JP4789020B2 - Image processing system and image processing program - Google Patents

Description

  The present invention relates to an image processing system and an image processing program, and more particularly to a filter processing with correction suitable for block division coding.

  Conventionally, there has been known a method of compressing / decompressing image data by block division coding, that is, an image is divided into a plurality of blocks, and processing is performed using blocks as processing units. In the typical JPEG (Joint Photographic Experts Group), an image is divided into blocks of a fixed size (8 × 8 pixels), and image data is separated from the spatial domain by DCT (Discrete Cosine Transform) in units of blocks. Converted to the frequency domain. Then, entropy coding is performed on the frequency component obtained by DCT after quantization. Another example of this encoding is recursive AC component predictive encoding (RACP: Recursive ACP) disclosed in Patent Documents 1 and 2. The RACP encoding is a technique for hierarchically compressing / decompressing image data by sequentially subdividing blocks to be processed and using AC component prediction (ACP: AC Componet Prediction). Patent Document 3 discloses RACP encoding using a low-pass filter in order to mitigate block noise and the like caused by block division encoding.

  By the way, as pointed out in Patent Document 4, for example, when using a low-pass filter, not only the noise component superimposed in the encoding process but also the high-frequency component inherent in the original image is removed. End up. Therefore, there is a problem that the image after filtering is blurred as a whole. In order to solve this problem, Patent Document 4 discloses that a deringing filter is used instead of a low-pass filter in block division coding. The deringing filter is a filter that detects noise by comparison with a determination threshold value and replaces the pixel value of the area where the noise is generated with a value calculated using information of the surrounding area. Used for applications such as ringing noise mitigation. Here, the block noise is a block-like distortion in which a boundary between blocks that does not exist in the original image appears, and the ringing noise is a haze-like distortion generated around the contour of the image, that is, the edge.

Japanese Patent No. 4000157 Japanese Patent No. 3774201 Japanese Patent No. 3700976 Japanese Patent Laid-Open No. 2000-102020

  The conventional deringing filter described above has a drawback that the average pixel value of the block, that is, the DC value (DC component) changes before and after filtering. For example, in JPEG, DCT is performed in units of 8 × 8 pixel blocks, but since DCT is a good approximation of KL conversion, information is concentrated on low-frequency components. The DC value, which is the first component of DCT, is closer to the true value than the AC value (alternating current component), which is another component, even if it is quantized. Therefore, if the DC value of the block itself changes due to the filtering process (filtering) using the output pixel value generated including the AC value, the PSNR (Peak signal-to-noise ratio) is lowered. Here, PSNR refers to a ratio (usually dB) between an image signal and mixed noise. The lower this value, the more noise is mixed in the image signal. The same is true for RACP encoding. In the RACP encoding, a DC value is supplied from an upper layer, and an output pixel value is generated together with an AC value. Since the quantization coefficient is smaller in the upper layer, the DC value is closer to the true value than the AC value. Therefore, if the DC value of the block itself is changed by the filtering process using the output pixel value generated including the AC value, the PSNR is lowered.

  The present invention has been made in view of such a point, and an object thereof is to effectively hold a DC value of a block before the filtering process even when the filtering process is performed.

In order to solve such a problem, the first invention provides an image processing system including a filtering unit, a subtracter, and a block correction unit. The filtering unit performs a filtering process using information on a peripheral area located around the area for each area to be filtered. For a block that is a set of a plurality of regions adjacent to each other, the subtracter uses a first DC value that is an average pixel value of the block before the filtering process and a second DC value that is an average pixel value of the block after the filtering process. The difference value subtracted from the value is calculated. The block correction unit divides the difference value by the number of areas in the block and assigns the divided values equally to each area in the block, and subtracts the divided value from the average pixel value in each area in the block after the filter processing. By calculating the average pixel value of each area after correction, the block composed of each area after the filter processing is corrected. Alternatively, the block correction unit apportions the difference value according to the ratio of the average pixel value of each area in the block after the filtering process and variably assigns the difference value to each area in the block. By calculating the average pixel value of each area after correction by subtracting from the average pixel value of each area, the block consisting of each area after the filter processing is corrected.

  Here, in the first invention, the filtering unit preferably performs any one of a deringing filter, an edge enhancement filter, and a smoothing filter as the filter processing.

  In the first invention, the block is a block that is a block division coding processing unit that divides an image into a plurality of blocks and compresses or decompresses image data by processing each block as a processing unit. Is preferred. The first invention may be applied to a plurality of hierarchical structures in which the size of the block gradually decreases as the hierarchy becomes lower, and the block division coding is recursively performed between the hierarchies. In this case, at least one layer includes the filtering unit, the subtractor, and the block correction unit. Then, it is preferable to supply the average pixel value after the fill process corrected on the upper layer side to the lower layer side. Here, the block division encoding may be an encoding process for compressing an image using AC component prediction, or a decoding process for expanding an image using AC component prediction.

The second invention provides an image processing program for causing a computer to execute the following image processing. This image processing includes a first step of performing filtering processing using information on a peripheral region located in the periphery of each region to be filtered, and a set of a plurality of regions adjacent to each other. A second step of calculating an average pixel value of the block before the filtering process relating to the block as the first DC value, and a third step of calculating an average pixel value of the block after the filtering process as the second DC value And a fourth step of calculating a difference value obtained by subtracting the first DC value from the second DC value , and dividing the difference value by the number of areas in the block and equally assigning to each area in the block, the division value by calculating an average pixel value of each region corrected by subtracting from each of the average pixel values of the respective regions in the block after filtering consists of the region after the filtering And a fifth step of correcting the lock.
Alternatively, in the fifth step, the difference value is apportioned in accordance with the ratio of the average pixel value of each region in the block after the filter process, and is variably assigned to each region in the block. A block composed of each area after the filter processing is corrected by calculating the average pixel value of each area after correction by subtracting from the average pixel value of the area after processing.

  Here, in the second invention, the first step preferably performs any one of a deringing filter, an edge enhancement filter, and a smoothing filter as the filter processing.

  In the second invention, the block is a block that is a block division coding processing unit that divides an image into a plurality of blocks and compresses or decompresses image data by processing each block as a processing unit. Is preferred. The second invention may be applied to a hierarchical structure in which the size of the block decreases stepwise as the hierarchy becomes lower, and the block division coding is recursively performed between the hierarchies. In this case, the image processing is performed in at least one hierarchy. Then, the average pixel value after the filter processing corrected by the image processing on the upper layer side is supplied to the lower layer side. Here, the processing of each layer may be an encoding process for compressing an image using AC component prediction or a decoding process for expanding an image using AC component prediction.

  According to the present invention, each value in the block after the filter process is corrected so that the average pixel value of the block after the filter process approaches the first DC value. This correction is performed by distributing the difference between the DC values before and after the filtering process to each region. As a result, the corrected block can effectively maintain the DC value before the filter processing while maintaining the good characteristics after the filter processing. In particular, when applied to block division coding, a decrease in PSNR due to filtering can be suppressed without impairing the filtering effect.

  In the present embodiment, RACP encoding will be described as an example of block division encoding. FIG. 1 is an overall configuration diagram of a RACP encoder. This encoder is mainly composed of a DC calculation unit 1, a DC encoding unit 2, hierarchical processing units 3a to 3c, and filter processing units 4a and 4b, and has a hierarchical structure including four layers. Data TDCn, DCn, DC ″ n (n = 0, 1, 2, 3) output from these units 1, 2, 3a to 3c, 4a, 4b are temporarily stored in a buffer (storage unit) (not shown). And data transfer between units is performed through this buffer.

  The DC calculation unit 1 divides the input image to be compressed into blocks of a preset size, and calculates the average pixel value (DC value) of each block, that is, the average value of the pixels included in the block, from TDC0 to Output as TDC3. Here, TDC0 is an average pixel value of an 8 × 8 pixel block (8 × 8 block), TDC1 is an average pixel value of a 4 × 4 pixel block (4 × 4 block), and TDC2 is a 2 × 2 pixel block ( 2 × 2 block) average pixel value. TDC3 is an average pixel value of a 1 × 1 pixel block (1 × 1 block), that is, a pixel value itself of a pixel which is a minimum unit of an image. The average pixel values TDC0 to TDC3 (true values) calculated directly from the input image are distinguished from the average pixel values DC0 to DC3 (restored values) restored through the processing in the units 2 and 3a to 3c. Is done. Both values are accompanied by irreversible transformation (including quantization) in the process of encoding, so they do not exactly match. Note that in this specification, the “pixel value” includes not only the value of the pixel itself, which is the minimum unit of the image, but also a representative value representing the characteristics of a pixel block that is a set of a plurality of adjacent pixels.

  The DC encoding unit 2 and the hierarchical processing units 3a to 3c include 8 × 8 blocks (top layer), 4 × 4 blocks (second layer), 2 × 2 blocks (third layer), and 1 × 1 blocks. In the four-layer structure composed of (the lowest layer), the layer processing assigned to itself is performed.

The highest layer is configured by the DC encoding unit 2. The D C marks Goka unit 2, to the average pixel value TDC0 for each 8 × 8 block is read from the buffer, differential pulse code modulation (DPCM: Differential Pulse Code Modulation) and subjected to entropy coding. Difference pulse code modulation encodes the difference between the average pixel values TDC0 for blocks adjacent to each other on the image plane. Then, entropy coding such as Huffman coding or arithmetic coding is performed on the coded difference after quantization. The data that has undergone such processing is output as a DC component DC0 of an image that is a part of the compressed data, and an average pixel value of 8 × 8 blocks is supplied to the hierarchical processing unit 3a of the immediately lower hierarchy. A restored (inverse quantized) value DC0 is output.

  The three hierarchical processing units 3a to 3c include an average pixel value TDCn (n = 1, 2, 3, and so on) generated by the DC calculation unit 1 and an average pixel value DCn−1 generated by the immediately higher hierarchy. (Or DC ″ n−1) as an input, and an AC component ACn of the image as a part of the compressed data is output by AC component predictive coding. As is well known, the general process flow of AC component predictive coding includes prediction pixel value calculation, prediction residual calculation, lossy conversion of prediction residuals (Hadamard transform, quantization, etc.), and entropy coding. (Huffman coding or the like), and the quantization coefficient is smaller in the upper layer. In addition, the hierarchical processing units 3a to 3c (3c is only when necessary) performs inverse transformation of irreversible transformation to restore a prediction residual, and adds this to the prediction pixel value, thereby obtaining an average pixel value. Restore DCn. The restored average pixel value DCn is output for supply to the immediately lower hierarchy. Note that in the AC component prediction of a certain hierarchy, when the sub-block of that hierarchy is referred to as a peripheral region, the restored average pixel value DCn is fed back to the same hierarchy.

  The second hierarchy is composed of a hierarchy processing unit 3a and a filter processing unit 4a. In this hierarchy, the AC component AC1 of 4 × 4 blocks and the average pixel values DC1 and DC ″ 1 of 4 × 4 blocks are output by processing for 8 × 8 blocks. The hierarchy processing unit 3a uses the average pixel value DC0 of the 8 × 8 block generated in the highest hierarchy as a peripheral area to be referred to in the AC component prediction, and when referring to the sub-blocks together, the same hierarchy The average pixel value DC1 of the 4 × 4 block fed back in step 1 is also used. Further, the average pixel value TDC1 of 4 × 4 blocks calculated by the DC calculation unit 1 is used for calculating the prediction residual. The filter processing unit 4a performs a filter process with correction on the average pixel value DC1 in order to reduce block noise and ringing noise, and outputs an average pixel value DC ″ 1.

  The third hierarchy is composed of a hierarchy processing unit 3b and a filter processing unit 4b. In this hierarchy, the 2 × 2 block AC component AC2 and the 2 × 2 block average pixel values DC2 and DC ″ 2 are output by the processing targeted for the 4 × 4 block. The hierarchical processing unit 3b uses the average pixel value DC ″ 1 (after filtering) of the 4 × 4 block generated in the second hierarchical level as a peripheral region to be referred to in the AC component prediction, and combines the sub blocks. For reference, an average pixel value DC2 of 2 × 2 blocks fed back in the same hierarchy is also used. Further, the average pixel value TDC2 of 2 × 2 blocks calculated by the DC calculation unit 1 is used for calculating the prediction residual. The filter processing unit 4b performs a filter process with correction on the average pixel value DC to reduce block noise and the like, and outputs an average pixel value DC ″ 2.

  The lowest hierarchy is configured by the hierarchy processing unit 3c. In this hierarchy, the AC component AC3 of 1 × 1 block is output by the processing targeted for 2 × 2 blocks. The hierarchy processing unit 3c uses the average pixel value DC ″ 2 (after filtering) of the 2 × 2 block generated in the third hierarchy as information to be referred to in the AC component prediction, and also refers to the sub blocks. In this case, the average pixel value DC3 of 1 × 1 block fed back in the same hierarchy is also used. Further, the average pixel value TDC3 of the 1 × 1 block calculated by the DC calculation unit 1 is used for calculating the prediction residual.

  In such a hierarchical structure, the DC encoding unit 2 and the hierarchical processing units 3a to 3c are linked to each other and perform pipeline parallel processing, whereby a series of processing is recursively executed between the hierarchical levels. As a result, the DC component DC0 of the image and the AC components AC1 to AC3 are output as compressed data. In addition to these, the compressed data includes accompanying information such as a Huffman table. Note that the filter processing units 4a and 4b do not need to be provided in a plurality of layers, but may be provided in at least one layer.

  FIG. 2 is an explanatory diagram of blocks set on the image plane. A plurality of blocks are set on the image plane by dividing an image (the whole image of one frame or a part thereof) vertically and horizontally. Block division coding (AC component predictive coding) in each layer is performed in units of blocks. The block size is set to decrease step by step as the hierarchy becomes lower, and the size of the block to be processed in a certain hierarchy is the size of the sub-block subdivided in the immediately higher hierarchy Matches.

  Processing of the entire image in a certain hierarchy is achieved by repeating the processing while sequentially shifting the blocks to be processed on the screen and processing all the blocks in the image. As shown in the figure, the direction of this shift may be a line-sequential scanning along the horizontal direction, but may be arbitrarily set including that along the vertical direction. Further, the shift directions in the respective hierarchies are not necessarily the same. Here, a region located around the processing target is referred to as a “peripheral region”, and information on the peripheral region is referred to when the AC component prediction of the processing target is performed. Peripheral areas are classified into areas that have not yet undergone sequential shift processing, ie, “unprocessed peripheral areas”, and areas that have already undergone sequential shift processing, ie, “processed peripheral areas”. In the case of the figure, if the processing target is a block S, adjacent blocks T, B, L, and R on the top, bottom, left and right are peripheral regions. The blocks B and R adjacent to the lower right are unprocessed peripheral areas, and the blocks T and L adjacent to the upper left are processed peripheral areas. In the unprocessed peripheral area (blocks B and R), the individual average pixel values DCn of the sub-blocks obtained by dividing the block into four have not yet been calculated, and the average pixel values of the entire block calculated in the immediately higher hierarchy Only DCn-1 is known. Further, in the processing-periphered side areas (blocks T and L), the average pixel value DCn of each sub-block obtained by subdividing the block has been calculated by the previous process in the same hierarchy. Hereinafter, regarding four sub-blocks obtained by subdividing a block, the upper left of the block is identified by subscripts “00”, the upper right is “01”, the lower right is “11”, and the lower left is “10”. In the present embodiment, the upper, lower, left, and right blocks T, B, L, and R are set as the peripheral area to be processed, but this is an example, and a part thereof (for example, upper and left and right) It may also be included other than that (for example, diagonal direction, one jump, etc.).

  Next, the AC component predictive encoding executed by the hierarchical processing units 3a to 3c will be schematically described with reference to FIG. In the present predictive coding, (1) a predicted image S ′ of the processing target S is generated by referring to information on the peripheral regions T, B, L, and R, and (2) the predicted image S ′, the original image, Is calculated as a prediction residual PR, and (3) irreversible transformation (Hadamard transform, quantization, etc.) of the prediction residual PR is performed, and then entropy coding (Huffman coding, etc.) is performed. In this encoding, the point in increasing the compression rate is to reduce the prediction residual PR, in other words, to improve the accuracy of the predicted image S ′. The closer the predicted image S ′ (a set of adjacent predicted pixel values) generated by the AC component prediction is to the original image (a set of adjacent true pixel values), the smaller the prediction residual PR. The smaller the prediction residual PR, the more likely the appearance frequency of the prediction residual PR statistically tends to be close to 0, which is advantageous in performing entropy coding.

  The AC component prediction uses the spatial correlation between the processing target S and the surrounding regions T, B, L, and R to predict the predicted image S ′, that is, the regions s00 to s11 obtained by subdividing the processing target S. The predicted pixel value (average pixel value in the region) is calculated. There are various specific calculation methods, and any method including the following examples may be adopted.

(Calculation method 1)
Using the average pixel value DCn-1 of the blocks T, B, L, and R (calculated at the immediately higher hierarchy) as a peripheral region to be referred to in the AC component prediction, the predicted pixel value s00 to s00 to s11 is calculated. In addition, regarding Expression 1, the notation of S, T, B, L, and R does not indicate the block itself but indicates the average pixel value DCn-1 (the same applies to Expression 2 described later). The details of this method are disclosed in Japanese Patent No. 3774201 cited as Patent Document 2, and should be referred to if necessary.

(Formula 1)
s00 = S + (T−B + LR) / 8
s01 = S + (T−B−L + R) / 8
s10 = S + (-T + B + LR) / 8
s11 = S + (− T + B−L + R) / 8

(Calculation method 2)
As peripheral regions to be referred to in the AC component prediction, the average pixel value DCn-1 of blocks B and R (calculated in the immediately higher hierarchy) and sub-blocks t10, t11, l01, and l11 adjacent to the processing target S The predicted pixel values s00 to s11 are calculated by the following formula 2 using the average pixel value DCn (calculated in the previous process of the same hierarchy). In addition, regarding Expression 2, the notation of t10, t11, l01, and l11 does not indicate the sub-block itself, but indicates the average pixel value DCn thereof. In general, the sub-blocks t10 and t11 that are directly adjacent to the processing target S are spatially separated from the processing target s compared to the block T that also includes the sub-blocks t00 and t01 that are not directly adjacent to the processing target S. It can be said that the correlation is higher. Therefore, by using only the sub-blocks t10 and t11 excluding the redundant sub-blocks t00 and t01, further improvement in prediction accuracy can be expected. Note that details of this method are disclosed in Japanese Patent No. 4000157 cited as Patent Document 1, and should be referred to if necessary.

(Formula 2)
s00 = S + ((t10 + t11) / 2-B + (l01 + l11) / 2-R) / 8
s01 = S + ((t10 + t11) / 2-B- (l01 + l11) / 2 + R) / 8
s10 = S + (-(t10 + t11) / 2 + B + (l01 + l11) / 2-R) / 8
s11 = S + (-(t10 + t11) / 2 + B- (l01 + l11) / 2 + R) / 8

  Note that the average pixel value DCn supplied to the immediately lower hierarchy is calculated by adding a prediction residual to the predicted image S ′. As the value of the prediction residual, the above-described prediction residual PR itself is used. Rather, a restored value PR ′ (slightly different from PR) obtained by performing inverse conversion after irreversible conversion is used. The reason for using the restoration value PR ′ is to prevent the accumulation of error Δ (Δ = PR−PR ′) due to repetitive processing at the time of decoding and to ensure consistency with the decoding processing.

  FIG. 4 is a configuration diagram of the filter processing units 4a and 4b. The filter processing units 4a and 4b in the respective layers have the same basic configuration and the same process flow except that they handle different block sizes. Hereinafter, the filter processing unit 4a in the second hierarchy will be described as an example. The filter processing unit 4a includes a pre-processing DC value calculation unit 41, a filtering unit 42, a post-processing DC value calculation unit 43, a subtractor 44, and a block correction unit 45.

  The pre-processing DC value calculation unit 41 calculates an average pixel value (DC value) before filtering of blocks processed by block division coding as DC0. In the second hierarchy, an average pixel value DC1 in units of 4 × 4 blocks is calculated from an average pixel value DC0 in units of 8 × 8 blocks. The input block is a set of four regions that are adjacent in position, and the average of DC1 (four values) of these regions is DC0. Note that DC0 is equivalent to the average pixel value DC0 (8 × 8 blocks) calculated in the immediately higher layer (DC encoding unit 2) in relation to the configuration of FIG. Therefore, instead of performing the calculation here, the average pixel value DC0 calculated in the immediately higher hierarchy may be read from the buffer and used.

  The filtering unit 42 is located in the vicinity of each region to be filtered, specifically, each region obtained by subdividing the input block into four (the value of each region is DC1). Filter processing using information on the peripheral area is performed individually. In this embodiment, as the filtering process, a deringing filter is used in which noise is detected by comparison with a determination threshold value, and a pixel value in a region where the noise is generated is replaced with a calculated value using information on the surrounding region. .

  FIG. 5 is an explanatory diagram of the operation of the deringing filter. In order to process one DC value g0 (= DC1) in the input block, eight DC values g1 to g8 (= DC1) located in the vicinity thereof are used as a peripheral region. The subtractor 42a individually calculates a difference between each DC value g1 to g8 that is information on the peripheral region and a DC value g0 that is information on the processing target S, and outputs a difference value Δ1 to Δ8 (absolute value). . The threshold value determination unit 42b compares each of the difference values Δ1 to Δ8 with a predetermined threshold value Th to determine the presence or absence of noise. Difference values Δ1, Δ2, Δ5, Δ7, and Δ8 that are smaller than the threshold value Th are considered to be due to noise superimposed in the encoding process, while those that are equal to or greater than the threshold value Th are the differences. This is considered to be due to contour components that are inherently present in the image. The adder 42c adds the difference values Δ1, Δ2, Δ5, Δ7, Δ8 caused by noise and the value obtained by multiplying the DC value g0 of the region to be processed by the coefficient λ. Then, a value g′5 obtained by dividing the addition result by the coefficient (λ + n) is output as a filtered value DC′1. Here, n is the number of outputs of the threshold value determination unit 42b, that is, the number of DC values g smaller than the threshold value Th. In the case of FIG. The above processing is performed on all four areas constituting the input block, and thereby a processed block consisting of four values DC′1 is obtained.

  FIG. 6 is a diagram illustrating changes in values on the scan line before and after the filter processing. Black circles in the figure indicate regions on the scan line, and the vertical direction indicates changes in the DC value g in each region. As shown in FIG. 5A, when the difference value | Δ | is equal to or greater than the threshold value Th, the DC value g does not change before and after filtering, and the outline of the original image is maintained as it is. On the other hand, as shown in FIG. 6B, when the difference value | Δ | is smaller than the threshold value Th, the output is obtained by removing the step by filtering, and block noise and ringing noise are effectively obtained. Can be removed.

The post-processing DC value calculation unit 43 calculates the average pixel value of the processed block as DC′0. Then, the subtractor 44 calculates a difference value α (= DC′0−DC0) between the average pixel value DC0 before filtering and the average pixel value DC′0 after filtering. When DC′0 (after filtering) is larger than D C0 (before filtering), that is, when the DC value is increased by filtering, the sign of the difference value α becomes positive. On the other hand, when DC′0 (after filtering) is smaller than D C0 (before filtering), that is, when the DC value is reduced by the filtering process, the sign of the difference value α is negative.

  The block correction unit 45 corrects the processed block by distributing the difference value α to each region in the block so that the average pixel value of the processed block approaches DC0 (before filtering). For example, the difference value α is divided by the number of areas 4 in the block. This division value corresponds to a correction amount per area. Then, the division value (α / 4) is evenly assigned to each area in the block according to the following formula 3, and the value obtained by subtracting the division value (α / 4) from the value DC′1 of each area is corrected. The value of DC ″ 1.

(Formula 3)
DC ″ 1 = DC′1−α / 4

  When the DC value is increased by the filtering process, since the sign of the difference value α is positive, (α / 4) is subtracted from the value DC′1 before correction. As a result, the difference value α corresponding to the entire output block is subtracted, and the average pixel value is equivalent to DC0 (before filtering). On the other hand, when the DC value is reduced by the filtering process, since the sign of the difference value α is negative, (α / 4) is added from the value DC′1 before correction. As a result, the difference value α is added to the entire output block, and the average pixel value is equivalent to DC0 (before filtering). As a result, the output block consisting of the four values DC ″ 1 can be said to maintain good characteristics after filtering, although the values in the region individually change, when viewed as a whole block.

  As another distribution method of the difference value α, the difference value α may be variably assigned to each region according to the ratio of the value DC′1 of each region in the processed block. In this case, the difference value α is apportioned according to the ratio of log 2 of the value DC′1 after filtering. For example, when the pixel values in the block are 100, 100, 150, and 200, the difference value α is apportioned by 2: 2: 3: 4. Then, each apportioned value is assigned to each area in the block and subtracted from each DC′1.

  In the two examples described above, the average pixel value of the output block is set to DC0 by correction (however, there is a rounding error due to division, so it is not strictly DC0). Although matching with DC0 is the most preferred embodiment, the present invention is not limited to this, and it is sufficient that at least DC0 is approached. This is because as it approaches DC0, it contributes to the suppression of the decrease in PSNR. Through the above processing, the filtered value DC ″ 1 corrected by the block correction unit 45 on the upper layer side is supplied to the lower layer side.

  The function equivalent to the filter processing unit 4 shown in FIG. 4 can also be realized by software processing by a computer. FIG. 7 is a flowchart of a computer program that realizes the above-described filter processing with correction. Since there is no essential difference between hardware processing and software processing, only a brief description will be given here, and the details will be referred to the above description. First, in step 1, the average pixel value before filtering the block is calculated as DC0. In step 2, a filtering process (deringing filter) using information on a peripheral area located in the vicinity of the target area to be filtered is performed. In step 3, the average pixel value after the filter processing of the block is calculated as DC′0. In step 4, a difference value α between the value DC 0 calculated in step 1 and the value DC ′ 0 calculated in step 3 is calculated. Finally, in step 5, the difference value α is distributed to each area in the block so that the average pixel value after filtering of the block approaches DC0, thereby correcting the block after filtering. .

  As described above, according to this embodiment, the filter processing units 4a and 4b are added to the conventional RACP encoder, and the filter processing is performed on the blocks processed by the block division coding in each layer. Then, the value of each area in the block is corrected so that the DC value of the block after the filtering process approaches the value DC0 before the filtering process. This correction is performed by distributing the DC value difference α before and after the filtering process to each region. As a result, the corrected block maintains good characteristics after filtering. In RACP encoding, a DC value is supplied from an upper layer and an output pixel value is generated together with an AC value. Since the quantization coefficient is smaller in the upper layer, the DC value is closer to the true value than the AC value. According to the present embodiment, DC0 before the filter process can be effectively held by correcting the block after the filter process. As a result, it is possible to effectively suppress a decrease in PSNR caused by filtering without impairing the filtering effect.

  In the above-described embodiment, a deringing filter is cited as an example of the filtering process. However, a filter that uses information on a peripheral region located around the target region, including an edge enhancement filter and a smoothing filter, is used. May be used. As the edge enhancement filter, for example, a Laplacian filter as shown in FIG. 8 can be used. As the smoothing filter, for example, a simple average filter as shown in FIG. 9 or a Gaussian filter as shown in FIG. 10 can be used. In applying these filters, the center of the filter is placed so as to be located in the region to be processed, the pixel value corresponding to the position is multiplied by the filter element individually, and the sum of these is multiplied by the target region. Value.

  In the above-described embodiment, an example in which the filter processing unit 4 is added to an encoder that compresses an image using AC component prediction has been described. However, this is added to a decoder that expands an image using AC component prediction. Also good.

  The present invention is not limited to AC component predictive coding, but can be widely applied to block division coding based on block unit processing, including JPEG and the like. Furthermore, the present invention can also be applied to image processing such as generating a high-resolution image (high-definition image) while subdividing blocks.

Overall configuration of RACP encoder Illustration of blocks set on the image plane Illustration of AC component predictive coding Configuration diagram of the filter processing unit Operation diagram of deringing filter The figure which shows the change of each value on the scan line before and after the processing of the filter Flowchart of computer program that realizes filter processing with correction Illustration of Laplacian filter Illustration of simple average filter Illustration of Gaussian filter

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 DC calculation part 2 DC encoding part 3a-3c Hierarchical process part 4a, 4b Filter process part 41 Pre-process DC value calculation part 42 Filtering part 42a Subtractor 42b Threshold value determination part 42c Adder 43 Post-process DC value calculation part 44 Subtractor 45 Block correction unit

Claims (12)

In an image processing system,
A filtering unit that performs a filtering process using information on a peripheral area located around the area for each of the areas to be filtered;
For a block that is a set of a plurality of regions adjacent to each other, a first DC value that is an average pixel value of the block before the filtering process is used as a second DC value that is an average pixel value of the block after the filtering process. A subtractor for calculating a difference value subtracted from
The difference value is divided by the number of areas in the block and assigned to each area in the block equally, and the divided value is subtracted from the average pixel value of each area in the block after filtering. An image processing system comprising: a block correction unit that corrects the block including each of the regions after the filter processing by calculating an average pixel value of each of the regions after correction. In an image processing system,
A filtering unit that performs a filtering process using information on a peripheral area located around the area for each of the areas to be filtered;
For a block that is a set of a plurality of regions adjacent to each other, a first DC value that is an average pixel value of the block before the filtering process is used as a second DC value that is an average pixel value of the block after the filtering process. A subtractor for calculating a difference value subtracted from
The difference value is apportioned in accordance with the ratio of the average pixel value of each region in the block after the filtering process and is variably assigned to each region in the block. A block correction unit that corrects the block composed of each of the regions after filter processing by calculating an average pixel value of each of the regions after correction by subtracting from the average pixel value of the region;
An image processing system comprising: The filtering unit, wherein the filtering, de-ringing filtering, image processing system according to claim 1 or 2, characterized in that to do any of the edge emphasis filter and smoothing filter. The block is a block which is a block division coding processing unit for compressing or expanding image data by dividing an image into a plurality of blocks and processing each block as a processing unit. The image processing system described in any one of 1 to 3 . In the hierarchical structure in which the size of the block is gradually reduced as the hierarchy becomes lower, and the block division coding is performed recursively between the hierarchies.
At least one layer includes the filtering unit, the subtracter, and the block correction unit,
5. The image processing system according to claim 4 , wherein the average pixel value after filter processing corrected on the upper layer side is supplied to the lower layer side. 6. The image processing according to claim 5 , wherein the block division encoding is an encoding process for compressing an image using AC component prediction or a decoding process for expanding an image using AC component prediction. system. In an image processing program,
A first step of applying a filtering process using information on a peripheral area located around the area to each of the areas to be filtered;
A second step of calculating, as a first DC value, an average pixel value of the block before the filtering process related to a block that is a set of a plurality of regions adjacent to each other;
A third step of calculating an average pixel value after filtering related to the block as a second DC value;
A fourth step of calculating a difference value by subtracting said first DC value from said second DC values,
The difference value is divided by the number of areas in the block and assigned to each area in the block equally, and the divided value is subtracted from the average pixel value of each area in the block after filtering. An image processing program for causing a computer to perform image processing including calculating a mean pixel value of each of the regions after correction and correcting the block including the regions after filter processing. . In an image processing program,
A first step of applying a filtering process using information on a peripheral area located around the area to each of the areas to be filtered;
A second step of calculating, as a first DC value, an average pixel value of the block before the filtering process related to a block that is a set of a plurality of regions adjacent to each other;
A third step of calculating an average pixel value after filtering related to the block as a second DC value;
A fourth step of calculating a difference value obtained by subtracting the first DC value from the second DC value;
The difference value is apportioned in accordance with the ratio of the average pixel value of each region in the block after the filtering process and is variably assigned to each region in the block. A fifth step of correcting the block consisting of each of the areas after filtering by calculating an average pixel value of each of the areas after correction by subtracting from the average pixel value of each area;
An image processing program for causing a computer to execute image processing having The image processing program according to claim 7, wherein the first step performs any one of a deringing filter, an edge enhancement filter, and a smoothing filter as the filter processing. The block is a block which is a block division coding processing unit for compressing or expanding image data by dividing an image into a plurality of blocks and processing each block as a processing unit. The image processing program according to any one of 7 to 9 . In the hierarchical structure in which the size of the block is gradually reduced as the hierarchy becomes lower, and the block division coding is performed recursively between the hierarchies.
The image processing is performed in at least one hierarchy,
11. The image processing program according to claim 10 , wherein the average pixel value after the filter processing corrected on the upper layer side is supplied to the lower layer side. 12. The image processing according to claim 11 , wherein the block division coding is an encoding process for compressing an image using AC component prediction or a decoding process for expanding an image using AC component prediction. program.

Images