and the same effects as those described above can also be obtained when the present invention is applied to, for example, an image signal decoding apparatus that decodes and reproduces image data recorded by compression encoding. In short, any image signal decoding apparatus that decodes compression-encoded image data can be widely applied.

  The present invention can be used, for example, when decoding an image signal that has been subjected to high-efficiency compression coding using a block coding method.

It is a block diagram which shows the structure of the image signal decoding apparatus by one Example of this invention. It is a basic diagram which shows an example of the coefficient group output from a coefficient table circuit. It is a basic diagram which shows an example of the coefficient group output from a coefficient table circuit. It is a block diagram which shows the structure of the image signal decoding apparatus by 2nd Example. It is a basic diagram with which it uses for description about the relative position of the pixel in a block. It is a block diagram which shows the structure of a coefficient production circuit. It is a block diagram which shows the structure of the image signal decoding apparatus by another Example. It is a block diagram which shows the structure of the coefficient preparation circuit by another Example.

Explanation of symbols

1, 20, 40... Image signal decoding device, 3, 23... Control circuit, 4... Block decoding circuit, 6... Dynamic range detection circuit, 7. , 24... Product-sum operation circuit, 10, 25, 43... Coefficient table circuit, 21... Delay / prediction tap memory, 22, 42, 54... Class generation circuit, 30, 50. 51 …… Coefficient selection circuit, 41 …… ADRC decoder.

Claims (2)

In an image signal decoding apparatus for decoding compressed image data in which image data is compression-encoded in units of blocks by predetermined compression encoding means,
Dynamic range detection means for detecting the dynamic range of each block of the decoded image data obtained by decoding the compressed image data in units of blocks;
The by a value obtained by the peripheral pixel compression coding, the classification code representing a class to be classified when each pixel classified spatial pattern centering on the target pixel in each block of the decoded image data Class detection means for detecting a first class to which the decoded image data belongs;
Position information detecting means for detecting position information representing a relative position of the target pixel in the block of the decoded image data;
For each second class obtained by adding the dynamic range element and the position information element to the class classification code of the first class , the correlation between the image data and the decoded image data is represented by a linear primary combination. Coefficient sets that are subjected to a stronger low-pass filter as the block dynamic range is smaller are stored in advance, and the first class classification code obtained by the class detection unit is stored in the dynamic range detection unit. A coefficient table for reading out and outputting the coefficient set according to the second class to which the element of the dynamic range obtained and the element of the position information obtained by the position information detecting means is added;
An image signal decoding apparatus comprising: a product-sum operation unit that corrects the decoded image data by performing a product-sum operation on the coefficient set output from the coefficient table and the decoded image data. In an image signal decoding method for decoding compressed image data obtained by compressing and encoding image data in units of blocks by predetermined compression encoding means,
A dynamic range detection step for detecting a dynamic range of each block of the decoded image data obtained by decoding the compressed image data in units of blocks;
A value obtained by compressing and encoding peripheral pixels centering on the target pixel in each block of the decoded image data is used as a class classification code representing a class classified when each pixel is classified into a spatial pattern. A class detection step for detecting a first class to which the decoded image data belongs;
A position information detection step for detecting position information indicating a relative position of the target pixel in the block of the decoded image data;
For each second class obtained by adding the dynamic range element and the position information element to the class classification code of the first class, the correlation between the image data and the decoded image data is represented by a linear primary combination. From the coefficient table in which coefficient combinations that are subjected to a strong low-pass filter as the block dynamic range is smaller are stored in advance, the first class classification code obtained by the class detection step is added to the dynamic class. Coefficient reading that reads out and outputs the coefficient set according to the second class to which the element of the dynamic range obtained by the microphone range detection step and the element of the position information obtained by the position information detection step are added. Steps,
A product-sum operation step for correcting the decoded image data by performing a product-sum operation on the coefficient set output from the coefficient table and the decoded image data;
An image signal decoding method comprising:

Images