JP4201192B2 - Image decoding apparatus, program, and information recording medium - Google Patents

Description

The present invention relates to the field of image processing, and more particularly to a technique for decoding code data of an image.

  One of the problems in the decoding of code data encoded by, for example, JPEG2000, which includes a frequency conversion process for converting image data into a plurality of subband coefficients and a quantization process for quantizing the subband coefficients. Is how to improve image quality degradation due to quantization error during encoding.

  Regarding this problem, in decoding of JPEG2000 code data, a parameter for inverse quantization of subband coefficients (called a reconstruction parameter in JPEG2000) is adjusted by using a neighborhood coefficient before inverse quantization. An apparatus and method for performing such a process is described in Japanese Patent Application Laid-Open No. H10-228707.

  Since JPEG2000 relates to the embodiments of the present invention described later, an outline of JPEG2000 will be described here.

  FIG. 24 is a block diagram for explaining the algorithm of JPEG2000. In JPEG2000, at the time of encoding, image data is divided into one or two or more rectangular areas called tiles that do not overlap and processed for each tile. Assuming a color image, the image data of each component of each tile is subjected to DC level shift and color space conversion by the DC level shift / inverse shift / color space conversion / inverse conversion unit 111, and then wavelet conversion is performed for each component. The unit 112 applies a two-dimensional wavelet transform (discrete wavelet transform) to divide the space into a plurality of subbands (frequency bands).

  FIG. 25 shows a state of subband division when the number of decomposition levels (number of conversions) is three. By applying the first two-dimensional wavelet transform to the tile image data (decomposition level 0) shown in (a), the subbands of 1LL, 1HL, 1LH, and 1HH of the decomposition level 1 shown in (b) It is divided into. Next, by applying the second two-dimensional wavelet transform to the coefficients of the 1LL subband, the 1LL subband is divided into 2LL, 2HL, 2LH, and 2HH subbands of decomposition level 2 as shown in (c). Is done. By applying the third two-dimensional wavelet transform to the coefficients of the 2LL subband, the 2LL subband is divided into 3LL, 3HL, 3LH, and 3HH subbands at the decomposition level 3 as shown in (d).

  JPEG2000 defines an irreversible wavelet transform called 9 × 7 transform and a reversible wavelet transform called 5 × 3 transform. When the 9 × 7 transform is used, the wavelet coefficient is linearly quantized for each subband by the quantization / inverse quantization unit 113.

  Next, each subband coefficient is entropy encoded by the entropy encoding / decoding unit 114 for each bit plane (strictly, each bit plane is divided into three sub bit planes called encoding paths). The context for this encoding is generated by the quantization / inverse quantization unit 113.

  In this entropy coding, each subband is divided into non-overlapping rectangles called “precincts”. This was introduced to use memory efficiently in implementation. The precinct is further divided into rectangular “code blocks” that do not overlap. This code block is the basic unit of entropy coding. FIG. 27 shows the relationship among tiles, subbands, precincts, and code blocks. Precincts corresponding to LH, HL, and HH at the same decomposition level are handled as one set, but precincts of the LL subband are handled independently.

  The tag processing unit 115 collects a part of the code of the code block in the precinct to generate a packet, arranges the packet in a predetermined order, and adds a necessary tag and tag information, thereby adding one code stream (code). Data). As shown in FIG. 26, this code stream has a structure in which tag information called a tile part header and a packet sequence are arranged for each tile, followed by tag information called a main header, and an end tag is added at the end.

  The decoding process is the reverse of the encoding process. The tag processing unit 115 interprets tag information added to the code stream, and decomposes the code stream into code streams of each tile of each component. Then, the entropy encoding / decoding unit 114 performs entropy decoding for each code stream of each tile of each component. At this time, the position of the bit to be decoded is determined in the order based on the tag information in the code stream, and the inverse quantization unit 113 generates a context from the sequence of the peripheral bits (already decoded) of the target bit. The Decoding is performed by probability estimation from the context and the code stream to generate a target bit, which is written in the position of the target bit. When the wavelet coefficients decoded in this way are linearly quantized at the time of encoding, the quantization / inverse quantization unit 113 uses the same number of quantization steps as that at the time of quantization for each subband. Inverse quantization. The wavelet coefficients after the inverse quantization are returned to tile image data by applying a two-dimensional inverse wavelet transform by the wavelet transform / inverse transform unit 112. The tile image data is subjected to reverse color conversion and reverse DC level shift in the DC level shift / inverse shift / color space conversion / inverse conversion unit 111, and returned to the original color image tile image data.

JP 2003-189107 A

  In the apparatus and method described in Patent Document 1, the recontraction parameter for the coefficient whose value before inverse quantization is 0 is always determined to be 0 (see FIGS. 3 to 7 in Patent Document 1). The coefficient quantized to 0 at the time of conversion is always dequantized to 0 at the time of decoding. In JPEG2000, there is a promise that a reconstruction parameter for dequantizing a coefficient having a value of 0 (a parameter indicating where the coefficient is dequantized in the quantization interval) is 0. The apparatus and method follow that convention, and do not assume that a coefficient quantized to 0 at the time of encoding is dequantized to a value other than 0 at the time of decoding.

  Now, the main cause of image quality degradation of the decoded image of JPEG2000 code data is that a coefficient that is not 0 is quantized to 0 at the encoding stage, and the coefficient is dequantized to 0 as it is at the decoding stage. Yes (this is not limited to JPEG2000 code data only, and the same applies to code data in other encoding methods including subband division and subband coefficient quantization). However, in the apparatus and method described in Patent Document 1, since such a coefficient is always inversely quantized to 0, there is a problem in that image quality degradation (such as edge breakage) caused by such a coefficient cannot be improved.

An object of the present invention is to provide a novel image decoding apparatus, a program, and an information recording medium that can solve the problems described above.

According to the first aspect of the present invention, code data encoded by an encoding method including a frequency conversion step of converting image data into a plurality of subband coefficients and a quantization step of quantizing the subband coefficients are obtained. In an image decoding apparatus that performs decoding by inverse quantization and inverse frequency transformation, prior to inverse frequency transformation, subband coefficients to which the coefficients belong are subordinate to the subband coefficients to which the coefficients belong. An image decoding apparatus comprising coefficient correction means for correcting using a neighborhood coefficient in a direction highly correlated with the coefficient.

  According to a second aspect of the present invention, in the image decoding apparatus according to the first aspect of the invention, the coefficient correction means includes means for determining whether or not the coefficient value is 0, and excludes a coefficient whose value is not 0 from the correction target. An image decoding apparatus characterized by the above.

According to a third aspect of the present invention, in the image decoding device according to the first or second aspect of the present invention, the coefficient correction means includes means for determining whether or not the number of quantization steps for the subband to which the coefficient belongs is smaller than a predetermined value, An image decoding apparatus characterized in that a coefficient of a subband whose number of quantization steps is smaller than a predetermined value is excluded from correction targets.

According to a fourth aspect of the present invention, in the image decoding device according to any one of the first to third aspects, the frequency transformation is a two-dimensional discrete wavelet transformation, and the inverse frequency transformation is a two-dimensional inverse discrete wavelet transformation. An image decoding apparatus characterized by the above.

According to a fifth aspect of the present invention, in the image decoding device according to the fourth aspect of the present invention, the HL coefficient is an average of the upper and lower HL coefficients of the correction target coefficient, and the correction value of the coefficient is used as the HL coefficient. It is.

According to a sixth aspect of the present invention, in the image decoding apparatus according to the fourth aspect of the present invention, the LH coefficient is an average of the left and right LH coefficients of the coefficient to be corrected as a correction value of the coefficient. It is.

According to a seventh aspect of the present invention, in the image decoding device according to the fourth aspect of the present invention, the HH coefficient is an average of the upper right, upper left, lower right, and lower left HH coefficients of the coefficient to be corrected as a correction value of the coefficient. An image decoding apparatus characterized by the above.

The invention of claim 8 is a program for causing a computer to function as coefficient correction means according to any one of claims 1 to 7 .

A ninth aspect of the invention is a computer-readable information recording medium on which a program according to the eighth aspect of the invention is recorded.

In the present invention , after the inverse quantization, the coefficient is corrected by using the neighborhood coefficient, so there is no restriction like the reconstruction parameter described above. Therefore, “the coefficient that is 0 after dequantization, which is originally 0”, which is the main cause of image quality degradation, can be corrected to an original value or a value close to the original value, and such coefficients are decoded without being corrected. In such a case, it is possible to restore the breakage of the edge generated in the decoded image and improve the image quality. In addition, since the degree of freedom of the correction content using the coefficient near the coefficient is high, it is possible to improve the image quality such as sharpening the edge.

Further, since the direction in which the coefficient is strongly correlated and the strength of the correlation vary depending on the type of subband to which the coefficient belongs, the position of the neighborhood coefficient to be used for correcting the coefficient should be selected according to the type of subband. According to the present invention, it is possible to correct a coefficient by using a neighborhood coefficient at a position optimal for the type of subband to which the coefficient belongs. In particular, by using the neighborhood coefficient in the (high) direction where the coefficient correlation is strong, it is possible to improve the image quality by appropriately correcting the coefficient particularly at the edge portion or the like.

Further, in the present invention, only coefficients whose dequantized value is 0, which is the main cause of image quality degradation, are subject to correction processing, and other coefficients are excluded from correction targets. Compared with the case where this is done, more efficient correction processing is possible.

Furthermore, the effects of quantization error or the need for coefficient correction or the effects of incorrect coefficient correction are not uniform across all subbands. In the present invention, it is possible to rationally narrow down the subbands to be corrected in consideration of the above , and to improve the processing efficiency while effectively suppressing the influence of the quantization error.

  Hereinafter, a typical embodiment of the present invention will be described with reference to an example in which JPEG2000 code data is a processing target. The present invention is widely applied not only to the case where JPEG 2000 code data is to be processed, but also to the case where code data based on an encoding method including subband division and subband coefficient quantization steps is to be processed. To get.

  FIG. 1 is a block diagram for explaining a typical configuration of an image decoding apparatus according to the present invention. Since this image decoding apparatus is assumed to decode JPEG2000 code data, tag processing means 11, entropy decoding means 12, inverse quantization means 13, and inverse wavelet transform means 15 are used as its basic components. And an inverse color space conversion / inverse DC level shift means 16. Since the function of each of these means is clear from the description related to FIG. 24, the description thereof will not be repeated here.

  A feature of this image decoding apparatus is that a coefficient correction unit 14 is added between the inverse quantization unit 13 and the inverse wavelet transform unit 15, and the coefficient correction unit 14 performs a correction process on the wavelet coefficient after the inverse quantization. The inverse wavelet transform means 15 applies the inverse wavelet transform to the coefficient after this correction processing. Note that the information regarding the number of quantization steps or the quantization interval necessary for performing inverse quantization in the inverse quantization means 13 is acquired from the tag information in the code data by the tag processing means 11, and this information will be described later. As described above, it is also used as necessary in the processing in the coefficient correction means 14.

  An overall processing flow in this image decoding apparatus is shown in FIG. The flowchart of FIG. 2 is also a flowchart showing a typical processing procedure of the image decoding method according to the present invention.

  The processing flow (or processing procedure) will be described with reference to FIG. 2. For example, code data to be processed (here, JPEG2000 code data) is input to the tag processing means 11 from a storage device inside or outside the device. (Step S1). Tag information in the code data is interpreted by the tag processing means 11, and based on the result, the code data is decomposed into a code stream of each tile of each component (step S2). Entropy decoding is performed by the entropy decoding means 12 (step S3). As described above, the context for entropy decoding is generated based on the wavelet coefficients decoded by the inverse quantization means 13. The decoded wavelet coefficients are dequantized by the dequantization means 13 for each subband (step S4). The number of quantization steps for each subband used in this inverse quantization is the information extracted from the tag information (a pair of mantissa and exponent corresponding to the subband in JPEG2000, or one set corresponding to the LL subband. It is determined by calculation based on the mantissa and the exponent. Correction processing is performed on the coefficient after inverse quantization by the coefficient correction unit 14 (step S5), and then inverse wavelet transform is performed by the inverse wavelet transform unit 15 (step S6). The image data of each tile of each component restored in this way is subjected to inverse color space conversion and inverse DC level shift by the inverse color space conversion / inverse DC level shift means 16 to obtain the original color system image data. (Step S7). The image data decoded in this way is output, for example, to storage means inside or outside the apparatus.

  The image decoding apparatus or the image decoding method according to the present invention as described above can be realized by hardware, software, or a combination thereof. With reference to FIG. 3, a description will be given of an embodiment in which an image decoding apparatus or an image decoding method according to the present invention is realized using a general-purpose computer such as a personal computer or a computer such as a built-in microcomputer.

  In FIG. 3, 21 is a CPU (Central Processing Unit), 22 is a main memory including a RAM for temporarily storing a program executed by the CPU 21, processing data, and the like, 23 is a program and data accumulated. An auxiliary storage device 25 including a hard disk device for storing data and the like is a bus for transferring data and control information between the above-described units.

  On such a computer, the image decoding apparatus shown in FIG. 1 or the processing shown in FIG. 2 can be realized by one or more programs. In this case, according to the program, the computer functions as the units 11 to 16 shown in FIG. 1 to execute the processes of steps S2 to S7 in FIG. In this case, for example, code data is read from the auxiliary storage device 23 into the main memory 22 and processed, and image data as a processing result is obtained on the main memory 22. This image data is written out to the auxiliary storage device 23, for example. Such a program and a computer-readable information recording (storage) medium such as a magnetic disk, an optical disk, a magneto-optical disk, and a semiconductor storage element on which the program is recorded are also included in the present invention.

  As indicated by a broken line in FIG. 3, when the computer has a JPEG2000-compliant hardware decoder 24, only the coefficient correction unit 14 shown in FIG. 1 or step S5 which is the coefficient correction processing step of FIG. Can be realized by a program. In this case, the processing up to the inverse quantization means 13, that is, the processing of steps S2 to S4 is executed by the decoder 24, and the generated inverse quantized coefficient is read into the main memory 22, and the coefficient correction process is performed according to the program. Is made. For the corrected coefficient, the decoder 24 executes processing after the inverse wavelet transform. Such a program and various information recording (storage) media on which the program is recorded are also included in the present invention.

  It is also possible to incorporate the function of the coefficient correction means 14 into the decoder 24. Such a decoder 34 is also an embodiment of the image decoding apparatus according to the present invention. More generally, any one or more of the means shown in FIG. 1 can be realized by hardware, and the remaining means can be realized by various programs.

  FIG. 4 is a flowchart showing a typical flow of the process of the coefficient correction means 14, that is, the process of step S5 (coefficient correction process). Each step in FIG. 4 corresponds to the function of the means constituting the coefficient correction means 14. That is, the coefficient correction means 14 includes means having a function corresponding to each step in FIG. Each step in FIG. 4 corresponds to the processing step in step S5. Naturally, one or more programs for realizing such means or processing steps by a computer and various information recording (storage) media on which the programs are recorded are also included in the present invention.

  In FIG. 4, one coefficient after inverse quantization is selected in step S11. When there are no more coefficients to be selected (step S12, YES), the process ends.

  In step S13, it is determined whether or not the selected coefficient (coefficient of interest) satisfies a condition for correction. When the condition is not satisfied (step S13, NO), no processing is performed on the current attention coefficient, and the process returns to step S11 to select another coefficient. Note that an aspect in which this condition determination step is omitted, that is, an aspect in which all coefficients are to be corrected is also included in the present invention.

  If the coefficient of interest satisfies the conditions for correction (step S13, YES), in the next step S14, the correction value of the coefficient of interest using the value of the coefficient near the coefficient of interest in the same subband as the coefficient of interest. Ask for. In step S15, it is determined whether or not clipping processing of the correction value is necessary. If the clipping process is necessary, the clipping process is performed in step S16, and the value of the attention coefficient is rewritten with the correction value after clipping in step S17. If the clipping process is unnecessary (NO in step S15), in step S17, the value of the attention coefficient is rewritten with the correction value calculated in step S14. Returning from step S17 to step S11, the same processing is repeated for another coefficient.

  5 and 6 are diagrams for explaining steps S15 and S16. The number of quantization steps of the subband to which the attention coefficient belongs is Q (> 0), and the value of the attention coefficient after inverse quantization is 0 as shown in FIGS. 5 (a) and 6 (a). To do.

  When the correction value for this attention coefficient protrudes in the positive direction from the quantization interval [−Q, + Q] as shown in FIG. 5B, the correction value is changed to the positive value of the quantization interval as shown in FIG. Correct to the side edge value. Conversely, when the correction value protrudes in the negative direction from the quantization interval as shown in FIG. 6B, the correction value is changed to the value at the negative end of the quantization interval as shown in FIG. 6C. Correct it. That is, when the correction value becomes an abnormal value that protrudes from the quantization interval, it is determined in step S15 that clipping processing is necessary, and in step S16, the correction value is clipped to the end near the quantization interval.

  If the coefficient is corrected to an abnormal value that protrudes from the quantization interval, the image quality of an image decoded using the corrected coefficient may be deteriorated. In order to avoid such image quality deterioration, the clipping process as described above is performed.

  Hereinafter, steps S13 and S14 will be described more specifically.

  In this embodiment, in step S13, as shown in FIG. 7, when the value (X) of the attention coefficient is 0, it is determined as a correction target (step S1300). That is, the coefficient correction unit 14 according to the present embodiment includes a determination unit corresponding to step S1300. As described above, the main cause of image quality degradation of JPEG2000 is that a coefficient that is not 0 is quantized to 0 at the time of encoding, and the value after inverse quantization of the coefficient becomes 0 at the time of decoding. . In view of this, in this embodiment, by selecting only the coefficient for which the value after inverse quantization, which is the primary cause of image quality degradation, is 0, the effect of image quality improvement by correction is achieved. On the other hand, the processing target coefficient is reduced to improve the processing efficiency.

  In this embodiment, the process shown in FIG. 8 is performed in step S14. First, the type of subband to which the attention coefficient belongs is determined (step S1401). That is, the coefficient correction means 14 according to the present embodiment includes means for determining the subband type.

If the subband to which the coefficient of interest belongs is an HL subband, this subband is a vertical edge component (vertical low frequency / horizontal high frequency component), and the coefficient has a strong (high) correlation in the vertical direction. Accordingly, in the HL subband to which the attention coefficient belongs, as shown in FIG. 9, the vertical vicinity coefficients a and b of the attention coefficient are selected and the average value is calculated, and this is used as the correction value (step S1403). .

  In the case of the LH subband, this subband is a horizontal edge component (high frequency component in the vertical direction and low frequency component in the horizontal direction), and the coefficient has a strong horizontal correlation. Therefore, in the HL subband to which the attention coefficient belongs, as shown in FIG. 10, the horizontal vicinity coefficients a and b of the attention coefficient are selected, the average value thereof is calculated, and this is used as the correction value (step S1404). .

  In the case of the HH subband, this subband is an edge component in the oblique direction (high frequency component in the oblique direction), and the correlation of the coefficient in the oblique direction is strong. Accordingly, in the HH subband to which the attention coefficient belongs, as shown in FIG. 11, the neighboring coefficients a, b, c, and d in the diagonal direction of the attention coefficient are selected, and the average value is calculated as a correction value (step S1405). .

  In the case of the LL subband, this is a component of a low frequency in the vertical direction and a low frequency in the horizontal direction and has a low directivity. However, since the correlation between the coefficients is relatively strong in the vertical direction and the horizontal direction, FIG. As shown, the horizontal and vertical neighborhood coefficients a, b, c, and d of the attention coefficient are selected, and the average value is calculated as a correction value (step S1402).

  As described above, the coefficient correction means 14 according to the present embodiment includes means for selecting the position of the neighborhood coefficient used for correcting the attention coefficient according to the type of the subband to which the attention coefficient belongs as described above. It is out.

  Note that the present embodiment includes the following modifications.

<Modification 1_1>
In step S1402, the average value of the neighboring coefficients in the diagonal direction may be calculated in the same manner as the attention coefficient belonging to the HH subband, or the average value of the eight neighboring coefficients of the attention coefficient may be calculated.

<Modification 1_2>
The condition of step S13 can be unconditional. That is, all coefficients are to be corrected. In this case, the average value including the attention coefficient may be calculated in steps S1402 to S1405.

<Modification 1_3>
In step S13, it is possible to determine whether the attention coefficient is a coefficient of the LL subband, and to exclude the coefficient of the LL subband from the correction target. Since the number of quantization steps for the LL subband coefficients is generally small, the quantization error is usually small, and when the correction for the LL subband coefficients is inappropriate (the coefficient of truly 0 is changed to a value other than 0). In consideration of the large influence on the image quality (when corrected), the LL subband coefficient is excluded from the correction target in this modification. For example, as shown in FIG. 13, in addition to 0 determination of the attention coefficient (X) (step S1301), it is determined whether the subband to which the attention coefficient belongs is an LL subband (step S1302). Only the coefficient having a value of 0 among the subband coefficients may be processed. In this case, step S1402 (FIG. 8) is omitted. In this case, the determination in step S1301 can be omitted.

  The feature of the coefficient correction means 14 according to this modification is that it includes a determination means corresponding to step S1302.

<Modification 1_4>
In step S13, it is determined whether the quantization step number (Q) of the subband to which the target coefficient belongs is smaller than a predetermined value (TH), and the subband coefficient smaller than the predetermined value is excluded from the correction target. It may be. That is, since the coefficient of the subband having a small number of quantization steps has a small quantization error, the necessity for correction is small, so that the possibility of an adverse effect of erroneous correction is taken into consideration and excluded from the correction target. For example, as shown in FIG. 14, in addition to 0 determination of the attention coefficient (X) (step S1303), comparison determination (step) with a predetermined value (TH) of the number of quantization steps (Q) of the subband to which the attention coefficient belongs. S1304) is performed, and it is possible to process only the coefficient of the subband having a quantization step number larger than a predetermined value and having a value of 0. In this case, the determination in step S1303 can be omitted. The feature of the coefficient correction means 14 according to this modification is that it includes a determination means corresponding to step S1304.

<Modification 1_5>
In step S13, it may be determined whether the subband hierarchy (decomposition level) to which the attention coefficient belongs is lower than the predetermined hierarchy (k), and only the coefficients of the hierarchy lower than the predetermined hierarchy may be processed. Compared to the lower layer subbands, the number of quantization steps in the upper layer subbands is generally quite small, so the coefficient quantization error is small and less need to be corrected. In consideration of this, only the coefficients in the lower hierarchy than the predetermined hierarchy are processed. For example, as shown in FIG. 15, in addition to the determination of the attention coefficient 0 (step S1305), a comparison determination (step S1306) with the number of hierarchies of the predetermined hierarchy (k) is performed. For example, when the number of layers (the number of decomposition levels) is 4, by setting the predetermined layer k = 3, only the coefficients having values of 0 belonging to the subbands of layer 2 and layer 1 are subject to correction. be able to. In this case, the determination in step S1305 can be omitted.

  The feature of the coefficient correction means 14 according to this modification is that it includes a determination means corresponding to step S1306.

<Modification 1_6>
In step S13, it is determined whether or not the attention coefficient is a coefficient corresponding to a specific image area, and only the coefficient corresponding to the specific image area can be processed. For example, as shown in FIG. 16, in addition to the determination of the attention coefficient 0 (step S1307), it is possible to determine whether or not the attention coefficient corresponds to a specific image region (step S1308). The specific image region is, for example, an image region designated by the user or designated by default, and the image region can be designated in units of tiles, code blocks, or precincts. In addition, an image area including the ROI area can be designated as a specific image area. In the JPEG2000 code code, it is possible to identify a tile including the ROI area only by referring to the tag information. By checking the dynamic range of the coefficient, it is possible to identify whether or not the ROI area is included in code block units. In this case, the determination in step S1307 can be omitted.

  The feature of the coefficient correction means 14 according to this modification is that it includes a determination means corresponding to step S1308.

  Note that in the modified examples 1_1 to 1_5, the determination in step S1308 can be added.

  In this embodiment, processing as shown in FIG. 17 is performed in step S14. First, the type of subband to which the attention coefficient belongs is determined (step S1411).

  In the case of the HL subband, since this subband has a vertical edge component and a strong correlation of coefficients in the vertical direction, in the subband to which the target coefficient belongs, the vertical centered on the target coefficient (shaded portion) as shown in FIG. By applying the direction smoothing filter, the correction value of the attention coefficient is obtained (step S1403). The numbers in FIG. 18 are filter coefficients (weighting coefficients) at each coefficient position. A correction value is obtained by dividing the sum of products of filter coefficients corresponding to the values of the coefficients by the sum of filter coefficients.

  In the case of the LH subband, the coefficient correlation is strong in the horizontal direction. Therefore, in the subband to which the target coefficient belongs, by applying a horizontal smoothing filter centered on the target coefficient (shaded portion) as shown in FIG. Then, a correction value of the attention coefficient is obtained (step S1414). The numbers in FIG. 19 are filter coefficients (weighting coefficients) at each coefficient position.

  In the case of the HH subband, since the coefficient correlation of the subband is strong in the diagonal direction, an oblique direction smoothing filter centered on the target coefficient as shown in FIG. 20 is applied to the subband to which the target coefficient belongs. Then, a correction value of the attention coefficient is obtained (step S1415). The numbers in FIG. 20 are filter coefficients (weighting coefficients) at each coefficient position.

  In the case of the LL subband, the directionality is weak, but the correlation between the coefficients in the vertical and horizontal directions is relatively strong. Therefore, in the LL subband, the horizontal centered on the attention coefficient (shaded portion) as shown in FIG. The correction value of the attention coefficient is obtained by applying the direction and vertical smoothing filters (step S1412). The numbers in FIG. 21 are filter coefficients (weighting coefficients) at each coefficient position.

  Note that the smoothing filter shown in FIGS. 18 to 21 is merely an example. A smoothing filter different from the above example can be used as long as the filter coefficient (weighting coefficient) decreases as the neighboring coefficient is farther from the attention coefficient. Further, the image quality can be adjusted by controlling the strength of correction by each neighborhood coefficient by selecting the filter coefficient.

  Step S13 may be the same as that in the first embodiment and its modifications 1_1 to 1_6, and will not be described.

  In this embodiment, in Step S13, as in Modification 1_3 of Embodiment 1, as shown in FIG. 13, the determination of the value (X) of the attention coefficient (X) is 0 (Step S1301) and the subband to which the attention coefficient belongs. It is determined whether or not it is an LL subband (step S1302), coefficients other than the LL subband are excluded from the processing target, and coefficients whose values are not 0 are excluded from the processing target.

  The processing in step S14 is basically the same as in the first embodiment, but is different from the first embodiment in that exception processing is included. Further, since the coefficient of the LL subband is excluded in step S13, the processing for the LL subband coefficient is not included.

  Hereinafter, the process of step S14 is demonstrated along the flowchart shown in FIG. First, the type of subband to which the attention coefficient belongs is determined (step S1421). That is, the coefficient correction unit 14 according to the present embodiment also includes a subband type determination unit.

  When the attention coefficient is a coefficient of the HL subband, 0 determination is performed for the vertical neighboring coefficients a and b of the attention coefficient as shown in FIG. 9 in the HL subband to which the attention coefficient belongs (step S1422). If neither a nor b is 0, the average value of a and b is obtained as a correction value as in the first embodiment (step S1423). However, when at least two values of a and b are 0 (the attention coefficient is also 0), the value of the attention coefficient is directly used as a correction value (step S1424). That is, the attention coefficient is not corrected, and this is one of the exception processes. Such exception processing is performed when the attention coefficient is 0 and the neighborhood coefficient a or b is 0, and if the attention coefficient is corrected to the average value of a and b, the edge of the vertical edge does not appear. This is because there is a high possibility that the side effect of rounding will occur.

  When the attention coefficient is a coefficient of the LH subband, 0 determination is performed on the horizontal vicinity coefficients a and b of the attention coefficient as shown in FIG. 10 in the LH subband to which the attention coefficient belongs (step S1425). If neither a nor b is 0, the average value of a and b is calculated and used as a correction value, as in the first embodiment (step S1426). However, when at least one value of a and b is 0 (the attention coefficient is also 0), the value of the attention coefficient is directly used as a correction value (step S1427). That is, the attention coefficient is not corrected, and this is one of the exception processes. Such exceptional processing is performed when the attention coefficient is 0 and the neighborhood coefficient a or b is 0, and if the attention coefficient is corrected to the average value of a and b, the edge of the horizontal direction edge does not appear. This is because there is a high possibility that the side effect of rounding will occur.

  When the attention coefficient is a coefficient of the HH subband, 0 determination is performed on the neighboring coefficients a, b, c, and d in the diagonal direction of the attention coefficient as shown in FIG. 11 in the HH subband to which the attention coefficient belongs (step S1428). ). If none of a, b, c, and d is 0, the average value of a, b, c, and d is calculated and used as a correction value as in the first embodiment (step S1429). However, when at least one of a, b, c, and d is 0 (the attention coefficient is also 0), the value of the attention coefficient is directly used as a correction value (step S1430). That is, the attention coefficient is not corrected, and this is one of the exception processes. Such exception processing is performed when the attention coefficient is 0, and when any one of the neighborhood coefficients a, b, c, d is 0, the attention coefficient is corrected to the average value of a, b, c, d. This is because there is a high possibility that a side effect that the edge of the edge in the oblique direction is rounded will occur.

  As described above, the coefficient correction unit 14 according to the present embodiment includes a unit for determining the exception process, and suppresses the correction for the coefficient corresponding to the exception process condition, thereby causing a side effect due to an inappropriate coefficient correction. Is to prevent.

  In step S13, it is also possible not to exclude a coefficient whose value is not 0 from the correction target without performing 0 determination of the coefficient of interest. In this case, steps S1422, S1425, and S1428 are changed so that exception processing is performed when the attention coefficient is 0 and at least one of the neighborhood coefficients is 0.

  Also in the present embodiment, as in Modification 1_4 of the first embodiment, the number of quantization steps is determined in step S13, and subband coefficients having a quantization step number smaller than a predetermined value are excluded from the processing target. It may be. Further, similarly to the modified example 1_5 of the first embodiment, in step S13, the hierarchy may be determined, and the coefficient higher than the predetermined hierarchy may be excluded from the processing target. Further, similarly to the modified example 1_6 of the first embodiment, a region may be determined in step S13, and a coefficient that does not correspond to a predetermined image region may be excluded from the processing target.

  In this embodiment, in Step S13, as in Modification 1_3 of Embodiment 1, as shown in FIG. 13, the determination of the attention coefficient (X) is 0 (Step S1301) and the subband to which the attention coefficient belongs is the LL sub It is determined whether or not it is a band (step S1302), coefficients other than the LL subband are excluded from the processing target, and coefficients whose values are not 0 are excluded from the processing target.

  The processing in step S14 is basically the same as in the second embodiment, but is different from the second embodiment in that exception processing is included. Further, since the coefficient of the LL subband is excluded in step S13, the processing for the LL subband coefficient is not included.

  Hereinafter, the process of step S14 is demonstrated along the flowchart shown in FIG. First, the type of subband to which the attention coefficient belongs is determined (step S1441).

  When the attention coefficient is a coefficient of the HL subband, 0 determination is performed for the vertical neighboring coefficients a and b of the attention coefficient as shown in FIG. 9 in the HL subband to which the attention coefficient belongs (step S1442). If neither a nor b is 0, as in the second embodiment, the smoothing filter centered on the attention coefficient as shown in FIG. 18 is applied to the HL subband to which the attention coefficient belongs. A correction value is obtained (step S1443). However, when the value of at least one of the neighborhood coefficients a and b in the vertical direction is 0 (the attention coefficient is also 0), the value of the attention coefficient is directly used as the correction value (step S1444). That is, the attention coefficient is not corrected, and this is one of the exception processes. Such exception processing is performed when the attention coefficient is 0 and the neighborhood coefficient a or b is 0. If the correction is applied to a correction value to which a smoothing filter as shown in FIG. This is because there is a possibility that a side effect that the end of the edge is rounded may occur.

  When the attention coefficient is a coefficient of the LH subband, 0 determination is performed on the horizontal vicinity coefficients a and b of the attention coefficient as shown in FIG. 10 in the LH subband to which the attention coefficient belongs (step S1445). If neither a nor b is 0, as in the second embodiment, a horizontal smoothing filter centered on the attention coefficient as shown in FIG. 19 is applied to the LH subband to which the attention coefficient belongs. A correction value is obtained (step S1446). However, when at least one value of a and b is 0 (the attention coefficient is also 0), the value of the attention coefficient is directly used as a correction value (step S1447). That is, the attention coefficient is not corrected, and this is one of the exception processes. Such exception processing is performed when the attention coefficient is 0 and the neighborhood coefficient a or b is 0. If the correction is applied to a correction value to which a smoothing filter as shown in FIG. This is because there is a high possibility that the end of the corner will be rounded.

  When the attention coefficient is a coefficient of the HH subband, 0 determination is performed on the neighboring coefficients a, b, c, and d in the diagonal direction of the attention coefficient as shown in FIG. 11 in the HH subband to which the attention coefficient belongs (step S1448). ). If any of a, b, c, and d is not 0, as in the second embodiment, in the HH subband to which the attention coefficient belongs, an oblique direction smoothing filter centered on the attention coefficient as shown in FIG. The correction value is obtained by applying (step S1449). However, when at least one value of a, b, c, and d is 0 (the attention coefficient is also 0), the value of the attention coefficient is directly used as a correction value (step S1450). That is, the attention coefficient is not corrected, and this is one of the exception processes. Such exception processing is performed when the attention coefficient is 0 and at least one of the neighborhood coefficients a, b, c, and d is corrected to a correction value to which a smoothing filter as shown in FIG. 20 is applied. This is because there is a high possibility that a side effect that the edge of the edge in the oblique direction is rounded will occur.

  As described above, the coefficient correction unit 14 according to the present embodiment includes a unit for determining the exception process, and suppresses a correction for a coefficient that satisfies the exception process, thereby preventing a side effect due to an inappropriate coefficient correction. It is to prevent.

  In step S13, it is also possible not to exclude a coefficient whose value is not 0 from the correction target without performing 0 determination of the coefficient of interest. In this case, steps S1442, S1445, and S1448 are changed so that exception processing is performed when the attention coefficient is 0 and at least one of the neighborhood coefficients is 0.

  Also in the present embodiment, as in Modification 1_4 of the first embodiment, the number of quantization steps is determined in step S13, and subband coefficients having a quantization step number smaller than a predetermined value are excluded from the processing target. It may be. Further, similarly to the modified example 1_5 of the first embodiment, the hierarchy may be determined in step S13, and the coefficient higher than the predetermined hierarchy may be excluded from the processing target. Further, similarly to the modified example 1_6 of the first embodiment, a region may be determined in step S13, and a coefficient that does not correspond to a predetermined image region may be excluded from the processing target.

It is a block diagram which shows the typical apparatus structure which concerns on this invention. It is a flowchart which shows the typical process flow which concerns on this invention. It is a block diagram for demonstrating embodiment of this invention using a computer. It is a flowchart which shows the typical process flow of a coefficient correction process. It is explanatory drawing of clipping with respect to the correction value of a coefficient. It is explanatory drawing of clipping with respect to the correction value of a coefficient. It is a flowchart which shows the example of step S13. It is a flowchart which shows the example of step S14. It is explanatory drawing of correction value calculation with respect to a HL subband coefficient. It is explanatory drawing of the correction value calculation with respect to a LH subband coefficient. It is explanatory drawing of correction value calculation with respect to a HH subband coefficient. It is explanatory drawing of correction value calculation with respect to LL subband coefficient. It is a flowchart which shows the example of step S13. It is a flowchart which shows the example of step S13. It is a flowchart which shows the example of step S13. It is a flowchart which shows the example of step S13. It is a flowchart which shows the example of step S14. It is a figure which shows the smoothing filter applied to a HL subband coefficient. It is a figure which shows the smoothing filter applied to a LH subband coefficient. It is a figure which shows the smoothing filter applied to a HH subband coefficient. It is a figure which shows the smoothing filter applied to a LL subband coefficient. It is a flowchart which shows the example of step S14. It is a flowchart which shows the example of step S14. It is a block diagram for demonstrating the algorithm of JPEG2000. It is explanatory drawing of a two-dimensional wavelet transform. It is a figure which shows the format of the code stream of JPEG2000. It is a figure which shows the relationship between a tile, a precinct, and a code block.

Explanation of symbols

11 Tag processing means 12 Entropy decoding means 13 Inverse quantization means 14 Coefficient correction means 15 Inverse wavelet transform means 16 Inverse color space conversion / DC level inverse shift means 21 CPU
22 Main memory 23 Auxiliary storage device 24 JPEG2000 compliant decoder

Claims (9)

Inverse quantization and inverse frequency conversion of code data encoded by an encoding method including a frequency conversion process for converting image data into a plurality of subband coefficients and a quantization process for quantizing the subband coefficients. In the image decoding apparatus for decoding
Prior to inverse frequency transformation, the sub-band coefficients after inverse quantization are corrected using the neighborhood coefficient in the direction highly correlated with the coefficient of the sub-band to which the coefficient belongs according to the sub-band to which the coefficient belongs. An image decoding apparatus comprising coefficient correction means for performing the processing.   2. The image decoding device according to claim 1, wherein the coefficient correction means includes means for determining whether or not the value of the coefficient is 0, and excludes a coefficient whose value is not 0 from the correction target. 3. The image decoding device according to claim 1, wherein the coefficient correction means includes means for determining whether or not the number of quantization steps for the subband to which the coefficient belongs is smaller than a predetermined value, wherein the number of quantization steps is less than the predetermined value. An image decoding apparatus, wherein small subband coefficients are excluded from correction targets. 4. The image decoding device according to claim 1, wherein the frequency transformation is a two-dimensional discrete wavelet transformation, and the inverse frequency transformation is a two-dimensional inverse discrete wavelet transformation. 5. 5. The image decoding apparatus according to claim 4, wherein the HL coefficient has an average of the upper and lower HL coefficients of the coefficient to be corrected as a correction value of the coefficient. 5. The image decoding apparatus according to claim 4, wherein the LH coefficient uses an average of left and right LH coefficients of a coefficient to be corrected as a correction value of the coefficient. 5. The image decoding apparatus according to claim 4, wherein the HH coefficient is obtained by using an average of the upper right, upper left, lower right, and lower left HH coefficients of the correction target coefficient as a correction value of the coefficient. A program that causes a computer to function as the coefficient correction means according to claim 1. A computer-readable information recording medium on which the program according to claim 8 is recorded.

Images