KR100598754B1 - Device for coding original image data, device for decoding coded image data, method for coding original image data, method for decoding coded image data, method for transmitting image data, computer-readable storing media - Google Patents

Abstract

The present invention provides an apparatus and method configured to be able to provide a decoded picture almost identical to an original picture to be encoded. In particular, correction data having fewer pixels than the number of pixels of the original image is calculated by performing a predetermined calculation using pixel values for the pixels of each block of the image and mapping coefficients corresponding to the class of the block. Then, a prediction value for the original image is calculated at the local decoder based on the correction data. Then, in the error calculating unit, a prediction error with respect to the predicted value of the original image is detected. In the mapping setting unit, the mapping coefficient corresponding to the class of the block is changed based on the prediction error. By repeating this process, mapping coefficients that are generated when the prediction error is smaller than a predetermined threshold value are obtained. Thus, the number of pixels of the encoded picture is minimized by including mapping coefficients instead of at least some picture information.

Description

An original image data encoding apparatus, an encoded image data decoding apparatus, an original image data encoding method, an encoded image data decoding method, an image data transmission method, a computer-readable recording medium TECHNICAL FIELD DEVICE FOR CODING ORIGINAL IMAGE DATA, DEVICE FOR DECODING CODED IMAGE DATA, METHOD FOR CODING ORIGINAL IMAGE DATA, METHOD FOR DECODING CODED IMAGE DATA, METHOD FOR TRANSMITTING IMAGE DATA, COMPUTER-READABLE STORING MEDIA}

The present invention relates to a picture coding apparatus, a picture coding method, a picture decoding apparatus, a recording medium and a picture processing apparatus, and a picture transmitting method, and more specifically, thinning-out a picture so that a decoded picture almost identical to an original picture is obtained. out) (Subsampling) and compression-encoding, and relates to a picture coding apparatus and a picture coding method, a picture decoding apparatus, a recording medium and an image processing apparatus, and an image transmission method.

Conventionally, various methods have been proposed as a method of compressing an image, and one of these methods is a method of compressing an image by thinning out the number of pixels of the image.

However, when the thinned-out compressed image is stretched only by interpolation, the resolution of the resulting decoded image is lowered.

There are at least two reasons for the deterioration of the resolution of the decoded image. One reason is that the high frequency components included in the original image are not included in the thinned-out image. The second reason is that the pixel values of the pixels constituting the image after thinning out are not particularly suitable for decoding the original image. Therefore, there is a need to provide a method and apparatus for solving these problems.

The present invention has been proposed to solve this problem, and encodes an image by thinning-out (subsampling) to obtain a decoded image that is at least or nearly identical to the original image data.

The picture coding method and apparatus according to the present invention generates class information corresponding to a block of pixels of the original picture data. By the class information, mapping coefficients corresponding to the class information are read out from the memory. The coded picture data is calculated in response to the original picture data and the read mapping coefficients.

An image decoding method and apparatus according to the present invention receives encoded image data and extracts a block of pixels. Class information is generated corresponding to the extracted pixel block. The prediction coefficients corresponding to the generated class information are read out from the memory. The decoded picture data is calculated in response to the coded picture data and the read prediction coefficients.

Other features of the present invention will become more apparent upon consideration with reference to the following description in conjunction with the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS In order to clarify the corresponding relationship between the means of the various embodiments disclosed herein, certain features of the present invention will first be briefly described with reference to the drawings. It is to be understood that the term "unit" is broadly understood to include a hard-wired circuit, a mainframe computer loaded with appropriate software, a programmed microprocessor or microcontroller, or a combination thereof. will be.

The picture coding apparatus of the present invention provides a block means (for example, a blocking unit 11 shown in Fig. 2) for blocking a picture into a prescribed block, in response to the characteristics of the blocks. Means for classifying the data into prescribed classes (e.g., class classification unit 13 shown in Fig. 2), and storage means for storing the prescribed mapping coefficients of each block (e.g., in Fig. 2). The illustrated mapping coefficient memory 14 and calculation means (for example, the calculation unit 16 shown in FIG. 2) for calculating the encoded data having a small number of pack cells for the encoded image. In particular, the calculation means performs a prescribed calculation on the pixel value of the pixels of one block and one or more mapping coefficients corresponding to the class of the block to which the block belongs.

The foregoing should not be limited to each of the above listed means.

Referring to FIG. 1, an image processing system is shown in FIG. In operation, the digitized image data is supplied to the transmission device 1. The transmission device 1 prescribes and calculates the input image data, and records the resultant data on the recording medium 2 including, for example, an optical disk, an optical-magnetic disk, a magnetic tape, or the like, or The resultant data is transmitted via a transmission path 3 such as a terrestrial wave, satellite circuit, telephone line or CATV network.

Subsequently, the encoded data recorded on the recording medium 2 is reproduced in the reception device 4 or the encoded data transmitted via the transmission path 3 is received. The coded data is decompressed and decoded, and the resulting decoded picture is supplied to a display not shown in the figure.

The above-described image processing system is, for example, a device for recording / reproducing images such as an optical disk apparatus, a magneto-optical disk apparatus, a magnetic tape apparatus, and the transmission of images such as, for example, a video phone apparatus, a television broadcasting system, a CATV system, and the like. It can be applied to a device for executing. In addition, the image processing system of FIG. 1 can be used in mobile applications, and can also be applied to portable terminals having a low transmission rate, such as, for example, a portable telephone.

Referring to FIG. 2, in FIG. 2, image data to be encoded is input to the blocking unit 11. The blocking unit 11 includes a block centered on a pixel of interest (eg, each block is 3 × 3 pixels). Can be blocks). The blocks are then supplied to an adaptive dynamic range coding (ADRC) processing unit 12 and a delay unit 15.

The ADRC processing unit 12 then performs ADRC processing on the blocks provided from the blocking unit 11 and outputs the resulting ADRC code to the class classification unit 13.

The number of bits needed to represent the pixels of the blocks is reduced with ADRC processing.

For simplicity, an example is described using a block having only four pixels. Referring to FIG. 3A, the maximum value MAX and the minimum value MIN for four pixels are detected in the ADRC process. DR = MAX-MIN is then taken as the localized dynamic range of the block, and the pixel values of the pixels comprising this block are quantized to K-bits based on the dynamic range DR.

That is, the minimum value MIN is subtracted from the value of each of the pixels in the block, and the result of this subtraction is divided by DR / 2K and then converted into a code corresponding to the resulting divided value. For example, when K = 2, as shown in FIG. 3B, the dynamic range DR is divided into 4 (2 ² ) equal parts, and a determination is made as to the corresponding portion of the range to which each pixel value belongs. When the divided value belongs to the lowest level range, the second lowest level range, the third lowest level range, or the highest level range, for example, each encoding is performed by two bits of 00B, 01B, 10B, or 11B. (B represents a binary number). Subsequently, in decoding, ADRC codes 00B, 01B, 10B, and 11B are assigned to the center value L00 of the lowest level range, the center value L01 of the second lowest level range, the center value L10 of the third lowest level range, and the highest level range. Is executed by converting to the center value L11. The minimum value MIN is added to these values.

The form of ADRC shown in FIG. 3B is called non-edge matching. 3C shows improved ADRC processing. As shown in Fig. 3C, the dynamic range DR is divided into quarters and the average value MIN 'of pixel values in the range of the lowest level and the average value MAX' of pixel values in the range of the highest level are converted into ADRC codes 00B and 11B. do. The level of dividing the dynamic range DR 'defined by MAX'-MIN' (in three equal parts) is converted into ADRC codes 01B and 10B so that ADRC decoding is performed.

The detailed processing of ADRC is disclosed in detail in Unexamined-Japanese-Patent No. 3-53778, for example.

The number of bits can be reduced by performing ADRC processing to re-quantize using fewer bits than the number of bits allocated to the pixels constituting the original block.

Referring again to FIG. 2, the class classification unit 13 performs class classification processing to classify blocks from the ADRC processing unit 12 in response to the characteristics of these blocks. Information about the class to which the block belongs is provided to the mapping coefficient memory 14 as class information.

The detailed class classification process will now be described with reference to the configuration examples. As shown in Fig. 4A, a noted pixel and three neighboring pixels constitute a 2x2 pixel block (class classification block), each of which has one bit (zero ( 0) or 1 level). At this time, these 2x2 pixel blocks are classified into 16 (= (2 ¹ ) ⁴ ) patterns as shown in Fig. 4B by the level distribution of the respective pixels. This pattern division is an example of class classification processing that can be performed by the class classification unit 13.

The class classification process may be done to further consider the activity (picture complexity) of the picture (picture in the block).

For example, when a class classification block is composed of 9 pixels of 3 x 3 and 8 bits are allocated to image data to be encoded so that class classification processing is performed for such a class classification block, a considerable number of classes are used. The classification pattern 2 ^{8 9} may result as a result.

Therefore, in the ADRC processing unit 12, the number of bits of pixels constituting the class classification block can be reduced and the number of classes can be further reduced with respect to the class classification block. For example, assuming that 1-bit ADRC processing is performed, ADRC processing is performed so that the number of classes is reduced from (2 ⁸ ) ⁹ to (2 ¹ ) ⁹ , that is, 512.

In the above-described embodiment, class classification processing is performed in the class classification unit 13 based on the ADRC code output from the ADRC processing unit 12. However, class classification processing may include, for example, DPCM (estimation coding), BTC (block truncation coding), VQ (vector quantization), DCT (Discrete Cosine Transforms), or Adamar Transformation. transforms) or other transforms may be performed using the data.

Referring again to FIG. 2, the mapping coefficient memory 14 stores mapping coefficients obtained by learning processing (mapping coefficient learning) for each item of class information. The learning process will be described later. The class information supplied from the class classification unit 13 is provided as an address, and the mapping coefficient stored in the provided address is supplied to the calculation unit 16.

The delay unit 15 delays the blocks supplied from the blocking circuit 11 until the mapping coefficients corresponding to these block class information are read out from the mapping coefficient memory 14. The delayed blocks are then supplied to the calculation unit 16.

The calculating unit 16 defines an operation using pixel values for pixels constituting the block supplied from the delay unit 15 and mapping coefficients corresponding to corresponding classes of the corresponding blocks supplied from the mapping coefficient memory 14. Is done. In this way, the number of pixels for the image is thinned out (reduced) so that the encoded data is calculated. That is, the pixel values for each of the pixels constituting the block output by the blocking unit 11 are y1, y2,... The mapping coefficients corresponding to the class of the block outputted by the mapping coefficient memory 14 are K1, K2,... In other words, the calculation unit 16 determines the prescribed count value regarded as the thinning-out number output as the pixel value of one or more representative pixels (eg, center pixels) of the block output by the blocking unit 11. f (y1, y2, ..., K1, K2, ...) is calculated.

When the number of pixels constituting the class classification block output by the blocking unit 11 is N, the calculation unit 16 reduces the image data by 1 / N and outputs the reduced image data as encoded data. .

Therefore, the encoded data output by the calculation unit 16 is not obtained by performing so-called simple thinning-out when the center pixel of a block including N pixels is simply extracted and output. Able to know. Rather, the encoded data are coefficient values f (y1, y2, ..., K1, K2, ...) defined by the N pixels constituting the block as described above. In other respects, it is thought that the coefficient values f (y1, y2, ..., K1, K2, ...) can be obtained by a simple thinning-out process, but the pixel value of the pixel at the center of the block is the peripheral pixel value. Correction (adjustment) on the basis of Encoded data, which is data obtained from the result of calculating the mapping coefficient and the pixels constituting the block, is referred to as corrected data hereinafter.

The arithmetic processing generated in the calculation unit 16 calculates the pixel value of each of the pixels constituting the class classification block outputted by the blocking unit 11 by the coefficient values f (y1, y2, ..., K1, K2, ...). ) Can be considered as a process of mapping. Coefficients K1, K2,... Used in this processing. Is called the mapping coefficient.

The transmission / recording device 17 records the corrected data supplied as the encoded data from the calculation unit 16 on the recording medium 2 and / or transmits the corrected data via the transmission path 3.

Next, the above operation will be described with reference to the flowchart of FIG. 5.

Image data is supplied to the blocking unit 11 in units of one frame, for example. Subsequently, in step S1, the blocking unit 11 displays image data as 3x3 (horizontal, vertical), for example, as shown by a rectangle surrounding pixels Y33 (1) to Y33 (9) in FIG. It is divided into nine pixel class classification blocks arranged. These blocks are supplied to the ADRC processing unit 12 and the delay unit 15 in that order.

Although the blocks for class classification are shown as square blocks of 3x3 pixels, it is to be understood that these blocks need not necessarily be square. For example, the blocks may be rectangular, intersecting or any other shape. In addition, the number of pixels constituting the block is not limited to nine pixels arranged as 3x3. In addition, the class classification block may include pixels that are far apart from each other in addition to including pixels that are directly neighboring. However, it is preferable that the shape and number of the pixels match the shape and number of the pixels of the block used for the map coefficient learning as described below.

When the ADRC processing unit 12 receives a class classification block from the blocking unit 11, the ADRC processing (for example, 1-bit ADRC processing) is performed on the block in step S2, whereby the original image data A block with a pixel represented by fewer bits than the pixel of is determined. The ADRC processed block is then supplied to the class classification unit 13.

In step S3, the class classification block provided from the ADRC processing unit 12 is class classified in the class classification unit 13. The resulting class information is supplied to the mapping coefficient memory 14 as an address. In step S3A, the mapping coefficients corresponding to the class information supplied by the class classification unit 13 are supplied to the calculation unit 16 after being read from the mapping coefficient memory 14.

On the other hand, in the delay unit 15, blocks provided from the blocking circuit 11 are delayed to wait for reading of mapping coefficients corresponding to the class information for blocks from the mapping coefficient memory 14. After an appropriate delay time, blocks are supplied to the calculation unit 16. In step S4, in the calculation unit 16, it is assumed that the above-described function value f (●) (in the parenthesis of this coefficient f represents a set of pixel values y1, y2, ... and mapping coefficients k1, k2, ...). Case) is calculated using the pixel value of each of the pixels constituting the class classification block provided from the delay unit 15 and the mapping coefficient provided from the mapping coefficient memory 14. Then, the corrected data (e.g., the corrected form of the pixel value of the pixel at the center of the pixels of the blocks) is calculated. The corrected data is then supplied to the transmission / recording device 17 as encoded data for the image.

In step S5, in the transmission / recording device 17, the encoded data provided from the calculation unit 16 is recorded on the recording medium 2 or transmitted via the transmission path 3.

Next, in step S6, a determination is made as to whether or not the processing for one frame of the image data is completed. In step S6, when it is determined that the processing for one frame of image data has not yet been completed, the processing returns to step S2, and the processing starting from step S2 is repeated for class separation for the next block. When it is determined in step S6 that the processing for one frame of the image data is completed, the processing returns to step S1 and the processing starting from step S1 is repeated for the next frame.

Next, FIG. 7 shows a structural example of an image processing apparatus for performing a learning (mapping coefficient learning) process for calculating a mapping coefficient. For example, this process can be used to determine the mapping coefficients to be stored in the mapping coefficient memory 14 of FIG.

Referring to FIG. 7, one frame or more frames of digital image data (hereinafter, simply referred to as learning images) are stored in the memory 21. Then, the blocking unit 22 reads the image data stored in the memory 21 and transfers the block (of the same structure as the block for class classification output from the blocking unit 11 of FIG. 2 unit) to the ADRC processing unit 23 and Supply to the calculation unit 26.

The ADRC processing unit 23 and the class classification unit 24 perform processing similar to the processing executed by the ADRC processing unit 12 and the class classification unit 13 in FIG. The class information for the block output by the blocking unit 22 is output from the class classification unit 24 and supplied to the mapping coefficient memory 31 as an address.

The calculation unit 26 performs the same calculation as the calculation unit 16 in FIG. 2 using the pixels constituting the block supplied from the blocking unit 22 and the mapping coefficient supplied from the mapping coefficient memory 31. The resulting corrected data (coefficient f ()) is supplied to the local decoder 27.

The local decoder 27 predicts the predicted value of the original image data for learning (the predicted value of the pixel value of the pixel constituting the block output by the blocking unit 22) based on the corrected data supplied from the calculation unit 26. (Calculation) and these predicted values are supplied to the error calculation unit 28. The error calculator 28 reads from the memory 21 a pixel value (absolute value) for the image for learning corresponding to the predicted value supplied from the local decoder 27, and stores the pixel value for these images for the learning. The prediction error of the predicted value of the predicted value corresponding to is calculated (detected) and these prediction errors are supplied to the determination unit 29 as error information.

The determination unit 29 compares the error information from the error calculator 28 and the prescribed threshold value ε 1 and controls the mapping coefficient setting unit 30 to correspond to the comparison result thereof. The mapping coefficient setting unit 30 sets (updates) the setting of the number of mapping coefficients equal to the number of classes obtained from the result of class classification generated in the class classification unit 24 under the control of the determination unit 29. The mapping coefficients are supplied to the mapping coefficient memory 31.

The mapping coefficient memory 31 temporarily stores the mapping coefficient supplied from the mapping coefficient setting unit 30. The mapping coefficient memory 31 has a storage area in which the class classification unit 24 can store as many mapping coefficients (set of mapping coefficients) as the number of classes for classifying blocks. Subsequently, when each new mapping coefficient is supplied from the mapping coefficient setting unit 30 to each of the storage areas, it is stored instead of the mapping coefficient already stored.

The mapping memory 31 supplies the mapping coefficients stored at the addresses corresponding to the class information supplied from the class classification unit 24 to the AD input of the mapping coefficient memory 31, and the mapping coefficient memory 31 supplies these coefficients. It supplies to the calculation unit 26.

The learning operation of the unit of FIG. 7 will be described below with reference to FIG. 8.

First, in step S51, the mapping coefficient setting unit 30 sets mapping coefficient memory initial value values for mapping coefficients for mapping coefficients as many as the number of classes for which the class classification unit 24 classifies a block. Specifically, the mapping coefficient (initial value) from the mapping coefficient setting unit 30 is stored in the address of the area in the mapping coefficient memory 31 which holds the mapping coefficient for the class to which the mapping coefficient corresponds.

In step S52, the blocking unit 22 is a 3x3 pixel in which all the images for learning stored in the memory 21 are centered in the same manner as the blocking unit 11 in FIG. 2 operates. Write in the form of a block. The blocking unit 21 then reads these blocks from the memory 21 and sequentially supplies these blocks to the ADRC processing unit 23 and the calculation unit 26.

In step S53, 1-bit ADRC processing is performed on the blocks from the blocking unit 22 in a manner similar to the manner in which the ADRC processor 12 of FIG. 2 operates in the ADRC processing unit 23. The result of the ADRC processing is supplied to the class classification unit 24. In step S54, the class of the blocks supplied from the ADRC processing unit 23 is determined in the class classification unit 24, and the determined class information is supplied to the mapping coefficient memory 31 as addresses. Next, in step S55, the mapping coefficient is read from the addresses of the mapping memory 31 corresponding to the class information supplied from the class classification unit 24. The mapping coefficient is then supplied to the calculation unit 26.

When the blocks are received from the blocking unit 22 and a mapping coefficient corresponding to the class of the received blocks is received from the mapping coefficient memory 31, the calculation unit 26 calculates the function value f (β) ( Step S56). In particular, the calculation unit 26 calculates the coefficient value using the mapping coefficient and the pixel values of the pixels constituting the blocks supplied from the blocking unit 22. The result of this calculation is then supplied to the local decoder 27 as corrected data corrected from the pixel value of the pixel at the center of the block supplied from the blocking unit 22.

For example, when the shape of a 3x3 pixel block shown enclosed in a quadrangle as shown in Fig. 6 is regarded as being output from the block unit 22, the pixel value of the pixel shown by? In Fig. 6 Corrected data corrected with respect to the result is obtained in the calculation unit 26 and output to the local decoder 27.

The number of pixels constituting the training picture is thinned out in the calculation unit 26 in 1/9 so that this thinned-out picture is supplied to the local decoder 27.

In Figure 6, when the expression above illustrated by mark ● in the i-th and the correction of the correction data corresponding to the j-th pixel from the left of data X _ij is considered to be the corrected data X _ij is in the center, learning circle The pixel values for the nine pixels arranged as 3x3 generated from the image data (original image data) are Y _ij (1), Y _ij (2), Y _ij (from the leftmost right direction and from top to bottom). 3), Y _ij (4), Y _ij (5), Y _ij (6), Y _ij (7), Y _ij (8), and Y _ij (9).

8, processing continues in step S57 after the data corrected in step S56 is calculated. In step S57, a determination is made as to whether corrected data has been obtained for all the images for learning stored in the memory 21. If it is determined in step S57 that the corrected data has not yet been obtained for all the images for learning, the process returns to step S53 until the corrected data for all the learning images is obtained in steps S53 to S57. The processing of is repeated.

When it is determined in step S57 that corrected data has been obtained for all the images for learning, that is, an image thinned out by 1/9 for all the images for learning stored in the memory 21 is obtained, step S58. Processing continues at. Note that these thinned-out pictures are not just learning images thinned out by 1/9, but rather these thinned-out pictures are based on pixel values obtained using mapping coefficients and calculations. Should be). Subsequently, the predicted values of the original image data for training are calculated by locally decoding the picture thinned out at the local decoder 27, and the predicted value is supplied to the error calculator 28.

The image containing the predicted value obtained by the local decoder 27 (as described later, when the error information output from the error information calculation unit 28 is smaller than the threshold value? 1), the side of the receiving device 4 (FIG. 1) Similar to the decoded picture obtained from.

In the error calculator 28, the learning image is read from the memory 21 in step S59, and a prediction error with respect to the predicted value supplied from the local decoder 27 is calculated in association with these images for learning. That is, when the pixel value for the training image is expressed by Y _ij and the predicted value output from the local decoder 27 is expressed by Σ (Y _ij ), the error calculation unit 28 expresses an error variance (error) expressed by the following equation. Squared sum) Q is calculated and supplied to the determination unit 29 as error information.

Q = ∑ (Y _ij -∑ [Y _ij ]) ²

In the above formula,? Represents the sum of all pixels of the learning image.

When the error information is received from the error calculating unit 28, the determination unit 29 compares the received error information with the prescribed threshold value epsilon 1 to determine the magnitude relationship (step S60). If the determination result error information in step S60 is larger than the threshold value epsilon 1, that is, if it is recognized that the image including the predicted value obtained by the local decoder 27 is not substantially the same as the original image data for learning, the determination unit 29 This control signal is output to the mapping coefficient setting unit 30. Next, the mapping coefficient setting unit 30 updates the mapping coefficient in accordance with the control signal from the determination unit in step S61 and newly stores the updated mapping coefficient in the mapping coefficient memory 31.

Subsequently, the process returns to step S53 and the process from step S53 is repeated using the updated mapping coefficient stored in the mapping coefficient memory 31.

The mapping coefficients generated by the mapping coefficient setting unit 30 may be changed randomly. Also, when the current error information is smaller than the previous error information, a change may be made to have the same tendency as before, and when the current error information is larger than the previous error information, the change may be reversed.

In addition, the change of the mapping coefficient can be made only for all classes or only some classes. For example, when changing the mapping coefficient for only some classes, it is desirable that a class that strongly affects the error information is detected, and that only the mapping coefficient for this type of class is changed. A class that strongly influences the error information can be found, for example, as follows. First, error information is obtained by performing error processing using the initial value for the mapping coefficient. The mapping coefficient is then changed by the same amount for each class and the resulting error information is compared with the error information obtained when the initial value is used. A class whose difference is greater than or equal to the prescribed value may be detected as a class having a strong influence with respect to the error information.

As described above, K1, K2,... When a plurality of mapping coefficients are set to 1, only mapping coefficients having a strong influence with respect to this information are updated.

In the above case, mapping coefficients are set for each class, but these mapping coefficients may be set independently for each block or for each unit of neighboring block units, for example.

For example, when mapping coefficients are set independently for each block, a plurality of mapping coefficient sets may be obtained for any one class (although there may be classes in which no set of mapping coefficients is obtained). . Since it is necessary to finally determine the mapping coefficients for each class, when a set of plural mapping coefficients is obtained for a particular class, one set of mapping coefficients is performed by performing some form of processing that targets the plural mapping coefficient sets. It is necessary to decide.

On the other hand, if the error information is smaller than the threshold value epsilon 1 as a result of the determination in step S60, that is, when the image including the predicted value obtained by the local decoder 27 is confirmed to be the same as the original image data for learning, the process is completed.

At this time, the mapping coefficient stored in the mapping coefficient memory 31 for each class is considered to be optimal for obtaining corrected data capable of restoring the decoded image, and is stored in the mapping coefficient memory 14 of FIG.

An image almost identical to the original image data can be obtained on the receiving device 4 (FIG. 1) side by generating corrected data using the learning mapping coefficient stored in the mapping coefficient memory 14.

In the embodiment of Fig. 7, as described above, an image is created in the form of blocks of nine pixels arranged as 3x3 with the pixel of interest in the blocking unit 22 as the center. In addition, one-bit ADRC processing is performed in the ADRC processor 23. Since the number of classes obtained by class classification using the class classification unit is 512 (= (2 ¹ )) ⁹ , 512 sets of mapping coefficients can be obtained.

Next, a configuration example of the local decoder 27 of FIG. 7 is shown in FIG. 9.

The corrected data from the calculation unit 26 is supplied to the blocking unit 41 for class classification and the blocking unit 42 for calculating the predicted value. The classifying blocking unit 41 creates the corrected data in the form of blocks centered on the corrected data noted in block units.

That is, as described above with reference to Fig. 6, when the i-th and j-th corrected data (compressed data) (or pixels) (parts indicated by marks? In Fig. 6) are expressed as X _ij , The classifying blocking unit 41 includes eight pixels X _{(i-1) (j-1)} , X adjacent to the upper left, upper, upper right, left, right, lower left, lower and lower right of the pixel X _ij noted. _{(i-1) j} , X _{(i-1) (j + 1)} , X _{i (j-1)} , X _{i (j + 1)} , X _{(i-1) (j-1)} , X _(i Create a class classification block that includes a total of nine pixels including _{-1) j} , X _{(i + 1) (j + 1)} . This class classification block is then supplied to the ADRC processing unit 43.

Since the class classification blocks obtained in the blocking unit 42 of FIG. 9 have a configuration for determining the class of the blocks obtained from the predicted value, the blocking unit of FIG. 2 in order to determine the class of the blocks for calculating the corrected data at this point. There is a difference from the blocks generated in (11).

The predictive value calculating blocking unit 42 creates the corrected data in the form of blocks for predictive value calculation centered on the corrected data of interest, which is a unit for calculating the predicted value for the original image data (the learning image). That is, in this embodiment, for example, in order to calculate the predicted values for the pixels Y _ij (1) to Y _ij (9), the image data generated from the original image data (original image) centered on the corrected data X _ij is used. Y _ij (1), Y _ij (2), Y _ij (3), Y _ij (4), Y _ij (5), Y _ij (6), Y _ij (7) , Y _ij (8) and Y _ij (9), the blocking unit 42 for calculating the predicted value is X _{(i-2) (j-2)} , X centered on the pixel X _ij in a 5 × 5 matrix. _{(i-2) (j-1)} , X _{(i-2) j} , X _{(i-2) (j + 1)} , X _{(i-2) (j + 2)} , X _{(i-1) ( j-2)} , X _{(i-1) (j-1)} , X _{(i-1) j} , X _{(i-1) (j + 1)} , X _{(i-1) (j + 2)} , X _{i (j-2)} , X _{i (j-1)} , X _ij , X _{i (j + 1)} , X _{i (j + 2)} , X _{(i + 1) (j-2)} , X _{(i + 1) (j-1)} , X _{(i + 1) j} , X _{(i + 1) (j + 1)} , X _{(i + 1) (j + 2)} , X _{(i + 2) (j-2 )} , X _{(i + 2) (j-1)} , X _{(i + 2) j} , X _{(i + 2) (j + 1)} and X _{(i + 2) (j + 2)} Create a rectangular block for calculating the predicted value to include.

Specifically, for example, the prediction error calculation block is calculated by using X ₁₁ , in order to calculate the predicted values for pixels Y _{33 (1)} to Y _{33 (9)} generated in the original image enclosed in a square shape in FIG. 6. X ₁₂ , X ₁₃ , X ₁₄ , X ₁₅ , X ₂₁ , X ₂₂ , X ₂₃ , X ₂₄ , X ₂₅ , X ₃₁ , X ₃₂ , X ₃₃ , X ₃₄ , X ₃₅ , X ₄₁ , X ₄₂ , X ₄₃ It consists of 25 pixels (corrected data) of X ₄₄ , X ₄₅ , X ₅₁ , X ₅₂ , X ₅₃ , X ₅₄ and X ₅₅ .

The predictive value calculating blocks obtained by the predictive value calculating blocking unit 42 are supplied to the predictive unit 46.

As in the case of the class classification block, the number and shape of pixels of the predictive value calculation block are not limited to the above. However, in the local decoder 27, the number of pixels constituting the prediction value calculation block is preferably larger than the number of pixels constituting the class classification block.

If the blocking is done (this is all the same except for the blocking process) and there are no corresponding pixels in the vicinity of the picture frame of the picture, for example, the same pixels as the pixels containing the picture frame are on the outside. The processing is performed by doing this.

The ADRC processing unit 43 performs one-bit ADRC processing on the block (using the class separating block) output by the class separating blocking unit 41 and supplies the results to the class separating unit 44. The class separating unit 44 sorts the blocks from the ADRC processing unit 43 and supplies the class information obtained from the sort to the address input of the prediction coefficient ROM 45. The prediction coefficient ROM 45 holds the prediction coefficients for each class, and reads the prediction coefficients held at an address corresponding to the class information when the class information is received from the class separation unit 44. The prediction coefficients are supplied to the prediction unit 46. The prediction coefficients for each class stored in the prediction coefficient ROM 45 are obtained from the learning (prediction coefficient learning) described later.

The prediction unit 46 calculates (predicts) the predicted value for the original image data by using the predictive value calculating blocks from the predictive value calculating block unit 42 and the predictive coefficient from the predictive coefficient ROM 45.

Next, with reference to the flowchart of FIG. 10, the operation | movement of the apparatus of FIG. 9 is demonstrated.

First, in step S21, the local decoder 27 creates corrected data from the calculation unit 26 in the form of a block. That is, in the class separation blocking unit 41, the corrected data is created in the form of 3x3 pixels for class classification centering on the corrected data of interest and supplied to the ADRC processing unit 43. In the predictive value calculation blocking unit 42, the corrected data is created in the form of a 5x5 pixel block for predictive value calculation and supplied to the prediction unit 46. A corresponding block for class classification and a block for predictive value calculation are generated in the class separating blocking unit 41 and the predictive value calculating blocking unit 42. That is, for example, to about the corrected data X of Figure 6 the 3 × 3 pixel block for class separation ₃₃ from the blocking unit 41 for the class separation, centered on the same correction data X ₃₃ A 5x5 pixel block for predictive value calculation is generated as a block for predictive value calculation.

In step S22, when the blocks for class classification are received, for example, the ADRC processing unit 43 performs 1 bit ADRC (ADRC performed using 1 bit quantization) on the blocks for class classification, The data corrected by this is converted (encoded) into one bit and output to the class classification unit 44. In step S23, the class classification unit 44 performs class classification processing on the blocks for class classification subjected to ADRC processing. That is, the class classification unit 44 detects each distribution state of the pixel level in the block and determines the class to which these class classification blocks belong based on this. The result of this class determination is supplied to the prediction coefficient ROM 45 as class information.

In the embodiment of Fig. 10, each of the class classification blocks is performed for a class classification block including nine pixels arranged as 3x3 with 1-bit ADRC processing, so that 512 (= (2 ¹ ) ) One of ^nine .

The process then proceeds to step S24 in which the 25x9 prediction coefficient is read from the address of the prediction coefficient ROM 45 corresponding to the class information supplied from the class classification unit 44. Next, in step S25, the prediction unit 46 uses 25 pixel values including these 25x9 prediction coefficients and the block for calculating the prediction value from the blocking unit 42 for calculating the prediction value, for example. The expected value for the pixel value of the original image data is calculated according to the following form of the linear linear equation.

= W ₁ X ₁ + W ₂ X ₂ + ...

Here, W ₁ , W ₂ ... are prediction coefficients, and X 1, X ₂ ... represent pixel values (corrected data) of pixels including blocks for predictive value calculation.

The pixel value of nine pixels is calculated from 25 pixels including the block for calculating the predicted value in the embodiment of FIG. 9 as described above.

Specifically, for example, class classification including 3x3 corrected data X ₂₂ to X ₂₄ , X ₃₂ to X ₃₄ , X ₄₂ to X ₄₄ centered on the corrected data X ₃₃ shown in FIG. 6. Block class information C for the dragon block is output from the class classification unit 44, and the corrected data X ₁₁ to X ₁₅ , X ₂₁ to X ₂₅ , and X ₃₁ for 5 x 5 pixels centered on the corrected data X ₃₃ . to X _35, X ₄₁ to X ₄₅ and X ₅₁ as to block for the predicted value calculation block for predicted value calculation comprising the X ₅₅ it is outputted from the blocking unit 42 for predicted value calculation.

Here, W ₁ (K) to W ₂₅ (K) are stored in the prediction coefficient ROM 45 at an address corresponding to the class information C, whereby they are set as prediction coefficients for the class.

The predicted value for the pixel value of the 3x3 pixel (the portion enclosed by the rectangle in FIG. 6) generated from the original image data centered on the corrected data X ₃₃ is calculated according to the following equation.

= w ₁ (k) X ₁₁ + w ₂ (k) X ₁₂ + w ₃ (k) X ₁₃

+ w ₄ (k) X ₁₄ + w ₅ (k) X ₁₅ + w ₆ (k) X ₂₁

+ w ₇ (k) X ₂₂ + w ₈ (k) X ₂₃ + w ₃ (k) X ₂₄

+ w ₁₀ (k) X ₂₅ + w ₁₁ (k) X ₃₁

+ w ₁₂ (k) X ₃₂ + w ₁₃ (k) X ₃₃

+ w ₁₄ (k) X ₃₄ + w ₁₅ (k) X ₃₅

+ w ₁₆ (k) X ₄₁ + w ₁₇ (k) X ₄₂

+ w ₁₈ (k) X ₄₃ + w ₁₉ (k) X ₄₄

+ w ₂₀ (k) X ₄₅ + w ₂₁ (k) X ₅₁

+ w ₂₂ (k) X ₅₂ + w ₂₃ (k) X ₅₃

+ w ₂₄ (k) X ₅₄ + w ₂₅ (k) X ₅₅

When nine predicted values are obtained in step S24, the process returns to step S21 and the processes of steps S21 to S24 are repeated to obtain predicted values in units of nine pixels.

Next, FIG. 11 shows an example of the configuration of an image processing apparatus that performs learning (prediction coefficient learning) to obtain prediction coefficients stored in the prediction coefficient ROM 45 of FIG. 9.

Image data for learning (learning picture) for obtaining prediction coefficients suitable for all pictures is supplied to the learning blocking unit 51 and the teacher blocking unit 52.

The learning blocking unit 51 calculates, from the input image data, 25 pixels (5 x 5 pixels) having a positional relationship shown by a mark? In Fig. 6 centering on the pixel of interest. Then, a block including these 25 pixels is supplied to the ADRC processor 53 and the learning data memory 56 as learning blocks.

In the teacher block unit 52, for example, nine pixels arranged in 3x3 centered on the pixel of interest are generated from the input image data. A block containing these nine pixels is then supplied to the teacher data memory 58 as a teacher block.

For example, in the case where the learning block including the 25 pixels in the positional relationship shown by the mark? In Fig. 6 is generated in the learning block unit 51, the learning 3 × 3 shown in Fig. 6 is surrounded by a rectangle. Pixel blocks are generated in the teacher's blocking unit 52.

The ADRC processor 53 extracts the central nine pixels (3 × 3 pixels) from the 25 pixels including the learning block, and applies them to the nine pixel blocks in the same manner as in the ADRC processing unit 43 of FIG. 1-bit ADRC processing is performed. Subsequently, an ADRC processed 3x3 pixel block is supplied to the class classification unit 54. The blocks from the ADRC processor 53 are classified in the class classification unit 54 in the same manner as the class classification unit 9, so that the obtained class information is passed through the terminal " a " And teacher data memory 58.

The learning blocks from the learning blocking unit 51 and the teacher blocks from the teacher blocking unit 52 are stored at addresses corresponding to the class information provided in the learning data memory 56 and the teacher data memory 58, respectively. .

In the learning data memory 56, for example, a block containing 5x5 pixels indicated by a mark? In Fig. 6 is regarded as a learning block and stored at a specific address. A × 3 pixel block is stored as a teacher block at a corresponding address in the teacher data memory 58.

The same processing is repeated for all the images for learning described above. In this way, in the local decoder 27 of FIG. 9 using a block for learning and a predictive value calculating block comprising 25 items of corrected data having a positional relationship such as 25 pixels comprising such block for learning. The teacher block including nine pixels from which the predicted value is obtained is stored in the address of the learning data memory 56 and the corresponding address in the teacher data memory 58.

A plurality of pieces of information can be stored at the same address of the learning data memory 56 and the teacher data memory 58. In this way, it is possible to store a plurality of sets of learning blocks and teacher blocks at the same address.

Subsequently, the selection terminal "a" of the switch 55 is switched to the terminal "b". The output of the counter 57 is supplied to the learning data memory 56 and the teacher data memory 58 as addresses. The counter 57 counts according to the prescribed clock and outputs the count value. The learning block and the teacher block stored at the address corresponding to the count value are read from the learning data memory 56 and the teacher data memory 58, respectively, and supplied to the calculation unit 59.

The set of blocks for learning and the set of blocks for teacher corresponding to the count value of the counter 57 are supplied to the calculation unit 59.

The calculation unit 59 receives a set of blocks for learning and a set of blocks for teaching for a specific class and calculates a prediction coefficient (e.g., using a least squares method) that minimizes the error.

For example, when the pixel values of the pixels including the learning block are X1, X2, X3, ... and the predictive coefficients to be obtained are W1, W2, W3, ..., the learning block is linear first order. It is composed by combining, and in order to obtain the pixel value y for a specific pixel, the prediction coefficients W1, W2, W3, ... are given by the equation y = W ₁ X ₁ + W ₂ X ₂ + W ₃ X ₃ + .. It is necessary to satisfy.

Computation unit 59, the prediction error from the learning block of the same class and the corresponding block for the teacher W for the true value y W ₁ X ₁ + W ₂ X ₂ + W ₃ X ₃ + ... The prediction coefficients W ₁ , W ₂ , W ₃ , ... are minimized. Therefore, a 25x9 prediction coefficient for each class is calculated by performing the above processing for each class.

The prediction coefficients obtained in the calculation unit 59 are supplied to the memory 60 for each class. In addition to the prediction coefficients from the calculation unit 59, coefficient values are supplied to the memory 60, whereby the measurement coefficients from the calculation unit 59 can be stored at addresses corresponding to the coefficient values from the counter 57. .

The most appropriate 25x9 prediction coefficients (prediction coefficients that minimize the error) for predicting a glass 3x3 pixel block are stored in the memory 60 at an address corresponding to the class.

Next, the learned prediction coefficients for each glass stored in the memory 60 are stored in the prediction coefficient ROM 45 of FIG. 9.

Rather than storing the actual prediction coefficients in the addresses corresponding to each class in the prediction coefficient ROM 45, it is possible to store the average value of pixel values including the teacher block. If this is regarded as class information, pixel values corresponding to this class are output. Thus, the local decoder 27 is completed even without providing the blocking unit 42 and the prediction unit 46 for calculating the predicted value.

In the embodiment of Fig. 9, in the local decoder 27, prediction coefficients already obtained by the above-described teacher (prediction coefficient teacher) are provided and stored in the prediction coefficient ROM 45, and specific prediction values are obtained by these prediction coefficients. However, as described below in connection with another embodiment, the predictive value may be obtained at the local decoder 27 by the learning image data from the memory and the corrected data from the arithmetic processing unit.

FIG. 12 is another configuration example of an image processing apparatus that performs a learning (mapping coefficient learning) process of calculating the mapping coefficient stored in the mapping coefficient memory 14 of FIG. 2.

According to the image processing apparatus of FIG. 7, for example, unlike the linear linear equation, it can be obtained by representing the function f as a nonlinear quadratic equation. However, in the image signal processing apparatus of Fig. 12, the coefficient f is represented only by a linear linear equation and an optimal prediction coefficient has been obtained.

The pixel value for each pixel including the 3x3 pixel block output by the blocking unit 11 of FIG. 2 centering on the pixel of interest is y ₁ , y ₂ , ..., y ₉ , and mapping When the mapping coefficients output by the coefficient memory 14 are k ₁ , k ₂ ,..., K ₉ , the image processing apparatus of FIG. 12 calculates the coefficient value f in order for the calculation unit 16 to obtain corrected data. Used to compute (y ₁ , y ₂ , ..., k ₁ , k ₂ , ...).

f (●) = k ₁ y ₁ + k ₂ y ₂ + ... + k ₉ y ₉

The learning image suitable for learning is supplied to the optimal corrected data calculation part 70 in units of one frame, for example. The optimum correction data calculator 70 includes a compression unit 71, a correction unit 72, a local decoder 73, an error calculator 74, and a determination unit 75. Subsequently, the image is compressed into a few pixels from the input image for learning and the pixel value including the optimal image for predicting the original image data is calculated and supplied to the buffer unit 76.

The training image supplied to the optimal corrected data calculation unit 70 is supplied to the compression unit 71 and the error calculator 74. The compression unit 71 simply thins out the learning image at the same rate as the calculation unit 16 of FIG. 2 thins out the pixels. That is, in this embodiment, the learning image is simply thinned out to 1/9 (when nine pixels arranged as 3x3 are one block, the corresponding pixel at the center of this block is calculated). This is compressed and supplied to the correction unit 72.

The correction unit 72 corrects the simple thinned out compressed data (hereinafter referred to as compressed data) provided from the compression unit 71 under the control of the determination unit 75. The resulting corrected data (this data is the corrected form of pixel values for the center pixel of the 3x3 pixel block in the same manner as the output of the calculation unit 16 of FIG. 2, hereinafter referred to as corrected data) is It is supplied from the correction unit 72 to the local decoder 73.

The local decoder 73 creates prediction values for the original image data (the image for learning) based on the corrected data from the correction unit 72 in a similar manner as the local decoder 27 of FIG. 7, and converts these predicted values. Supply to the error calculation unit 74.

The error calculator 74 calculates a prediction error with respect to the predicted value from the local decoder 73 with respect to the original image data input in the same manner as the error calculator 28 in FIG. This prediction error is supplied to the determination unit 75 as error information.

The determination unit 75 determines whether the corrected data output by the correction unit 72 is appropriate as a compression result for the original image data based on the error information from the error calculator 74. If it is determined by the judging unit 75 that the corrected data output by the correction unit 72 is not appropriate as a compression result for the original image data, the correction unit 72 is controlled to obtain a result for which the compressed data is obtained. It is corrected and output as new correction data. Further, if it is determined by the determination of the determination unit 75 that the corrected data output by the correction unit 72 is appropriate as a compression result for the original image data, then the corrected data supplied by the correction unit 72 is It is supplied to the buffer unit 76 as optimum correction data.

The buffer unit 76 includes a memory 76A that stores optimum correction data supplied from the correction unit. In addition, the buffer unit 76 reads the data of the optimal correction data stored in the memory 76A corresponding to the center pixel of the block read out from the memory 77A of the blocking unit 77, and stores this data in the memory 80. Supplies). When one frame portion of the corrected data is stored in the memory 76A, the buffer unit 76 outputs a control signal indicating this fact to the blocking unit 77.

The learning images are supplied to the blocking units 77 in units of one frame, respectively, in a manner similar to that for the optimal corrected data calculation unit 70. The blocking unit 77 includes a memory 77A for storing images for learning supplied thereto. Further, when a control signal is received from the buffer unit 76 by the blocking unit 77, the memory 77A is a block containing 3x3 pixels in the same manner as in the case of the blocking unit 11 of FIG. Remembered). These blocks of the learning image are subsequently read out and supplied to the ADRC processing unit 78 and the memory 80.

When a block is read from the memory 77A, a control signal indicative of the position of the read block is supplied to the buffer unit 76. Then, the 3x3 pixel block read from the memory 77A is confirmed in the buffer unit 76 based on this control signal, and the optimal correction data corresponding to the center pixel of this block is read out from the memory 76A. . That is, a specific 3x3 pixel block and the optimal corrected data corresponding to the block are supplied to the memory 80 at the same time. The ADRC processing unit 78 and the class classification unit 79 have the same configuration as the ADRC processing unit 12 and the class classification unit 13 in FIG. Subsequently, class information about a block from the blocking unit 77 output by the glass sorting unit 79 is supplied to the memory 80 as an address.

The memory 80 stores the optimal correction data supplied from the buffer unit 76 and the block supplied from the blocking unit 77 at an address corresponding to the class information supplied from the class classification unit 79. The memory 80 can store a plurality of information in a single address. Thus, optimally corrected data and blocks corresponding to specific class information can be stored as a plurality of sets.

The calculation unit 81 is adapted to optimally correct the corresponding blocks stored in the nine pixels y ₁ , y ₂ ,... Y ₉ and the memory 80 containing 3 × 3 blocks for the image for learning. The data y 'is read out, and the calculation unit 81 applies the minimum autonomous method such that the mapping coefficients k ₁ to k ₉ are obtained for each class and supplied to the memory 82. The memory 82 then stores the mapping coefficients k ₁ to k ₉ for each class supplied from the calculation unit 81 at addresses corresponding to these classes.

Next, the operation of the learning image processing apparatus of FIG. 12 will be described with reference to FIG. 13.

When the learning image is input, the learning image is stored in the memory 77A of the block unit 77 and is supplied to the optimum correction data calculation unit 70. The training image is received, and the optimal corrected data calculator 70 calculates the optimal corrected data for the training image (step S31).

Step S31 is shown in detail in the flowchart of FIG. Referring to Fig. 14, the compression unit 71 thins out the learning image to 1/9 in step S41 to generate compressed data, and at least initially does not correct the data, but through the correction unit 72, the local decoder. Output to (73). Then, the predicted value for the original image data is calculated in the local decoder 73 in step S42 based on the corrected data from the correction unit 72 (ie, local decoding is performed). (First, as described above, image data is simply thinned out to provide compressed data). This predicted value is supplied to the error calculator 74.

When the predicted value for the original image data is received from the local decoder 73, the error calculating unit 74 calculates a prediction error with respect to the predicted value from the local decoder 73 with respect to the original image data in step S43, and the error information. This is supplied to the determination unit 75 as a. In step S44, when the error information is received from the error calculator 74, the corrected data output from the correction unit 72 by the determination unit 75 is the compression result for the original image data based on this error information. A determination is made as to whether it is appropriate.

That is, a determination is made in step S44 as to whether the error information is smaller than the prescribed threshold value. When the determination result at step S44 indicates that the error information is smaller than the prescribed threshold value, it is confirmed that the corrected data output by the correction unit 72 is not suitable for use as compressed data for the original image data. Next, in step S55, the determination unit 75 controls the correction unit 72 to correct the compressed data output by the compression unit 71. The correction unit 72 changes the correction amount (correction value?) Under the control of the determination unit 75, corrects the compressed data and as a result, outputs the corrected data to the local decoder 73. Thereafter, the process returns to step S42 and is repeated.

As for the change of the mapping coefficient, for example, it is possible to correct the compressed data in the same manner as the method described above with reference to FIG. On the other hand, when it is determined in step S44 that the error information is smaller than the prescribed threshold value, it is confirmed that the corrected data output by the correction unit 72 is appropriate as a compression result for the original image data. The determination unit 75 then outputs the corrected data as the optimal correction data to the buffer unit in case the error information smaller than the prescribed threshold value is moved away from the correction unit 72. Subsequently, this data is stored in the memory 76A of the buffer unit 76 and the process of FIG. 14 is returned.

In the case where the error information is smaller than the prescribed threshold value, compressed data corrected as correction data is stored in the memory 76A as optimum corrected data. This optimal corrected data has error information smaller than the prescribed threshold value, and an image almost identical to the original image data (source image data) can be obtained by the corrected data to calculate a prediction value.

Returning to FIG. 13, when one frame portion of the optimal correction data is stored in the memory 76A, the buffer unit 76 outputs a control signal to the blocking unit 77. When the control signal is received from the buffer unit 76, the blocking unit 77 divides the learning image stored in the memory 77A into 3x3 pixel blocks in step S32. Next, the blocking unit 77 reads out the image learning blocks stored in the memory 77A and supplies these blocks to the ADRC processing unit 78 and the memory 80.

At the same time, when a block is read from the memory 77A, the blocking unit 77 supplies a control signal indicating the position of the read block to the buffer unit 76. The buffer unit 76 then checks the 3x3 pixel blocks read from the memory 77A in accordance with this control signal, reads the optimal corrected data corresponding to the center pixel for these blocks and stores these data in the memory. It supplies to 80.

Processing then proceeds to step S33 and after the block from the blocking unit 77 is ADRC processed in the ADRC processing unit 78, this block is classified in the class classification unit 79. The result of class classification is supplied to the memory 80 as an address.

In step S34, the optimal correction data supplied from the buffer unit 76 and the block (learning data) supplied from the blocking unit 77 in the memory 80 correspond to the class information supplied from the class classification unit 79. It is stored to correspond to the address.

Processing then proceeds to step S35, where a determination is made as to whether blocks for one frame portion and optimal correction data are stored in the memory 80. As a result of the determination in step S35, when the block for one frame portion and the optimal corrected data are not yet stored in the memory 50, the next block is read out from the blocking unit 77. The optimal corrected data corresponding to the block is also read from the buffer unit 76, and the process returns to step S33, and the process is repeated from here.

If it is determined in step S35 that one block portion of the block and the optimal corrected data are stored in the memory 80, the process proceeds to step S36 and whether the processing for all the images for learning are completed. A decision is made on. When it is determined in step S36 that the processing for all the images for learning is not completed, the processing in step S31 is returned and the processing from step S31 is repeated for the next learning image.

On the other hand, when it is determined that all the image processing for learning has been completed as a result of the determination in step S36, the process proceeds to step S37 for each class so that the calculation unit 81 satisfies the prescribed formula shown in equation (7). The optimal corrected data and blocks stored in the memory 80 are read. Further, in step S38, the calculation unit 81 solves the above formula to minimize the error and calculate the mapping coefficient of each class. These mapping coefficients are then supplied to the memory 82 for storage (step S39), and the process of Fig. 13 is completed.

When the function f is expressed as the linear linear equation described above, the mapping coefficients stored in the memory 82 are stored in the mapping coefficient memory 14 of Fig. 2 for encoding of the image.

In some cases, the number of expressions for obtaining the mapping coefficient cannot be obtained using the number of classes. In this case, the calculation unit 16 of FIG. 2 is output as an average value for _nine pixels, i.e., k ₁ to k ₉ = 1/9, etc., including a 3x3 block output from the blocking unit 11, for example. The mapping coefficient is set to default values.

Next, with reference to FIG. 15, the structure of the receiving apparatus 4 (FIG. 1) is demonstrated.

The encoded data recorded on the reproduced recording medium 2 or the encoded data transmitted via the transmission path 3 are reproduced or received by the reception / reproduction device 91 and supplied to the decoder 92.

Decoder 92 includes elements corresponding to elements of local decoder 27 (FIG. 9). The prediction value is obtained from the corrected data in a manner similar to the manner in which the local decoder 27 operates. The picture containing the predicted value is output as a decoded picture.

The corrected data is data having error information below a prescribed threshold. As a result, an image almost identical to the original image data is obtained in the receiving device 4.

The decoded picture can be obtained at the receiving side without the receiving device 4 shown in Fig. 15 by performing normal interpolation using an apparatus for decoding by interpolation of thinned-out pictures. However, the image quality (resolution) of the decoded image obtained in this case is inferior.

Embodiments of the present invention can be used to encode and decode many different types of image signals, i.e., from television signals in standard format (e.g., NTSC) to high-definition television signals containing large amounts of data. It should be noted that. Also, in these embodiments, although each process is described as being performed for one frame, the process may be performed for one field or two or more frames.

In addition, although block coding is described as being performed on one frame of image at a time, blocks may be constructed by combining pixels of the same position of a plurality of time-sequential frames together.

In addition, although the error information has been described as the square sum of the errors, for example, the absolute value of the error or the sum of the three powers (or higher) of the error can be used as the error information. Which type of information will be used as the error information can be determined based on statistical specificity (eg, convergence).

Another embodiment of the present invention will be described with reference to FIGS. 16 to 18. Only parts of the second embodiment that are different from the first embodiment are described below. Specifically, referring now to FIG. 16, the difference between the image processing apparatus shown in FIG. 16 and the image processing apparatus shown in FIG. 7 is that in FIG. 16, the training image data stored in the memory 21 is stored in the local decoder 27. ') Is provided. In the apparatus of FIG. 7, the learning image data stored in the memory 21 is not used by the local decoder 27.

Referring to Fig. 17, a detailed embodiment of a local decoder 27 'that uses image data for learning is shown. First, as described in detail with reference to FIG. 9, the local decoder 27 stores the prediction value based on the learned prediction coefficients provided from the ROM stored in the prediction coefficient ROM 45 and responsive to the class information of the block of corrected image data. Remember to create it. In contrast, the local decoder 27 'includes an adaptive processing unit 45' that is based on class information and also generates a predictive value based on digital picture data stored in and provided from the memory 21 (FIG. 16). do.

An example of the detailed operation of the adaptive processing unit 45 'will now be described. That is, for example, the predicted value for the pixel value y of the original image data is several pixels X ₁ , X ₂ ,. Pixel values for, (hereinafter referred to as learning data) and the specified prediction coefficients w ₁ , w ₂ ,. When considered to be obtained using a linear first-order combining model defined by the linear combination of, the predicted value can be expressed by the following equation.

If the matrix W comprising the set of prediction coefficients w, the matrix X comprising the set of training data x and the matrix Y 'comprising the set of prediction values are defined as follows:

The following observation equation holds.

When the least square method is applied to this observation equation, a predicted value close to the pixel value y of the original image data is obtained. At this time, if the matrix Y including the set y of the pixel values of the original image data (hereinafter referred to as teacher data) and the matrix E including the set of the residual e with respect to the predicted value of the pixel value y of the original image data are defined as follows: ,

The following residual equation is established from equation (2).

In this case, the prediction coefficient Wi for obtaining a prediction value close to the pixel value y of the original image data can be obtained to minimize the square error.

Therefore, the squared error is differentiated by the use of the prediction coefficient W _i to be zero. In other words, the prediction coefficient W _i which holds the following equation is an optimum value for obtaining a prediction value close to the pixel value y of the original image data.

The following equation is established by differentiating the differential equation (3) with the prediction coefficient W _i .

Equation 6 may be obtained from Equations 4 and 5.

The following normal equation can be obtained by considering the relationship between the learning data x, the prediction coefficient w, the teacher data y, and the residual e generated in Equation 3 from Equation 6 above.

The regular expression of Equation 7 has the same number as the number of prediction coefficients W to be obtained. The optimal prediction coefficient W can be obtained by solving Equation 7. Equation 7 can be solved by applying, for example, a release method (such as Gauss Jordan's elimination method).

In the adaptive processing, an optimum prediction coefficient w is obtained, and a prediction value close to the pixel value y for the original image data is obtained from equation (1) by using this prediction coefficient w. This adaptive processing can be done at the local decoder 27. At this time, it is not necessary to provide the prediction coefficient ROM to the local decoder 27.

Adaptive processing (which may include adaptive processing when the prediction coefficient ROM 45 is used) differs from interpolation processing in that components contained in original image data that are not included in the thinned-out image are reproduced. have. That is, in the case where only Equation 1 is used for the adaptive processing, this is the same as the so-called interpolation processing using an interpolation filter. However, since the prediction coefficient w corresponding to the tap number of this interpolation filter is obtained by learning using the teacher data y, the component included in the original image data can be reproduced. This adaptive process can be said to be an operation process of reproducing a so-called image.

Referring now to FIG. 18, this flowchart shows processing after the local decoder 27 'in FIG. In step S121, the classifying blocking unit 41 blocks the corrected image data. Next, in step S122, the ADRC processing unit 43 performs ADRC processing on blocks of the corrected image data. In step S123, the class classification unit 44 classifies the corrected image data block according to the output of the ADRC processing unit 43. In step S124, the adaptive processing unit 45 'calculates a prediction coefficient for each class (for internal use of the adaptive processing unit 45') and calculates a prediction value. In step S125, the predictive value is output from the adaptive processing unit 45 'of the local decoder 27'.

While the present invention has been described with reference to specific embodiments, it will be apparent to those skilled in the art that many changes may be made without departing from the teachings. However, all of such modifications are intended to be included within the scope of the following claims.

1 is a block diagram showing a configuration of an embodiment in which the present invention is applied to an image processing system.

FIG. 2 is a block diagram illustrating an exemplary configuration of the transmitter of FIG. 1. FIG.

3 shows ADRC processing.

4 is a diagram illustrating class classification processing.

FIG. 5 is a flowchart showing the operation of the transmitter of FIG. 2; FIG.

FIG. 6 shows a process of the blocking unit of FIG. 2. FIG.

Fig. 7 is a block diagram showing a configuration of a first embodiment of a learning image processing apparatus for obtaining mapping coefficients.

FIG. 8 is a flowchart showing an operation of the image processing apparatus of FIG. 7.

9 is a block diagram showing a configuration example of the local decoder 27 of FIG.

FIG. 10 is a flowchart showing processing of the local decoder 27 of FIG.

11 is a block diagram showing a configuration of an embodiment of an image processing apparatus that performs learning to obtain prediction coefficients.

Fig. 12 is a block diagram showing the construction of a second embodiment of an image processing apparatus that performs learning to obtain mapping coefficients.

FIG. 13 is a flowchart showing an operation of the image processing apparatus of FIG. 12. FIG.

FIG. 14 is a flow chart showing the detailed process of step S31 in FIG.

FIG. 15 is a block diagram illustrating an exemplary configuration of the receiving device 4 in FIG. 1.

Fig. 16 is a block diagram showing the construction of another embodiment of a learning image processing apparatus for obtaining mapping coefficients.

FIG. 17 is a flowchart showing an operation of the image processing apparatus of FIG. 16. FIG.

FIG. 18 is a block diagram showing an example of the configuration of a local decoder 27 'of FIG.

<Explanation of symbols for the main parts of the drawings>

1: transmitting device

2: recording medium

3: transmission path

4: receiving device

11: blocking unit

12: ADRC processing unit

13: class classification unit

14: mapping coefficient memory

15: delay unit

16: calculation unit

17: transmit / receive unit

21: memory

22: blocking unit

23: ADRC processing unit

24: class classification unit

27: local decoder

28: error calculation unit

29: judgment unit

30: mapping coefficient setting unit

31: mapping coefficient memory

Claims (69)

An apparatus for encoding original image data having a plurality of pixels to produce coded image data, the apparatus comprising: Means for extracting a plurality of pixels from the original image data and generating class information corresponding to characteristics of the extracted plurality of pixels and representing one of a plurality of classes; Means for storing mapping coefficients for each of the plurality of classes, the mapping coefficients being generated based on a prediction error of training original image data and predictive image data of the training original image data; Means for reading the mapping coefficient corresponding to the generated class information, and Means for producing the encoded image data in response to the original image data and the read mapping coefficients. Original image data encoding apparatus comprising a. The original image data encoding device according to claim 1, wherein the calculating means calculates encoded image data having a smaller number of pixels than the original image data. The method of claim 1, wherein the calculating means, Means for extracting a plurality of pixels from the original image data, and Means for calculating the encoded image data by performing calculations on pixel values of the pixels of one block and the read mapping coefficients. Original image data encoding apparatus comprising a. The original picture data encoding device according to claim 2, wherein mapping coefficients for each of the plurality of classes are generated in response to learning original picture data. The method of claim 2, wherein the mapping coefficients for each class are set so that the prediction error of the predictive image data for learning determined from the difference between the decoded coded data generated from the original image data for learning and the original image data for learning is minimized. A generated original image data encoding device. The method of claim 2, wherein the mapping coefficient for each class defines a prediction error of the predictive image data for learning determined from a difference between the decoded coded data generated from the original image data for learning and the original image data for learning. An original image data encoding device generated to be smaller than a threshold. The method of claim 2, The mapping coefficient for each class is Extracting a plurality of pixels of the original image data for learning, and generating class information corresponding to characteristics of the extracted plurality of pixels; Calculating encoded data in which the number of pixels of the original image data for learning is reduced in response to the learning original image data and the mapping coefficient corresponding to the class information; Generating predictive image data having a plurality of predictive pixels by generating predictive image data for learning according to the encoded data; Generating a prediction error of the predictive image data for learning related to the original image data for learning, Updating the mapping coefficients according to the generated prediction error until the updated mapping coefficients are optimal mapping coefficients, and Determining the optimal mapping coefficient An original image data encoding apparatus generated by the method. The method of claim 2, The mapping for each class is Reducing the number of pixels of the original image data to produce compressed data, Correcting the compressed data to produce corrected data, Predicting the original image data according to the corrected data to generate predictive data having a plurality of predicted pixels; Calculating a prediction error of the prediction data with respect to the original image data; Determining the suitability of the corrected data as encoded data based on the prediction error, Repeating said correcting step until said corrected data is optimal correction data, and Generating an optimal mapping coefficient for each class using the training original image data and the optimal correction data; An original image data encoding apparatus determined by the method. An apparatus for decoding encoded image data generated by encoding image data, Means for receiving the encoded image data, and Means for decoding the coded image data to produce decoded image data Including; The encoded data is Extracting a plurality of pixels from the original image data, and generating class information corresponding to characteristics of the extracted plurality of pixels and representing one of a plurality of classes; Storing mapping coefficients for each of the plurality of classes, wherein the mapping coefficients are generated based on a prediction error of training original image data and predictive image data of the training original image data; Reading a mapping coefficient corresponding to the generated class information, and Calculating the encoded image data in response to the original image data and the read mapping coefficients. The encoded image data decoding device generated by the. The encoded image data decoding device according to claim 9, wherein the encoded image data has fewer pixels than the original image data. The method of claim 10, The decoding means, Memory for storing prediction coefficients for each of a plurality of classes used to generate the mapping coefficients; Means for extracting a plurality of pixels of the encoded image data and generating class information corresponding to the extracted plurality of pixels; Means for reading prediction coefficients corresponding to the generated class information, and Means for calculating the decoded picture data using the coded picture data and the read prediction coefficients. Encoded image data decoding apparatus comprising a. The encoded image data decoding device according to claim 11, wherein the prediction coefficients for each class are generated using the original image data for learning. The encoded image data decoding device according to claim 10, wherein the mapping coefficients for each class are generated using the original image data for learning. 11. The method of claim 10, wherein the mapping coefficients for each class are generated such that the prediction error of the predictive image data for learning determined from the difference between the decoded encoded data generated from the original image data for learning and the original image data for learning is minimized. An encoded image data decoding device. The method of claim 10, wherein the mapping coefficient for each class is defined by a prediction error of the predictive image data for learning determined from the difference between the decoded coded data generated from the original image data for learning and the original image data for learning. An encoded image data decoding device generated to be smaller than a threshold value. A method of encoding original image data having a plurality of pixels to generate encoded image data, Extracting a plurality of pixels from the original image data, and generating class information corresponding to characteristics of the extracted plurality of pixels and representing one of a plurality of classes; Storing mapping coefficients for each of the plurality of classes, wherein the mapping coefficients are generated based on a prediction error of training original image data and predictive image data of the training original image data; Reading a mapping coefficient corresponding to the generated class information, and Calculating coded image data in response to the original image data and the read mapping coefficients Original image data encoding method comprising a. 17. The original image data encoding method according to claim 16, wherein the calculating step calculates encoded image data having a smaller number of pixels than the original image data. The method of claim 16, wherein the calculating step, Extracting a plurality of pixels from the original image data, and Calculating the encoded image data by performing calculation on pixel values of the pixels of the block and the read mapping coefficients. Original image data encoding method comprising a. 18. The original image data encoding method according to claim 17, wherein mapping coefficients for each of the plurality of classes are generated in response to original image data for learning. 18. The prediction coefficient according to claim 17, wherein the mapping coefficients for each class are minimized in the prediction error of the predictive image data for learning determined from the difference between the decoded encoded data generated from the original image data for learning and the original image data for learning. An original image data encoding method generated so as to be possible. 18. The prediction coefficient according to claim 17, wherein the mapping coefficient for each class defines a prediction error of the predictive image data for learning determined from the difference between the decoded coded data generated from the original image data for learning and the original image data for learning. An original image data encoding method generated to be smaller than a threshold. The method of claim 17, The mapping coefficient for each class is Extracting a plurality of pixels of the original image data for learning, and generating class information corresponding to characteristics of the extracted plurality of pixels; Calculating encoded data in which the number of pixels of the original image data for learning is reduced in response to the original image data for learning and the mapping coefficient corresponding to the class information; Generating predictive image data having a plurality of predictive pixels by generating predictive image data for learning according to the encoded data; Generating a prediction error of the predictive image data for learning related to the original image data for learning, Updating the mapping coefficients according to the generated prediction error until the updated mapping coefficients become optimal mapping coefficients, and Determining the optimal mapping coefficient An original image data encoding method generated by the method. The method of claim 17, The mapping for each class is Reducing the number of pixels of the original image data to produce compressed data, Correcting the compressed data to produce corrected data, Predicting the original image data according to the corrected data to generate predictive data having a plurality of predicted pixels; Calculating a prediction error of the prediction data in relation to the original image data, Determining the suitability of the corrected data as encoded data based on the prediction error, Repeating the correcting step until the corrected data becomes optimal correction data, and Generating an optimal mapping coefficient for each class using the training original image data and the optimal correction data; The original image data encoding method determined by the method. A method of decoding coded image data generated by encoding image data, the method comprising: Receiving the encoded image data, and Decoding the coded image data to produce decoded image data Including; The encoded data is, Extracting a plurality of pixels from the original image data, and generating class information corresponding to characteristics of the extracted plurality of pixels and representing one of a plurality of classes; Storing mapping coefficients for each of the plurality of classes, wherein the mapping coefficients are generated based on a prediction error of training original image data and predictive image data of the training original image data; Reading mapping coefficients corresponding to the generated class information; Calculating coded image data in response to the original image data and the read mapping coefficients Coded image data decoding method generated by the. The encoded picture data decoding method according to claim 24, wherein the encoded picture data has fewer pixels than the original picture data. The method of claim 25, wherein the decoding step, Storing prediction coefficients for each of the plurality of classes used to generate the mapping coefficients, Extracting a plurality of pixels of the encoded image data and generating class information corresponding to the extracted plurality of pixels; Reading prediction coefficients corresponding to the generated class information, and Calculating the decoded picture data by using the coded picture data and the read prediction coefficients. Coded image data decoding method comprising a. 27. The coded image data decoding method according to claim 26, wherein the prediction coefficients for each class are generated using the original image data for learning. The encoded image data decoding method according to claim 25, wherein the mapping coefficients for each class are generated using the original image data for learning. 26. The method of claim 25, wherein the mapping coefficients for each class have a minimum prediction error of the predictive image data for learning determined from the difference between the decoded encoded data generated from the original image data for learning and the original image data for learning. A coded image data decoding method generated so as to be generated. 27. The prediction coefficient of claim 25, wherein the mapping coefficient for each class is defined by a prediction error of the predictive prediction image data for learning determined from the difference between the decoded encoded data generated from the original image data for learning and the original image data for learning. A coded image data decoding method generated to be smaller than a predetermined threshold. In the method for transmitting image data having a plurality of pixels, Extracting a plurality of pixels from original image data, and generating class information corresponding to characteristics of the extracted plurality of pixels and representing one of a plurality of classes, Storing mapping coefficients for each of the plurality of classes, wherein the mapping coefficients are generated based on a prediction error of training original image data and predictive image data of the training original image data; Reading mapping coefficients corresponding to the generated class information; Calculating the encoded image data in response to the original image data and the read mapping coefficients, and Transmitting the encoded image data Image data transmission method comprising a. 32. The image data transmission method according to claim 31, wherein said calculating step calculates encoded image data having fewer pixels than said original image data. The method of claim 31, wherein the calculating step, Extracting a plurality of pixels from the original image data, and Calculating the encoded image data by performing calculation on pixel values of the pixels of the block and the read mapping coefficients. Image data transmission method comprising a. 33. The image data transmission method according to claim 32, wherein mapping coefficients for each of the plurality of classes are generated in response to the original image data for learning. 33. The method of claim 32, wherein the mapping coefficients for each class have a minimum prediction error of the predictive image data for learning determined from the difference between the decoded encoded data generated from the original image data for learning and the original image data for learning. The image data transmission method generated as much as possible. 33. The method of claim 32, wherein the mapping coefficient for each class defines a prediction error of the predictive prediction image data for learning determined from the difference between the decoded encoded data generated from the original image data for learning and the original image data for learning. An image data transmission method generated to be smaller than a threshold value. 33. The method of claim 32, The mapping coefficient for each class is Extracting a plurality of pixels of the original image data for learning, and generating class information corresponding to characteristics of the extracted plurality of pixels; Calculating encoded data in which the number of pixels of the original image data for learning is reduced in response to the mapping coefficients corresponding to the original image data for learning and the class information; Generating predictive image data having a plurality of predictive pixels by generating predictive image data for learning according to the encoded data; Generating a prediction error of the predictive image data for learning related to the original image data for learning, Updating the mapping coefficients according to the generated prediction error until the updated mapping coefficients become optimal mapping coefficients, and Determining the optimal mapping coefficient Image data transmission method generated by the. 33. The method of claim 32, The mapping for each class is Reducing the number of pixels of the original image data to produce compressed data, Correcting the compressed data to produce corrected data, Predicting the original image data according to the corrected data to generate predictive data having a plurality of predicted pixels; Calculating a prediction error of the prediction data in relation to the original image data, Determining the suitability of the corrected data as encoded data based on the prediction error, Repeating the correcting step until the corrected data becomes optimal correction data, and Generating an optimal mapping coefficient for each class using the training original image data and the optimal correction data; Image data transmission method determined by. A computer-readable recording medium having recorded thereon a program for encoding original image data having a plurality of pixels to produce coded image data, comprising: The program, Extracting a plurality of pixels from original image data and generating class information corresponding to characteristics of the extracted plurality of pixels and representing one of a plurality of classes; Storing mapping coefficients for each of the plurality of classes, the mapping coefficients being generated based on a prediction error of training original image data and predictive image data of the training original image data; Reading the mapping coefficients corresponding to the generated class information; Calculating coded image data in response to the original image data and the read mapping coefficients; And Recording the coded image data Computer-readable recording medium comprising a. 40. The computer readable recording medium of claim 39, wherein the calculating step generates the encoded image data having fewer pixels than the original image data. 40. The method of claim 39, wherein said calculating step Extracting a plurality of pixels from the original image data; And Calculating the encoded image data by performing calculation on pixel values of the pixels of the block and the read mapping coefficients. Computer-readable recording medium comprising a. 41. The computer program product of claim 40, wherein the mapping coefficients for each of the plurality of classes are generated in response to learning original image data. 41. The method of claim 40, wherein the mapping coefficients for each class are generated such that a prediction error of the predictive prediction image data determined from the difference between the decoded encoded data generated from the training original image data and the training original image data is minimized. Computer-readable recording medium. 41. The method according to claim 40, wherein the mapping coefficients for each class have a predetermined threshold of a prediction error of the learning prediction image data determined from the difference between the decoded encoded data generated from the learning original image data and the learning original image data. A computer readable recording medium generated to be smaller. The method of claim 40, The mapping coefficients for each class are Extracting a plurality of pixels of the learning original image data and generating class information corresponding to characteristics of the extracted plurality of pixels; Calculating encoded data in which the number of pixels of the learning original image data is reduced in response to the learning original image data and mapping coefficients corresponding to the class information; Generating predictive data having a plurality of predicted pixels by generating predictive predictive image data according to the encoded data; Generating in the prediction error of the learning predictive image data related to the learning original image data, Updating the mapping coefficients according to the generated prediction error until updated mapping coefficients become optimal mapping coefficients, and Determining the optimal mapping coefficients A computer-readable recording medium generated by the. The method of claim 40, The mapping for each class is Reducing the number of pixels of the original image data to generate compressed data; Correcting the compressed data to generate correction data; Predicting the original image data according to the correction data to generate prediction data having a plurality of prediction pixels; Calculating a prediction error of the prediction data in relation to the original image data; Determining the suitability of the correction data as encoded data based on the prediction error; Repeating the correction step until the correction data becomes optimal correction data; And Generating optimal mapping coefficients for each class using the training original image data and the optimal correction data. A computer-readable recording medium determined by. In the method for transmitting encoded image data, Receiving the encoded image data; And Transmitting the encoded image data Including; The coded image data is Extracting a plurality of pixels from original image data, and generating class information corresponding to characteristics of the extracted plurality of pixels and representing one of a plurality of classes, Storing mapping coefficients for each of the plurality of classes, wherein the mapping coefficients are generated based on prediction errors of training original image data and predicted image data of training original image data; Reading the mapping coefficients corresponding to the generated class information; Calculating coded image data in response to the original image data and the read mapping coefficients, and Recording the coded image data Image data transmission method generated by the. 48. The image data transmission method according to claim 47, wherein said calculating step generates said encoded image data having fewer pixels than said original image data. The method of claim 47, wherein the calculating step, Extracting a plurality of pixels from the original image data, and Calculating the encoded image data by performing calculation on pixel values of the pixels of the block and the read mapping coefficients. Image data transmission method comprising a. 49. The method of claim 48, wherein the mapping coefficients for each of the plurality of classes are generated in response to learning original image data. 49. The method of claim 48, wherein the mapping coefficients for each class are generated such that a prediction error of the predictive prediction image data determined from the difference between the decoded encoded data generated from the training original image data and the training original image data is minimized. Image data transmission method. 49. The method according to claim 48, wherein the mapping coefficients for each class are based on a predetermined threshold of a prediction error of the prediction prediction image data determined from the difference between the decoded encoded data generated from the training original image data and the training original image data. An image data transmission method generated to be smaller than a value. The method of claim 48, The mapping coefficients for each class are Extracting a plurality of pixels of the learning original image data and generating class information corresponding to characteristics of the extracted plurality of pixels; Calculating encoded data in which the number of pixels of the learning original image data is reduced in response to the learning original image data and mapping coefficients corresponding to the class information; Generating predictive data having a plurality of predicted pixels by generating predictive predictive image data according to the encoded data; Generating a prediction error of the learning predictive image data related to the learning original image data; Updating the mapping coefficients according to the generated prediction error until updated mapping coefficients become optimal mapping coefficients, and Determining the optimal mapping coefficients Image data transmission method generated by the. The method of claim 48, The mapping for each class is Reducing the number of pixels of the original image data to generate compressed data, Correcting the compressed data to generate correction data; Predicting the original image data according to the correction data to generate prediction data having a plurality of prediction pixels; Calculating a prediction error of the prediction data in relation to the original image data, Determining suitability of the correction data as encoded data based on the prediction error, Repeating the correction step until the correction data becomes optimal correction data, and Generating optimal mapping coefficients for each class using the training original image data and the optimal correction data. Image data transmission method determined by. An apparatus for encoding original image data having a plurality of pixels to produce encoded image data, An extraction circuit for extracting a plurality of pixels from the original image data and generating class information corresponding to characteristics of the extracted plurality of pixels and representing one of a plurality of classes; A memory circuit for storing mapping coefficients for each of the plurality of classes, the mapping coefficients being generated based on a prediction error of learning original image data and predictive image data of the learning original image data; A reading circuit that reads the mapping coefficients corresponding to the generated class information; And An encoding circuit for calculating the encoded image data in response to the original image data and the read mapping coefficients Original image data encoding apparatus comprising a. 56. The original image data encoding device according to claim 55, wherein the encoding circuit generates the encoded image data having fewer pixels than the original image data. 56. The apparatus of claim 55, wherein the encoding circuit is An extraction circuit for extracting a plurality of pixels from the original image data, and An encoding circuit for calculating the encoded image data by performing calculations on pixel values of the pixels of the block and the read mapping coefficients. Original image data encoding apparatus comprising a. 57. The apparatus of claim 56, wherein the mapping coefficients for each of the plurality of classes are generated in response to training original image data. 57. The method of claim 56, wherein the mapping coefficients for each class are generated such that a prediction error of the predictive prediction image data determined from the difference between the decoded encoded data generated from the training original image data and the training original image data is minimized. An original image data encoding device. 57. The method according to claim 56, wherein the mapping coefficients for each class have a predetermined threshold of a prediction error of the learning prediction image data determined from the difference between the decoded encoded data generated from the learning original image data and the learning original image data. An original image data encoding apparatus generated to be smaller than a value. The method of claim 56, wherein The mapping coefficients for each class are Extracting a plurality of pixels of the learning original image data to generate class information corresponding to characteristics of the extracted plurality of pixels; Calculating encoded data in which the number of pixels of the learning original image data is reduced in response to the learning original image data and mapping coefficients corresponding to the class information; Generating predictive data having a plurality of predicted pixels by generating predictive predictive image data according to the encoded data; Generating a prediction error of the learning predictive image data related to the learning original image data; Updating the mapping coefficients according to the generated prediction error until updated mapping coefficients become optimal mapping coefficients, and Determining the optimal mapping coefficients An original image data encoding device generated by the method. The method of claim 56, wherein The mapping for each class is Reducing the number of pixels of the original image data to generate compressed data; Correcting the compressed data to generate correction data; Predicting the original image data according to the correction data to generate prediction data having a plurality of prediction pixels; Calculating a prediction error of the prediction data in relation to the original image data; Determining the suitability of the correction data as encoded data based on the prediction error; Repeating the correction step until the correction data becomes optimal correction data; And Generating optimal mapping coefficients for each class using the training original image data and the optimal correction data. An original image data encoding apparatus determined by the method. An apparatus for decoding encoded image data generated by encoding image data, the apparatus comprising: A receiving circuit which receives the encoded image data, and Decoding circuit for decoding the coded image data to produce decoded image data Including; The coded image data is Extracting a plurality of pixels from original image data, and generating class information corresponding to characteristics of the extracted plurality of pixels and representing one of a plurality of classes, Storing mapping coefficients for each of the plurality of classes, the mapping coefficients being generated based on a prediction error of training original image data and predictive image data of the training original image data; Reading the mapping coefficients corresponding to the generated class information, and Calculating coded image data in response to the original image data and the read mapping coefficients The encoded image data decoding device generated by the. 66. The encoded picture data decoding apparatus according to claim 63, wherein said encoded picture data has fewer pixels than said original picture data. The method of claim 64, wherein the decoding circuit, A memory for storing prediction coefficients for each of the plurality of classes used to generate the mapping coefficients; An extraction circuit for extracting a plurality of pixels of the encoded image data to generate class information corresponding to the extracted plurality of pixels; A reading circuit for reading prediction coefficients corresponding to the generated class information, and An encoding circuit for calculating the decoded image data using the encoded image data and the read prediction coefficients Encoded image data decoding apparatus comprising a. 66. The encoded picture data decoding apparatus according to claim 65, wherein the prediction coefficients for each class are generated using learning original picture data. The encoded picture data decoding apparatus according to claim 64, wherein said mapping coefficients for each class are generated using said learning original picture data. 65. The method of claim 64, wherein the mapping coefficients for each class are generated such that a prediction error of the predictive prediction image data determined from the difference between the decoded encoded data generated from the training original image data and the training original image data is minimized. An encoded image data decoding device. 65. The method according to claim 64, wherein the mapping coefficients for each class have a predetermined threshold of a prediction error of the predictive prediction image data determined from the difference between the decoded encoded data generated from the training original image data and the training original image data. A coded image data decoding device generated to be smaller than a value.

Images