JP2013042385A - Image encoder and image decoder - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an encoder having a high coding efficiency, and to provide a corresponding decoder.SOLUTION: A unit block is split by adaptive split means 20 into small regions of respective split seeds, prediction means 11 performs prediction of each small region and compensation means 12 compensates each small region and then encodes each small region via prediction residual. A split seed of lowest coding cost is selected among the respective split seeds, and the unit block is encoded. Various types of small region in each split seed including non-square region are prepared, and optimum prediction information is determined for each small region thus enhancing the coding efficiency. Similar processing in units of small region is also performed in a decoder.

Description

  The present invention relates to an image encoding device and an image decoding device, and more particularly to an image encoding device and an image decoding device that improve encoding efficiency by adaptively changing the division of a unit block.

  In conventional image coding, there is a method of processing by using a local correlation by dividing an image into a plurality of blocks. Several methods have been proposed for dividing the block, and the coding efficiency is improved. For example, by using a large block in a flat area, not only can the correlation be utilized to the maximum, but also the control unit can be suppressed because the processing unit becomes large. On the other hand, by using a small block in the edge region, it is possible to efficiently perform frequency conversion while maintaining a relatively high correlation even in a pixel rich in change.

  In H.264 shown in Non-Patent Document 1, the block used for intra-picture coding can selectively use the square size of 16 × 16, 8 × 8, or 4 × 4, so the pixel value is changed Can be dealt with by dividing into smaller square sizes.

  In Patent Document 1, a non-square small area is prepared in H.264 block division. Encoding efficiency is improved by limiting the intra prediction mode according to the shape of the small area.

JP 2010-171729 A

Kakuno et al., “H.264 / AVC Textbook Impress Standard Textbook Series”, Impress Net Business Company, 2004

  Since Non-Patent Document 1 provides only three sizes of squares for a certain block to be encoded, even if the majority of the block is a flat region, if there is a change in part, it is flat. There is a problem that there is no choice but to divide it into small square sizes including the area. In addition, H.264 processes prediction errors, not pixels, so there is a problem that large changes are likely to occur. For example, H.264 Intra prediction generates a prediction value of a target block with reference to encoded neighboring pixels, so that there is a problem that a larger prediction error occurs as the distance from the prediction reference pixel increases. . Therefore, Non-Patent Document 1 has a problem that the encoding efficiency cannot be sufficiently increased.

  On the other hand, since Patent Document 1 introduces a non-square small region, the problem of Non-Patent Document 1 can be partially solved. However, in order not to output information on the target block division method and shape, frequency conversion method and quantization information, a restriction is imposed on the small region intra prediction mode. When pixel values are flowing due to camera work such as panning and tilting, the intra prediction mode constrained in Patent Document 1 may be optimal even if the pixel value is far from the reference pixel, so that the encoding efficiency is sufficiently increased. The problem of being unable to do so remains. In addition, since the prediction signal generation related information also serves as the designation of the small area dividing method, there is a problem that the Intra prediction mode cannot be individually designated when the square and non-square are mixed in the small area.

  An object of the present invention is to solve the above-described problems of the prior art and to provide an image encoding device with high encoding efficiency.

  Another object of the present invention is to provide an image decoding apparatus corresponding to the image encoding apparatus with high encoding efficiency.

  To achieve the above object, the present invention is based on prediction means for determining prediction information for predicting each pixel from an encoded pixel for each pixel of a unit block composed of a plurality of pixels, and the prediction information. And a compensation means for generating a prediction pixel of each pixel, and performing orthogonal transformation and quantization on the prediction residual obtained by performing the difference processing between each pixel and the prediction pixel to perform quantization In an image coding apparatus that performs coding for each unit block, the image processing apparatus includes adaptive division means that selects and divides a division type from a set of division types that divides a unit block into a plurality of small regions. The unit block is configured such that the prediction unit sequentially determines prediction information and the compensation unit generates a prediction pixel for each pixel constituting each of the small regions obtained by the division in the selected division type. Every In performing No., resolution type of information selected for each unit block quantization value and the prediction information for each and small regions are encoded, characterized in that it is encoded.

  The present invention is also an image decoding apparatus for the image encoding apparatus, wherein the decoding means decodes the information of the division type selected for each encoded unit block, and the quantization value and prediction information for each small area. And inverse quantization means for inversely quantizing the decoded quantized value to obtain a decoded transform coefficient; and inverse transform means for inversely transforming the decoded transform coefficient to obtain a decoded prediction residual; Based on the decoded division type information, decoding-side adaptive division means for resetting a small area determined by the division type to a unit block to be decoded, decoded pixels, and the decoded prediction information Compensating means for generating a decoded prediction pixel for each reset small area, and for each of the reset small areas, the decoded prediction residual and the decoded prediction pixel Addition to obtain decoded pixels The features.

  According to the image coding apparatus of the present invention, since the unit block is divided into a plurality of small areas and is predicted and coded in units of the small areas, the coding efficiency can be improved.

  Further, according to the image decoding device of the present invention, it is possible to perform decoding corresponding to the encoding in the image encoding device by decoding each small region.

It is a functional block diagram of an image coding apparatus. It is a functional block diagram of an image decoding apparatus. It is a figure which shows nine types of prediction modes in the prediction in a screen in the case of the unit block 4x4 in well-known H.264. It is a figure explaining performing prediction by limiting or expanding and applying the prediction of the conventional unit block object to the small area in the present invention. It is a figure which shows the 1st example of the set of a division | segmentation seed | species. It is a figure which shows the 2nd example of the set of a division | segmentation seed | species. It is a figure which shows the 3rd example of the set of a division | segmentation seed | species. It is a figure which shows the example explaining the summary description of the division | segmentation kind in FIG. It is a figure which shows the flow which calculates | requires the encoding cost in all the division | segmentation seed | species for every unit block which comprises an input image, finally determines the division | segmentation kind | species to apply, and encodes. It is a figure explaining that the order which encodes to the small area | region in each division | segmentation kind is predetermined. FIG. 10 is a flowchart illustrating the order in which code information actually received by the code output means is generated by encoding in the final determined division type according to the flow of FIG. 9. It is a functional block diagram of the detail of an adaptive division | segmentation means. It is a functional block diagram of the detail of a compensation means. It is a figure for demonstrating storage of the division | segmentation seed | species and prediction information by the encoded channel information storage part, and its application. It is a figure explaining re-assignment of the identification number of prediction mode with respect to each shape contained in the set of small fields. It is a figure which illustrates notionally the repetitive compensation by a pixel unit repetitive compensation part. It is a figure explaining the example which calculates | requires a prediction pixel by performing iterative compensation by applying the 4 * 4 prediction mode 0 (vertical prediction) to a small area | region by the pixel unit iterative compensation part. It is a figure explaining the example which calculates | requires a prediction pixel by performing iterative compensation by applying the 4 * 4 prediction mode 4 to a small area | region by a pixel unit iterative compensation part. It is a flowchart of the decoding in an image decoding apparatus.

  Functional block diagrams of an image encoding device and an image decoding device according to an embodiment of the present invention are shown in FIGS. 1 and 2, respectively.

  The image encoding device 1 according to the present invention provides a prediction residual obtained by performing a difference process on each pixel of a unit block composed of a plurality of pixels with each pixel predicted from an encoded pixel. A configuration in which an adaptive division function capable of efficiently dividing the unit block into a plurality of small regions is added to an image coding apparatus that performs orthogonal transform, quantization, and coding on the difference and performs coding for each unit block It is characterized by.

  That is, as shown in FIG. 1, the image encoding device 1 according to the present invention calculates image input means 10 that takes in an input pixel as an encoding target, and prediction information for predicting an input pixel from an encoded pixel. A prediction unit 11; a compensation unit 12 that generates a prediction pixel from the prediction information and the encoded pixel; a difference unit 13 that subtracts the prediction pixel from the input pixel and outputs a prediction residual; and converts the prediction residual into a frequency domain. Conversion means 14 for transform coefficients, quantization means 15 for quantizing the transform coefficients, encoding means 16 for variable-length coding the quantized transform coefficients and others, and outputting the code information The code output means 100, the inverse quantization means 17 that inversely quantizes the quantized transform coefficient, the inverse transform means 18 that inversely transforms the inversely quantized transform coefficient, and the prediction that is inversely transformed with the prediction pixel With the residual A unit block for encoding is further added to an image encoding device including an adder 19 that reconstructs the calculated and encoded pixels and an accumulating unit 21 that accumulates the reconstructed encoded pixels. An adaptive dividing means 20 for dividing the block into a plurality of small areas for performing sequential encoding within the block is provided.

  Note that the pixels stored in the storage unit 21 are pixels that are sequentially processed by the inverse quantization unit 17 through the adder 19 from the quantization unit 15 and stored in the storage unit 21 as a signal flow. At this time, an image input that has not yet been encoded in this way in the sense that it is input from the quantization means 15 to the inverse quantization means 17 and also input to the encoding means 16 and encoded. This is distinguished from the pixel immediately after being sent from the means 10 and is called an encoded pixel.

  On the other hand, the image decoding apparatus 3 corresponding to the image encoding apparatus 1 according to the present invention, as shown in FIG. 2, has code input means 30 to which code information from the code output means 100 is input, and variable with respect to the code information. Decoding means 31 that performs long decoding to obtain quantized values and prediction information, etc., inverse quantization means 37 that inversely quantizes the transform coefficients that have been quantized, and inverse transforms the inversely quantized transform coefficients into pixel regions An inverse transform means 38, a compensation means 32 for generating a prediction pixel from the prediction information and the decoded pixel, an adder 39 for adding the prediction pixel and the reconstructed prediction residual to reconstruct a decoded pixel, and decoding For an image decoding apparatus including an image output means 300 for outputting pixels and an accumulation means 41 for accumulating reconstructed decoded pixels (decoded pixels), unit blocks for decoding are further sequentially included in the blocks. Decrypt And it is configured to include an adaptive division means 40 for dividing the small regions of the eye.

  The outline of each means is as follows. Here, FIG. 2 corresponding to the functional block or signal of FIG. 1 is described in order to collectively explain the portion where the functional flow in FIG. 1 and FIG. A description such as “adder 19 (adder 39)” or “encoded pixel (decoded pixel)” in which the functional blocks or signals are enclosed in parentheses is used.

  The subtracter 13 receives the input pixel of the encoding target region sent from the image input unit 10 and the predicted pixel predicted from the encoded pixel sent from the compensation unit 12 using the prediction information of the prediction unit 11. Is done. For the input, the differentiator 13 calculates the difference between the input pixel and the prediction pixel as a prediction residual. The prediction residual is sent to the conversion means 14.

  The adder 19 (adder 39) receives the prediction residual sent from the inverse transform unit 18 (inverse transform unit 38) and the encoded pixel (decoded pixel) sent from the compensation unit 12 (compensation unit 32). The predicted pixel that has been predicted is input. The adder 19 (adder 39) calculates the sum of the prediction residual and the prediction pixel as an encoded pixel (decoded pixel) for the input.

  The calculated encoded pixels (decoded pixels) are accumulated in the accumulation means 21 (accumulation means 41), and in the coding configuration (FIG. 1), the prediction means 11, the compensation means 12, and the adaptive division means 20 are stored. On the other hand, in the decoding configuration (FIG. 2), it is output to the image output means 300 and also received from the compensation means 32 via the storage means 41.

  The transforming unit 14 transforms the prediction residual input from the differentiator 13 into the frequency domain by orthogonal transform, and outputs the transform coefficient obtained by the orthogonal transform to the quantizing unit 15. As the orthogonal transform, DCT to DCT approximate transform or DWT can be used.

  The quantization unit 15 quantizes the transform coefficient input from the transform unit 14. The quantized value obtained by the quantization is output to the encoding means 16 and the inverse quantization means 17. The quantization parameter used for the quantization process can be set as a combination of constant values. Alternatively, the output bit rate can be kept constant by controlling according to the information amount of the transform coefficient.

  The encoding unit 16 includes a quantized value sent from the quantizing unit 15, prediction information sent from the prediction unit 11 (illustrated by a dotted arrow), and division type information (dotted line) sent from the adaptive dividing unit 20, which will be described later. Is encoded) and output as code information. For the encoding, a variable length code or an arithmetic code that removes redundancy between codes can be used. In encoding the prediction information, different codes can be assigned according to the division type information as described later.

  Here, each picture (frame) of the input image is encoded by the encoding means 16 for each small region determined by the division by the division type set by the adaptive division means 20 described later. The input image may be a moving image composed of a series of frames or a still image composed of a single frame.

  The decoding unit 31 receives the code information from the code input unit 30 and outputs a quantized value, division type information, and prediction information by performing variable length decoding as the reverse procedure of the encoding unit. The quantization value, the prediction information, and the division type information obtained by the variable length decoding are sent to the inverse quantization unit 37, the compensation unit 32, and the adaptive division unit 40, respectively. Note that the decoding (decoding up to the image output means 300 and the storage means 41 as decoded pixels) is also performed for each of the above-described small regions.

  The inverse quantization means 17 (inverse quantization means 37) performs the reverse process of the quantization process, so that the quantization means 17 in the coding configuration (FIG. 1) and the decoding means in the decoding configuration (FIG. 2). The quantized value sent from 31 is inversely quantized into a transform coefficient. The transform coefficient obtained by inverse quantization is sent to the inverse transform means 18 (inverse transform means 38).

  The inverse transform unit 18 (inverse transform unit 38) performs an inverse orthogonal transform on the transform coefficient sent from the inverse quantization unit by performing an inverse process of the orthogonal transform. The prediction residual obtained by the inverse transformation is sent to the adder 19 (adder 39).

  The prediction unit 11 determines prediction information for reducing the redundancy of the image. The prediction unit 11 refers to the encoded pixels stored in the storage unit 21 and is an area set by the adaptive division unit 20 described later. Information for approximating the input signal of (each divided small area) is determined as prediction information. The determined prediction information is sent to the compensation means 12 and the encoding means 16 (illustrated by dotted arrows).

  Regarding the determination of the prediction information in the prediction means 11, a prediction mode is selected that is individually encoded in each prediction mode and minimizes the encoding cost calculated from the code amount and the distortion amount. Details of the method for minimizing the coding cost are described in Non-Patent Document 1.

  Various types of prediction modes can be used. For example, H.264 uses Intra prediction (intra-screen prediction). FIG. 3 shows nine types of prediction modes for H.264 intra prediction when the unit block size is 4 × 4. The nine types of prediction modes may be used, or the prediction mode in the case where the unit block size disclosed in Non-Patent Document 1 or the like is 8 × 8 or 16 × 16 may be used.

  In the present invention, the corresponding processes up to prediction and encoding are performed for each small area obtained by further dividing the unit block. Here, the small area is set to a rectangle as will be described later. Regarding the application of prediction for each small region, the positional relationship and calculation relationship between the pixel used for prediction and the predicted pixel defined in each conventional method is limited to or expanded by the small region. Application can make predictions.

  As an example, FIG. 4 shows an example in which prediction mode 0 (vertical prediction) and prediction mode 2 (average value prediction) of 4 × 4 intra prediction are applied to a small region. (1) in FIG. 4 is the prediction mode 0 of the conventional 4 × 4 intra prediction, and each pixel in the 4 × 4 unit block B01 is predicted in the vertical direction. That is, the values of the respective pixels (a), (b), (c), and (d) in the area of 4 × 1 (= horizontal pixel number × vertical method pixel number, the same applies in the following description) at the upper end of the block B01 are respectively Are used as predicted values for each of the 1 × 4 regions (A), (B), (C), and (D). Also, (4) in FIG. 4 is the prediction mode 2 of the conventional 4 × 4 intra prediction, and the common prediction value for all the pixels of the 4 × 4 unit block B01 is the adjacent inverted L-shaped region (a). It is an average value of ~ (d) and (e) ~ (i).

  (2) and (3) in FIG. 4 show cases where the prediction mode 0 described in (1) is applied to R02 and R03, which are examples of small regions. In (2), the small region R02 is a 2 × 1 region and is composed of pixels (X1) and (Y1). Due to the vertical prediction to the region R02 (prediction application limited to a region smaller than the 4 × 4 unit block B01), the predicted values of the pixels (X1) and (Y1) are encoded pixels ( The pixel values are a) and (b).

  In (3), the small region R03 is a 2 × 6 region, and the region is divided into two 1 × 6 pixel regions (X2) and (Y2). Due to the vertical prediction to the region R03 (the prediction application expanded including the portion exceeding the 4 × 4 unit block B01), the prediction values of the pixels in the pixel regions (X2) and (Y2) are present vertically above, respectively. The pixel values of the pixels (a) and (b) are obtained.

  In the cases (2) and (3) above, the pixels (a) and (b) that are commonly used for prediction are more limited than those in the case (1) by corresponding to the small regions R02 and R03, respectively.

  (5) and (6) in FIG. 4 show cases where the prediction mode 2 described in (4) is applied to small regions R02 and R03 similar to (2) and (3), respectively. In (5), each pixel of the small region R02 is predicted as an average value of the corresponding adjacent region pixels (a), (b), (e), and (f). The adjacent pixel region is limited compared to the case of (4) by corresponding to the small region R02.

  In (6), each pixel in the small region R03 is predicted as an average value of the corresponding adjacent region pixels (a), (b), and (e) to (k). Since the adjacent pixel region corresponds to the small region R03, the portions (c) and (d) are limited and the portions (j) and (k) are expanded as compared with the case of (4).

  Similarly to the example of FIG. 4 described above, in the other prediction modes shown in FIG. 3, even in the general case where the shape of the small region is not 4 × 4, it corresponds to the small region of the adjacent region pixel. Prediction can be performed by limitation or expansion and application of a prediction value calculation method in each prediction mode in accordance with the correspondence.

  That is, as shown in (7) of FIG. 4, when the small region R04 is size a × b, the adjacent pixel region as the region of the encoded pixel used for prediction is adjacent to the upper side of the small region. An a × 1 region R05, a 1 × b region R06 adjacent to the left side of the small region, and a 1 × 1 region R07 on the diagonal of the region R04 connecting both the regions R05 and R06 Should be set. In addition, when applying prediction mode 3 or 7 of 4 × 4 intra prediction, region R05 is expanded to the right, and when applying prediction mode 8 of 4 × 4 intra prediction, region R06 is downward. What is necessary is just to expand. The prediction value in each prediction mode may be calculated using the adjacent pixel region corresponding to the size of the small region.

  Furthermore, similarly, prediction modes in the sizes of 4 × 4, 8 × 8, and 16 × 16 in H.264 may be used when the unit block has a different size in the present invention. For example, in the present invention, the unit block is set to a size of 64 × 64, and the prediction modes consisting of 9 types of 4 × 4 in H.264 are applied to each small region within the size 64 × 64. Good.

  The compensation unit 12 (compensation unit 32) is a coded pixel (decoded pixel) obtained by referring to the storage unit 21 (storage unit 41) and prediction information determined by the prediction unit 11 (decoded by the decoding unit 31). Based on the prediction information), a prediction pixel is generated as an approximate value of the input pixel in the small region set by the adaptive division means 20 (adaptive division means 40) described later. The generated prediction pixel is sent to the differentiator 13 and the adder 19 in the encoding configuration (FIG. 1). On the other hand, it is sent to the adder 39 in the decoding configuration (FIG. 2).

  The division into small areas by the adaptive dividing means 20 which is a characteristic configuration of the image encoding device 1 of the present invention will be described.

  The role of the adaptive division means 20 is to select a division into small regions that is most suitable for the current block in order to predict and compensate the current block in units of small regions in the prediction means 11 and the compensation means 12 in the subsequent stage. To divide. However, if the encoding target block is divided into arbitrary shapes, it is necessary to encode the shape information, and there is a problem that the encoding efficiency cannot be improved. In order to solve this problem, a procedure is used in which a partition configured under all of the following constraints [1] to [4] is used or further selected from the partitions under the constraint.

[1] All small areas are rectangular
[2] The total number of divided small areas is n or less
[3] The maximum size of the small area is u or more or the minimum size v or more
[4] The number of small areas with the minimum size t is m or less

  In the above [1] to [4], n, u, v, t, and m are predetermined natural numbers. The size of the small area is defined as the number of pixels. For example, both the 2 × 2 small area and the 1 × 4 small area have the same size because the number of pixels is 4, and the 3 × 2 small area and the 1 × 5 small area have the same size as the 3 × 2 small area. The size is larger.

As a specific example, if the unit block size is 16x16,
n = 8; u = 12 × 8 or 8 × 12; v = 8 × 4 or 4 × 8; t = 4 × 4; m = 4
Then, there are 5398 types of division methods. Further, under the above conditions, if n = 4, there are 258 types of division, and n = 3, there are 49 types of division.

  In this way, the division obtained by setting predetermined values to n, u, v, t, and m as parameters under general constraints [1] to [4] may be used. As a further selection for the division, a division such as combining the divisions on behalf of the similar divisions manually may be used. In this way, each of the division methods prepared in advance and made available for selection by the adaptive division means 20 is called a division type, and a combination of the group of division types is called a division type set. Examples of the set of division types are shown in FIGS.

  The set of divided types shown in FIG. 5 is composed of a total of 12 types of divided types A0 to A11. The set of divided types shown in FIG. 6 is composed of a total of 28 types of divided types B0 to B27. In addition, the set of division types shown in FIG. 7 is shown as 33 types of division types C0 to C32. This is the above-mentioned n = 4, and all 258 types of divisions are summarized as rotations and symmetry as representatives. This shows the divided species. In FIG. 7, the fact that n small regions are generated by dividing a unit block by applying the division type to all the division types is also shown as [n division]. In FIG. 7, the division type drawn using the dotted line is a division type that gives a small area with the number of divisions to be written by selecting any one or more of the dotted lines as a solid line that is actually divided.

  FIG. 8 is a diagram showing how, for example, the division types C4 of FIG. 7 are collected as a supplementary explanation of FIG. By selecting one of the three dotted lines D0, E0, or F0, the division type with the division number of 3 is determined as D1, E1, or F1, respectively. In addition, those divided types D1, E1, and F1 rotated by 90 °, 180 °, and 270 ° (clockwise) are all different divided types, and are determined as divided types D2 to F4. In this way, the division type C4 is a group of twelve division types, ie, the division type D1 to the division type F4, by selecting one of the dotted lines and considering rotation.

  FIG. 7 shows an example of a set of 258 divided types by collectively expressing the divided type C4. FIG. 5 and FIG. 6 are examples of a set of divided types created by combining similar divided types from the set of 258 types of divided types. Although FIG. 7 shows a set of division types created with unit blocks of 16 × 16 as described above, it may be used for unit blocks of a size different from 16 × 16 by similar enlargement or reduction. The same applies to FIGS. 5 and 6.

  In the present invention, as an example of a feature of a preferable division type set that is further manually selected under the above constraints [1] to [4], (1) a division type in which the entire square unit block is directly used as one small region There is a set including a division type (for example, A4, A6, A9, and A11) that includes a square smaller than the unit block and a non-square that touches the square as a small region (for example, division type A0 in FIG. 5). According to (1), even if the unit block size is fixed, the same effect as that in which the block size is adaptively changed in the image according to the image can be obtained.

  Further, as an example of a feature of a preferable division type set, (2) a set including a division type whose boundary line between small areas is only in the horizontal direction (division type A2 or the like) and / or only in the vertical direction (division type A3 or the like). is there. By setting (1) and (2), it is possible to increase the encoding efficiency corresponding to more various images.

  As described above, the adaptive division unit 20 holds a set of given division types in advance, and individually encodes each division type included in the set, and calculates from the code amount and the distortion amount. Select the split type that minimizes the cost. FIG. 9 shows a specific flow of the selection, that is, a flow for performing encoding by selecting a division type for each unit block constituting the input image.

  The flow tries to divide each unit block i by all the division types j, and applies the prediction information that gives the minimum coding cost in each of the small regions k included in each division type j, depending on the division type j. The encoding cost of the unit block i is obtained, and the division type that gives the minimum coding cost is obtained. Note that the entity that manages the counters i, j, and k in each step and performs the operation linked to each functional block is the image encoding device 1 itself, in particular, the control means included in the device 1 (not shown in FIG. 1). ).

  In step S0, the flow is started. In step S1, the unit block counter i is set to the initial value 0. In step S2, the i-th unit block is read into the image encoding device 1 from the image input means 10. In step S3, the counter j for each division type included in a given division type set preset in the adaptive division means 20 is set to an initial value of zero.

  In step S4, the adaptive division means 20 divides the i-th unit block into a plurality of small regions (including a case where only one is included) defined by the j-th type of division type j. The division is a provisional division for calculating the minimum coding cost for the unit block i, and the division for actually encoding and sending to the code output means 100 as the output of the entire image coding apparatus 1 is the provisional division. It is determined in the following steps from among the above. In step S5, the counter k for each small region included in the division type j is set to an initial value of 0.

In step S <b> 6, the prediction unit 11 determines prediction information that provides the minimum encoding cost for the k-th small region k configured by the encoding target pixels. In calculating the cost, as described above, the prediction unit 11 obtains the encoding cost c (m) of the small region k for each prediction information, that is, each prediction mode m, and the prediction mode m = m that minimizes the cost. _k is determined as prediction information.

The determination is a provisional determination as described in step S4. In step S6, the prediction means 11 to the adder 19 perform each process according to the temporarily determined prediction mode m _k so that the prediction information of the subsequent k + 1-th and subsequent small regions can be provisionally determined. Thus, the encoded pixel relating to the temporary determination of the small region k is stored in the storage unit 21. The stored encoded pixels are provisionally determined as described in step S4 and generally do not match the final determined encoded pixels sent to the code output means 100.

  In step S7, it is determined whether or not the encoding process has been completed for all the small areas k. If an unprocessed small area remains, the counter k is added in step S8, and the provisional determination and the corresponding encoding are continued again in step S6 for the next small area to be encoded. If step S6 is completed for all the small regions k determined by the division type j, the process proceeds from step S7 to step 9.

  It is assumed that the order for the sequential encoding is determined in advance for each small region k of each division type j. FIG. 10 shows an example of the order. Here, the division type B24 of FIG. 6 is shown as an example, and is divided into four small regions (a) (b) (c) (d). In this case, the encoding order may be (a) (b) (c) (d) in alphabetical order given to the small regions (first case), (a) (c) (d) (b (In the second case), one of which is predetermined.

  The first case is a style in which the vertical direction is sequentially filled with the encoded region and sequentially proceeds in the horizontal direction, and the second case is the case in which the horizontal direction is filled with the encoded region and the vertical direction is sequentially advanced. Style. The first style or the second style or a mixed style may be set for each division type, starting from the upper leftmost small area in contact with another encoded unit block. In addition, since the applicable prediction mode is limited, an order in which a small region group that has already been encoded in the unit block appears apart from the region before encoding in the middle (in the middle, encoded small No order is set so that the region group may not be a single connected region. In FIG. 10, for example, the order of (a) (d) (c) (b) and the like is not set.

In step S9, the encoding cost C _j when the unit block i is divided by the division type j is calculated as the sum Σc (m _k ) of the encoding cost c (m _k ) of each small region k of the division type j. And remember.

In step S10, it is determined whether or not the processing in steps S4 to S9 has been completed for all of the division types j. . If completed, the process proceeds to step S12, and the division type j actually used for encoding is determined. In this determination, the division type that minimizes the coding cost C _j = Σc (m _k ) obtained in step S9 is selected for each division type j.

  In step S 12, the unit block i is actually encoded for each small region k based on the finally determined division type j, and sent to the code output unit 100 and the storage unit 21. At this time, also for the final determined division type j, each subregion k is already processed and encoded by the prediction unit 11 to the storage unit 21 when the encoding cost is calculated in step S6. If it is held, there is no need to re-encode.

  In step S13, it is determined whether the division type j is determined for all the unit blocks i and the encoding process according to the determination is performed. If completed, the process proceeds to step S15, and the flow ends. If not completed, the process proceeds to step S14, the counter i is added, the process returns to step S2, and the process is repeated for the next unit block.

  The unit block has a predetermined size, and each frame or still image of the moving image is sequentially read by the counter i in raster scan order in step S2.

  FIG. 11 shows the stream data actually received by the code output means 100 by calculating the encoding cost for each provisionally determined division type in the flow of FIG. 9 and finally determining the minimum cost division type. It is a flowchart explaining the order that the code | symbol information generate | occur | produces.

  When the flow starts in step S20, the counter i of the unit block is set to the initial value 0 in step S21. In step S22, encoding of the unit block i starts. In step S23, information of the division type j = j (i) applied to the unit block i by the final determination is encoded. In step S24, each small region k determined by the division type j (i) is sequentially encoded. In the encoding, prediction information (applied prediction mode) and a quantized value are encoded. The quantization value is obtained by processing the prediction pixel, the prediction residual, the transform coefficient, and the quantization value in order by the compensation unit 12 to the quantization unit 15 according to the prediction information.

  Note that, as described with reference to FIG. 10, the encoding order of each small region k is determined in a given order in the division j (i). That is, the information of the encoding order is included in the information of the division type j = j (i) encoded in the previous step S23.

  When all the small areas k are encoded, the encoding of the unit block i is completed in step S25. In step S26, it is determined whether or not the encoding of all the unit blocks i has been completed. If it has not been completed, the process proceeds to step S27, the counter i is added, and the process is repeated from step S22 for the next unit block. If completed in step S26, the process proceeds to step S28 and ends.

  FIG. 12 is a detailed functional block diagram of the adaptive dividing means 20, and FIG. 13 is a detailed functional block diagram of the compensating means 12. The adaptive division unit 20 includes a division type set holding unit 201, a coded channel information storage unit 202, a division type identification number assigning unit 203, and a prediction information identification number assignment unit 204, each of which is an additional part of the adaptive division unit 20. Take on the function. The compensation unit 12 includes a small area unit collective compensation unit 121, a pixel unit iterative compensation unit 122, and a switching unit 123, and each unit bears an additional function of the compensation unit 12. Hereinafter, the functions of these units will be described.

  The division type set holding unit 201 holds a plurality of sets of predetermined division types that differ depending on the size of the encoding unit block. In particular, a set that satisfies the relationship that a set corresponding to a larger unit block includes a larger number of division types, and conversely, a set corresponding to a smaller unit block includes a smaller number of division types. Hold a given set.

  For example, if the size of the unit block is 3 × 4 × 4, 16 × 16, and 64 × 64, the set of division types corresponding to the minimum size 4 × 4 is the first of all 12 types shown in FIG. The set of division types corresponding to the intermediate size 16 × 16 is the second set of all 27 types shown in FIG. 6, and the set of division types corresponding to the maximum size 64 × 64 is shown in FIG. The third set of all 258 types shown in FIG.

  By using a set that satisfies this relationship, the proportion of control information occupied by a larger size is relatively small. Therefore, by setting a larger number of divisions and a smaller number for a smaller size, efficient coding is possible for each size. Is effective. Similarly, in the present invention, since various divisions including a non-square rectangle are prepared as a set of division types, it is basically desirable to set the unit block size to be relatively large.

  The actual unit block size and corresponding division type set to be used may be set manually, or if it is a still image, encoding is attempted with all unit block sizes (and corresponding division type sets). Of these, the one with the lowest coding cost may be used. In the case of a moving image, the minimum frame size may be obtained for each of the first frame or every frame at a predetermined interval, and the unit size may be used for the subsequent frames.

  The encoded channel information storage unit 202 includes information on the division type finally determined and selected for the pixel of the first channel encoded by the flow of FIG. 9 for each unit block and the division type. The prediction information applied in each small area is stored, and when encoding the pixel of the second channel before encoding in the unit block, each small element corresponding to the stored division type in the first channel is stored. The prediction information of the area can be applied as it is.

  An example of this application is shown in FIG. Here, as an example, in the RGB channel of the three primary colors, the G signal frame which is the first channel shown in (1) and the R signal frame which is the second channel shown in (3) are shown. As shown in (1), the unit block IG0 of the G signal has already been encoded by the flow of FIG. The size of the unit block is 4 × 4, the set of FIG. 5 is used as the set of division types, and the prediction mode is the prediction mode of FIG. 3, so that the unit block IG0 is as shown in (2). Are divided into four 2 × 2 small areas by the division type A1, and prediction modes 0, 1, 2, and 3 are applied to each small area in the raster scan order.

  The encoded channel information storage unit 202 stores the result (2) of the encoding so that the encoding shown in (3) at the same position and the same time as the unit block IG0 of the encoded first channel The same division type and prediction information of each small region as in (2) are applied to the unit block IR0 of the previous second channel as shown in (4). In addition, although it has been described as a frame assuming a moving image, it may be a still image.

  With this application, the loop L1 and the loop L2 in FIG. 9 can be omitted in the second channel. That is, the prediction information that minimizes the coding cost of each small region is obtained for each division type (loop L2), and the division type that minimizes the coding cost of each division type is obtained (loop L1). By omitting, the calculation load can be reduced. Further, since there is a correlation between channels of normal signals, the omission has little adverse effect on coding efficiency.

  Also, the amount of code can be reduced by the application. That is, information on the division type shared by the first channel and the second channel and the prediction information of each small region, the quantization value (calculated from the prediction residual) of the first channel, and the second Since it is only necessary to encode the channel quantization value (calculated from the prediction residual), the amount of code of the shared part is saved compared to the case where the first and second channels are encoded separately. The Once the sharing is encoded as flag information, the processing related to sharing may be continued thereafter. Furthermore, for the third channel such as a B signal, the division type information and the prediction information of each small area may be shared with the first channel to reduce the code amount and reduce the calculation load.

  The division type identification number assigning unit 203 stores a selected ratio of each division type when a predetermined number of encoding target unit blocks are encoded in advance, and a division type set used when encoding the subsequent encoding target blocks The identification number of each division type is reassigned. Here, at the time of the re-assignment, the re-assignment of the smaller identification number to the divided type having a larger selected ratio allows the subsequent divided types to be selected at substantially the same rate. The amount of codes can be reduced.

  For example, a predetermined amount of encoding is performed using the division types A0 to A11 of FIG. 5 as the division type set, and the division types most frequently selected in actual encoding are the division types A3, A0, A1,. Suppose that In this case, in the subsequent encoding, the identification number 0 is assigned to the division type A3, the identification number 1 is assigned to the division type A0, the identification number 2 is assigned to the division type A1, and so on. It becomes.

  The prediction information identification number assigning unit 204 has a function of achieving a code amount reduction effect similar to that of the division type identification number giving unit 203, and is used as a set of division types when a predetermined number of encoding target unit blocks are encoded in advance. Stores the ratio of each prediction mode determined for each shape of the small area included in all the division types to which it belongs, and identifies the prediction mode for each shape of the small area used in the encoding of the subsequent encoding target block Reassign the number. Here, in the reassignment, separate reassignment is performed for each shape, and a smaller identification number is reassigned to a prediction mode with a larger selected ratio, so that the subsequent prediction modes are also applied to small regions of the shape. Therefore, it is possible to reduce the code amount of the prediction mode under the assumption that almost the same ratio is selected.

  FIG. 15 is a diagram illustrating an example for explaining reassignment of the identification number of the prediction mode for each shape. G1 to G12 in FIG. 15 represent the shapes of the small regions when A0 to A11 in FIG. 5 are used as the set of division types. That is, if the unit block sizes of A0 to A11 are 4 × 4, the shape G1 (size 4 × 4) is the division type A0, and the shape G2 (size 3 × 3) is the division types A4, A6, A9, and In A11, the shape G3 (size 2 × 2) is a shape of a small area included in the division type A1, and the shape G4 (size 1 × 1) is a shape of a small area included in each of the division types A4, A6, A9, and A11.

The correspondence between each of these shapes and the division type including the small region of the shape is as shown in the table T1. Similarly to the above, if the size of the unit block is 4 × 4, the sizes of the shapes G5 to G12 are as follows.
G5 (size 3x2); G6 (size 2x3);
G7 (size 4 × 1); G8 (size 3 × 1); G9 (size 2 × 1);
G10 (size 1x4); G11 (size 1x3); G12 (size 1x2)

  The prediction information identification number assigning unit 204 stores the ratio of the determined prediction mode for each such shape. For example, assuming that the prediction mode of FIG. 3 is used, the prediction mode 1 is most frequently determined for the shape G1 appearing in the division type A0, and 80% of the whole is determined, and the prediction mode is used for the shape G2 appearing in the division type A4, A6, A9, or A11. 2 is the largest and 90% of the whole is determined, and the determination ratio of each prediction mode in each shape such as etc. is stored. For the prediction mode applied to the shape G1, the smallest identification number is assigned to the prediction mode 1, and for the prediction mode applied to the shape G2, the smallest identification number is assigned to the prediction mode 2. Reduce the amount of code.

  In addition, in the division type identification number assigning unit 203 and the prediction information identification number assigning unit 204, a predetermined number of encoding target blocks to be pre-encoded in order to reassign the identification numbers are moving images having a start time of 1 For each frame or every frame at a predetermined interval, the re-assigned identification number may be used when encoding subsequent frames. If it is a still image, it may be a predetermined number on the younger side of the raster scan order for each still image, and an identification number for re-assignment to another still image that is encoded in advance and read afterwards. May be applied.

  In addition, when the predetermined number of encoding target unit blocks are encoded in advance, the division type identification number assigning unit 203 uses each prediction type for each division type, and the prediction information identification number assignment unit 204 uses each prediction mode for each shape. Is given an identification number as a given initial value.

  The small region unit collective compensation unit 121 generates a prediction pixel for each small region by a normal method as described with reference to FIG. That is, as shown in (2) and (3) in FIG. 4, the predicted value is obtained as it is without changing the value of the pixel in the adjacent pixel area, or the average calculation is performed as in (5) and (6). As a result, the prediction pixels are obtained for the entire small area at once.

  On the other hand, the pixel unit repetitive compensation unit 122 does not obtain the prediction pixels for the entire small area at once, but sequentially generates the prediction pixels in units of pixels. FIG. 16 conceptually shows the iterative generation. In FIG. 16, the repeated generation of the prediction pixel in the prediction mode 0 of FIG. 3 is shown in time series (1) to (4) for the pixels in the small region B100.

  In (1), the pixel P1 is predicted as a weighted average of the neighboring pixel region R1 rearward in the prediction mode 0, that is, in the vertical direction in the vertical prediction. Subsequently, the pixel P2 in the vertical direction from P1 is also predicted as a weighted average of the neighboring pixel region R2 as shown in (2), and similarly the pixel P3 in the vertical direction of P2 is predicted as the neighboring pixel region as shown in (3). The weighted average of R3 is predicted, and the pixel P4 in the vertical direction of P3 is predicted as the weighted average of the neighboring pixel region R4 as shown in (4).

  In general, the pixel value of P1 having a short distance is close to the pixel value of the nearest pixel in the adjacent pixel region R1, and the pixel value of P4 having the longest distance tends to be greatly different from the pixel value of the adjacent pixel region R1 with respect to P1. There is. Therefore, in particular, for the farthest pixel value of P4, it is often the case that the encoding efficiency increases as a whole when one of the pixels of R1 is not directly used as a predicted value but the average of R1 and other regions is adopted. This becomes more noticeable especially when the size of the unit block is increased and the small area is also increased.

  According to the pixel unit repetitive compensation unit 122, there is an effect of obtaining a predicted value in consideration of the whole of the pixel having the short distance and the pixel having the far distance. That is, the predicted value of the pixel P1 or the like that is close to the encoded adjacent pixel region R1 is close to the immediate pixel value itself of the adjacent pixel region R1, and is such as the pixel P4 that is far from the adjacent pixel region R1. The predicted value is close to the average value of the adjacent pixel region R1 and the like. In other words, the effect of the prediction mode 0 (vertical prediction) in FIG. 3 is strongly obtained for the close pixels such as P1, and the effect of the prediction mode 2 (average value prediction) is obtained for the distant pixels such as P4. Is strongly obtained, so that the prediction of the hybrid method is applied to the entire small region, and the coding efficiency is increased.

  In order to obtain the predicted values in (2) to (4) of FIG. 16, it is necessary to obtain the predicted values of R2 to R4 belonging to the small region B100 to be encoded. It is explained that a predicted value is also obtained. 17 and 18 show specific examples of iterative compensation by the pixel unit iterative compensation unit 122. FIG. In FIG. 17, pixels x (1,0) to x (4,0) and x (0,1) to x (0,3) to which the coordinate value of (A) is added are encoded adjacent pixel regions. In other words, the small region to be encoded is a region of size 3 × 3 of pixels y (1,1) to y (3,3) to which coordinate values are added. In addition, [1] and the like added to the upper part of y (1,1) indicate the order in which the iterative compensation is performed within the small region.

  Further, (B) shows the positions of pixels for which weighted averaging is performed when predicting each pixel y (i, j), and examples of weighting coefficients. That is, the pixels used for prediction are, for example, the encoded pixel x or the predicted pixel y at positions (i-1, j-1), (i, j-1), (i + 1, j-1). The weighting coefficients for the pixels at the respective positions are 1/4, 1/2, and 1/4. A direction D10 is a prediction direction, and in this example, iterative compensation is performed in the prediction mode 0 (vertical prediction) in FIG. From the description of (A) and (B) above, iterative prediction is performed sequentially as follows.

[1] y (1,1) = x (0,0) / 4 + x (1,0) / 2 + x (2,0) / 4
[2] y (2,1) = x (1,0) / 4 + x (2,0) / 2 + x (3,0) / 4
[3] y (3,1) = x (2,0) / 4 + x (3,0) / 2 + x (4,0) / 4
[4] y (1,2) = x (0,1) / 4 + y (1,1) / 2 + y (2,1) / 4
[5] y (2,2) = [... the same applies below and is omitted]

  As described above, in the case of the vertical prediction in the direction D10 shown in (B), the region R10 behind the straight line L10 perpendicular to D10 is already encoded with respect to the pixel y (i, j) for which the prediction value is obtained. [1], [4], [1], [4], and [1] to [3] while sequentially predicting the pixels arranged in the direction of the L10 as shown in (A) so as to be the region of the pixel x or the predicted pixel y. Just go in the direction of D10 as shown in [7].

  If the same is possible, the prediction may be performed by weighting a desired region in R10 with a desired coefficient. In this example, the weighting coefficient is fixed to 1/4, 1/2, 1/4 for each position in the figure, but the coefficient and the position of the pixel used for prediction are changed for each pixel for which the prediction value is obtained. Also good.

In particular, in the above example, when predicting y (3,2) for the [6] th, the pixel at the position of y (4,1) often belongs to the subsequent encoding target block or small region and is not generally available. Therefore, the weighting coefficient is 1/4, 1/2, 1/4 and the position of the pixel used for prediction is corrected, and prediction is performed as follows, for example.
[6] y (3,2) = {y (2,1) / 4 + y (3,1) / 2} × (4/3)
In addition, since [9] -th y (3,3) cannot be generally used, it may be the same as described above. If the small region is at a position as shown in (b) in the order of (2) in FIG. 10, y (4,1) and y (4,2) can also be used as encoded pixels. The prediction may be performed using the coefficients 1/4, 1/2, and 1/4 as in the case of the pixels.

  Note that as the pixel position moves away from the encoded pixel, the predicted value needs to converge to the average value of the encoded pixel region. Therefore, it is preferable to impose a condition that each weighting coefficient α for obtaining a predicted value of each pixel is 0 ≦ α ≦ 1, and the sum of coefficients in all pixels used for prediction is 1.

  FIG. 18 is an example of iterative compensation in the prediction mode 4 in FIG. 3, the direction of prediction is shown as a direction D20 in (B), and the encoded adjacent pixel region and the small region to be encoded shown in (A) are This is common with FIG. In this case, as a difference from the iterative compensation in FIG. 17, the order of the iterative compensation is changed as shown in (A), the position of the pixel used for prediction and the weighting coefficient are changed as shown in (B), This is because the direction D20 differs from the direction D10 in FIG. 1 by 45 °. In this case as well, iterative compensation can be performed sequentially as follows.

[1] y (1,1) = x (0,1) / 2 + x (1,0) / 2
[2] y (1,2) = x (0,2) / 2 + y (1,1) / 2
[3] y (2,1) = y (1,1) / 2 + x (2,0) / 2
[4] y (1,3) = x (0,3) / 2 + y (1,2) / 2
[5] y (2,2) = [... the same applies below and is omitted]

  Here, as shown in (A), the region R20 behind the straight line L20 perpendicular to the prediction direction D20 as shown in (B) becomes the region of the encoded pixel x or the predicted pixel y, as shown in (A). What is necessary is just to advance to the direction of D20, predicting the pixel lined up in the direction of L20 sequentially, and the setting of the order of prediction with respect to the direction of the said prediction is the same as that of FIG. The setting change of each coefficient is the same as in FIG.

  Similar to the examples of FIGS. 17 and 18, iterative prediction can be performed in the prediction modes in the other directions of FIG. 3. Note that when applying prediction mode 2 (average value prediction), the prediction direction is not defined in this mode, so that iterative prediction is omitted, and the average value is adopted as the prediction value of all pixels in the small region.

  In addition, when generating the prediction pixel by the pixel unit repetitive compensation unit 122, as a result of performing the repetitive calculation for each pixel as described above, each pixel in the adjacent pixel region that has been encoded for each point y in the small region A value obtained by multiplying x by a given weighting factor is used as a predicted value. Here, the given weighting coefficient has a property that the smaller the distance between the pixel y and the pixel x, the smaller the weight coefficient, and the larger the direction from the pixel x to the pixel y that is aligned with the direction defined by the prediction mode. Have.

  Instead of performing iterative calculation, a coefficient for predicting each pixel y from each of the above-mentioned pixels x may be obtained in advance, and the predicted pixels may be obtained in a lump. Therefore, it is disadvantageous in terms of code amount compared to the coefficients 1/4, 1/2, 1/4, etc. in FIG. If the coefficients are 1/4, 1/2, 1/4, etc., the relative position from the pixel to be predicted is predetermined in the direction defined by each prediction mode for all pairs of the shape of the small area and the prediction mode. Since the coefficients 1/4, 1/2, 1/4, etc. can be used in common for the pixel at the position of, the amount of code is reduced.

  The switching unit 123 switches which of the small region unit collective compensation unit 121 and the pixel unit iterative compensation unit 122 is used for encoding. As the switching determination, the coding cost is compared between the case where the small region unit collective compensation unit 121 is applied for each small region and the case where the pixel unit iterative compensation unit 122 is applied. Switch. That is, step S9 of FIG. 9 is performed for the application of the small region unit collective compensation unit 121 in each prediction mode and the application of the pixel unit iterative compensation unit 122 in each prediction mode, and obtaining the case of the minimum cost. Become.

  FIG. 19 is a flowchart of the sequential decoding for each small area of each unit block in the image decoding apparatus 3. Each step (S30 to S38) in FIG. 19 corresponds to each step (S20 to S28) in FIG.

  In step S30, the flow is started. In step S31, the counter i of the unit block to be decoded is set to an initial value 0, and decoding of the unit block i is started in step S32. In step S33, the decoding unit 31 decodes the code information encoded in step S23 of FIG. 11, and obtains corresponding information.

  That is, in step S33, information on the division type j = j (i) applied to the unit block i is obtained as the premise information for decoding the unit block i. As described above, the information includes information on the position and shape of each small region k constituting the division type, decoding order, and the like.

  In step S34, the small area k is sequentially decoded. Therefore, the decoding unit 31 sequentially decodes the information sequentially encoded in step S24 for each small region k, and obtains prediction information (applied prediction mode) and a quantization value.

  The prediction information includes information on the prediction mode applied to the small region k, information on which of the small region unit collective compensation unit 121 and the pixel unit iterative compensation unit 122 is applied, and the pixel unit iterative compensation unit 122. If applicable, information about the coefficient is included. When the encoded channel information storage unit 202 is used, the division type information includes information indicating sharing with the first and second channels.

  In step S34, when the small area k is decoded and sent as decoded pixels to the image output means 300, the small areas after k + 1 are sequentially stored in the storage means 41 for decoding. At this time, the (decoding side) inverse quantization means 37 to storage means 41 function in the same manner as the corresponding functional blocks on the encoding apparatus 1 side. However, it is not necessary for the (decoding side) adaptive division means 40 and the like to attempt division or the like in all division types as in the flow of FIG. 9 in the (encoding side) adaptive division means 20 and the like.

  That is, in step S34, the adaptive dividing means 40 only needs to reset the small area k in the unit block i as an area to be decoded according to the decoded division type j = j (i). The compensation unit 32 decodes the predicted pixel for the reset small region k using the decoded pixel of the storage unit 41 and the decoded prediction information. The adder 39 adds the decoded prediction residual and the decoded prediction pixel to obtain a decoded pixel. The decoded prediction residual is obtained by processing the decoded quantized value by the inverse quantization unit 37 and the inverse transform unit 38.

  When the processing is completed for all the small areas k in step S34, the decoding of the unit block i is completed in step S35, and it is confirmed in step S36 whether all the unit blocks have been processed. If incomplete, the counter i is added in step S37, and the next unit block is repeated from step S32. If completed, the process proceeds to step S38, and the flow ends.

  When each functional block (other than the encoded channel information storage unit 202) of the adaptive division unit 20 shown in FIG. 12 is used on the image encoding device 1 side, the initial value i = 0 or each functional block The corresponding information is decoded in the unit block decoding start step S32 to which i is applied, and the encoding apparatus 1 side encodes the unit block size, the identification number assignment to the prediction mode, and the like. Decoding should be performed with the same setting as in FIG.

  As described above, according to the image encoding device 1 of the present invention, when each unit block is divided, the unit block is adaptively divided into small areas including a non-square shape designated in advance, and prediction means and encoding according to the small areas are performed. -Encoding efficiency is improved by applying decoding means.

  In particular, in the present invention, by preparing various division types as shown in FIGS. 5 to 7, for example, in the case of using the set of FIG. 5, the division type A0 is automatically obtained in a region where the pixel value is flat. Such as the division type A2, A3, A1, etc. are selected in the region where the pixel value changes drastically, and is optimal for an intermediate region between them. A division type including a small region is also appropriately selected. In this way, even if the size of the unit block is fixed, the same effect as that of automatically changing the size of the unit block according to the distribution of pixel values can be obtained.

  Also, the shape of the small area is prepared in a variety of ways under the above constraints [1] to [4] so that the code amount does not increase excessively, and in the flow of FIG. Coding cost c (m) is estimated for each prediction mode m for k. In this case, since there is no restriction on the prediction mode m that can be used depending on the shape of the small region k, the optimum one is selected from all of the commonly used prediction modes m. improves.

  For example, in the prediction mode of FIG. 3, when the division type A2 including four horizontally long small areas of size 4 × 1 in FIG. 5 is applied, generally the nearest pixel is often close in value. In many cases, the prediction mode 0 (vertical prediction) gives a good prediction value.In the present invention, however, the opposite is true.For example, the prediction mode 1 (horizontal prediction) in which the pixels are far away gives a good prediction value. Thus, encoding efficiency can be increased.

DESCRIPTION OF SYMBOLS 1 ... Image coding apparatus, 11 ... Prediction means, 12 ... Compensation means, 13 ... Differentiator, 14 ... Conversion means, 15 ... Quantization means, 16 ... Encoding means, 17 ... Dequantization means, 18 ... Inverse transformation Means 19: Adder 20 ... Adaptive division means 21 ... Storage means 3 ... Image decoding device 31 ... Decoding means 37 ... Inverse quantization means 38 ... Inverse transform means 39 ... Adder 32 ... Compensation Means, 40 ... adaptive division means, 41 ... storage means

Claims (9)

Prediction means for determining prediction information for predicting each pixel from an encoded pixel for each pixel of a unit block composed of a plurality of pixels;
Compensating means for generating a prediction pixel of each pixel based on the prediction information,
Image coding that performs orthogonal transformation and quantization on the prediction residual obtained by performing difference processing between each pixel and its prediction pixel to produce a quantized value and performs coding for each unit block In the device
Adaptive division means for selecting and dividing a division type from a set of division types that determine division into a plurality of small areas of a unit block;
For each unit block, the prediction unit sequentially determines prediction information and the compensation unit generates a prediction pixel for each pixel constituting each of the small regions obtained by the division in the selected division type. An image encoding apparatus characterized in that when encoding is performed, information of a division type selected for each unit block is encoded, and a quantized value and prediction information are encoded for each small region.   Each division type in the set is that the small areas are all rectangular, the total number of small areas is less than or equal to a predetermined upper limit, the maximum size or the minimum size of the small areas is greater than or equal to a predetermined size, 2. The image encoding apparatus according to claim 1, wherein the image encoding apparatus satisfies that the number of small areas having a minimum size is not more than a predetermined number.   The adaptive division means temporarily selects each division type in the set for each unit block, and determines prediction information so that the prediction means minimizes the coding cost for each small region divided in the temporarily selected division type. The image coding apparatus according to claim 1, wherein an encoding cost as a whole of the unit block is obtained and the division type that minimizes the coding cost is selected. The adaptive division unit includes a division type set holding unit, and the division type set holding unit holds a plurality of sets of the division types for each unit block size, so that the adaptive division unit stores a unit block to be encoded. Select the division type from the set according to the size,
The image coding apparatus according to any one of claims 1 to 3, wherein a set having a larger unit block size in the plurality of retained sets has a larger number of division types belonging to the set. The adaptive division means includes an encoded channel information storage unit, and the encoded channel information storage unit includes information on the selected division type and the correspondence in the pixel of the first channel encoded for each unit block. By storing the prediction information for each small region to be performed, the image encoding device applies the stored division type and prediction information to the pixel of the second channel before encoding in the unit block. Find the quantization value for each region,
When encoding the pixels of the first and second channels, the division type information and the prediction information shared by the storage between the first and second channels, the first and second channels, 5. The image encoding device according to claim 1, wherein the quantized value of each channel is encoded.   The adaptive division means includes a division type identification number assigning unit, and the division type identification number assignment unit is a ratio in which each division type is selected when a predetermined number of unit blocks are encoded in advance by the image encoding device. 6. The image according to claim 1, wherein as the identification number used for the information of the division type for encoding, a smaller identification number is assigned to a division type with a higher ratio. Encoding device. The prediction information is information that specifies a prediction mode for predicting a small area to be encoded from the surrounding encoded pixel areas,
The adaptive division means includes a prediction information identification number assigning unit, and the prediction information identification number assignment unit is a small region in all division types belonging to the set when a predetermined number of unit blocks are encoded in advance by the encoding device. Based on the ratio of each prediction mode determined for each shape of the prediction mode as the prediction information for encoding, a smaller identification number for a prediction mode with a larger ratio determined for each shape of the small region The image encoding device according to claim 1, wherein the image encoding device is provided.   The compensation means includes a pixel unit repetitive compensation unit, and the pixel unit repetitive compensation unit reaches a prediction target pixel from an encoded pixel used for prediction determined by prediction information when generating a prediction pixel for each small region. A prediction pixel is generated in units of pixels by sequentially applying a predetermined weighted average to a coded pixel and / or a prediction pixel already generated in a predetermined vicinity position of a prediction pixel to be generated sequentially in the direction. The image encoding apparatus according to claim 1, wherein the already generated prediction pixel is combined with the encoded pixel and used for generation of a subsequent prediction pixel. An image decoding device for the image encoding device according to any one of claims 1 to 8,
Decoding means for decoding the information of the division type selected for each encoded unit block and the quantization value and prediction information for each small area;
Inverse quantization means for inversely quantizing the decoded quantized value to obtain a decoded transform coefficient, and inverse transform means for inversely transforming the decoded transform coefficient to obtain a decoded prediction residual;
From the decoded division type information, decoding-side adaptive division means for resetting a small area determined by the division type to a unit block to be decoded;
Compensation means for generating a decoded prediction pixel for each of the reset small regions based on the decoded pixel and the decoded prediction information,
An image decoding apparatus, wherein a decoded pixel is obtained by adding the decoded prediction residual and the decoded prediction pixel for each reset small region.