JP4529615B2 - Encoding apparatus, encoding method, encoding method program, and recording medium recording the encoding method program - Google Patents

Description

  The present invention relates to an encoding apparatus, an encoding method, an encoding method program, and a recording medium on which the encoding method program is recorded, and is applied to a video camera, an electronic still camera, a monitoring apparatus, and the like that record imaging results of moving images. be able to. The present invention detects the optimum mode for each macroblock from a plurality of intra prediction modes and a plurality of inter prediction modes by comparing the cost values with the cost function indicating the coding efficiency. By detecting the optimal mode by correcting the cost value by the activity, image quality degradation in the low activity area when the optimal mode is selected from the intra prediction mode and the inter prediction mode by the cost function and the image data is encoded. To be able to prevent.

  In recent years, in the transmission and recording of moving images related to broadcasting stations, general homes, etc., devices that efficiently transmit and store image data by effectively using the redundancy of image data are becoming popular. For example, in accordance with a method such as MPEG (Moving Picture Experts Group), image data is compressed by orthogonal transform such as discrete cosine transform and motion compensation.

  Here, MPEG2 (ISO / IEC 13818-2), which is one of such systems, is a system defined as a general-purpose image coding system so that it can handle both the interlace scanning system and the progressive scanning system. In addition, it is defined so as to be compatible with both standard resolution images and high-definition images, and is now widely used in a wide range of applications for professional use and consumer use. Specifically, according to MPEG2, for example, a standard resolution of 720 × 480 pixels and interlaced scanning image data are compressed to a bit rate of 4 to 8 Mbps, and a high resolution of 1920 × 1088 pixels and interlaced scanning are used. The image data of the system can be compressed to a bit rate of 18 to 22 [Mbps], and a high compression rate can be ensured with high image quality.

  However, MPEG2 is a high-quality encoding system suitable for broadcasting, and does not support a high compression rate encoding system with a smaller code amount than MPEG1. On the other hand, with the spread of portable terminals in recent years, it is expected that there will be an increasing need for an encoding method with a high compression rate with a smaller code amount than MPEG1. For this reason, an MPEG-4 encoding system standard was approved in December 1998 by ISO / IEC (International Organization for Standardization / International Electrotechnical Commission) 14496-2.

  In such a system, the standardization of H26L (ITU-T Q6 / 16 VCEG), which was originally intended for video coding for video conferencing, has progressed, compared to MPEG2 and MPEG4. Although the amount of computation increases, it becomes possible to secure higher encoding efficiency compared to MPEG2 and MPEG4. As part of MPEG4 activities, various functions are incorporated based on this H26L, and the encoding efficiency is even higher. Standardization of coding schemes to ensure the image quality is being promoted as Joint Model of Enhanced-Compression Video Coding. In these schemes, in March 2003, H264 and MPEG-4 Part 10 (AVC: Advanced Video Coding) Was set as an international standard.

  Here, FIG. 3 is a block diagram showing an encoding device based on this AVC. The encoding device 1 selects an optimal prediction mode from a plurality of intra prediction modes and a plurality of inter prediction modes, generates a difference data by subtracting a prediction value based on the selected prediction mode from image data, The difference data is subjected to orthogonal transform processing, quantization processing, and variable length coding processing, whereby the image data is coded by intra coding and inter coding.

  That is, in this encoding apparatus 1, the analog-digital conversion circuit (A / D) 2 performs analog-digital conversion processing on the video signal SV and outputs image data D1. The screen rearrangement buffer 3 receives the image data D1 output from the analog-digital conversion circuit 2, and the image data D1 according to the GOP (Group of Pictures) structure related to the encoding process of the encoding device 1. Sort and output the frames.

  The subtraction circuit 4 receives the image data D1 output from the screen rearrangement buffer 3, and generates and outputs difference data D2 from the prediction value generated by the intra prediction circuit 5 in intra coding. On the other hand, in inter coding, difference data D2 from the prediction value generated by the motion prediction / compensation circuit 6 is generated and output. The orthogonal transformation circuit 7 receives the output data D2 of the subtraction circuit 4, performs orthogonal transformation processing such as discrete cosine transformation and Karhunen-Labe transformation, and outputs transformation coefficient data D3 based on the processing result.

  The quantization circuit 8 quantizes the transform coefficient data D3 by the quantization scale by the rate control of the rate control circuit 9 and outputs it. The lossless encoding circuit 10 performs lossless encoding processing on the output data of the quantization circuit 8 by variable length encoding, arithmetic encoding, or the like, and outputs the result. Further, the lossless encoding circuit 10 acquires information on the intra prediction mode related to intra encoding, information about a motion vector related to inter encoding, and the like from the intra prediction circuit 5 and the motion prediction / compensation circuit 6, and outputs these information. Set in the header information of data D4 and output.

  The accumulation buffer 11 accumulates the output data D4 of the lossless encoding circuit 10 and outputs it at the transmission rate of the subsequent transmission path. The rate control circuit 9 monitors the amount of code generated by the encoding process by monitoring the free capacity of the storage buffer 11, and switches the quantization scale in the quantization circuit 8 based on the monitoring result, whereby the encoding device 1 Controls the amount of generated code.

  The inverse quantization circuit 13 performs inverse quantization processing on the output data of the quantization circuit 8, thereby reproducing the input data of the quantization circuit 8. The inverse orthogonal transform circuit 14 performs inverse orthogonal transform processing on the output data of the inverse quantization circuit 13, thereby reproducing the input data of the orthogonal transform circuit 7. The deblocking filter 15 removes block distortion from the output data of the inverse orthogonal transform circuit 14 and outputs the result. The frame memory 16 appropriately adds a prediction value generated by the intra prediction circuit 5 or the motion prediction / compensation circuit 6 to the output data of the deblocking filter 15 and records it as reference image information.

  Therefore, the motion prediction / compensation circuit 6 detects the motion vector of the image data output from the screen rearrangement buffer 3 from the prediction frame based on the reference image information held in the frame memory 16 in the inter coding. Based on the detected motion vector, the reference image information held in the frame memory 16 is motion compensated to generate predicted image information, and a predicted value based on the predicted image information is output to the subtraction circuit 4.

  In the intra coding, the intra prediction circuit 5 determines the intra prediction mode based on the reference image information stored in the frame memory 16, and generates and subtracts a predicted value of the predicted image information from the reference image information based on the determination result. Output to circuit 4.

  Accordingly, in this encoding method, difference data D2 by motion compensation related to inter prediction and difference data D2 by intra prediction are generated by inter encoding and intra encoding, respectively, and the difference data D2 is subjected to orthogonal transform processing. , Quantization processing, variable length coding processing, and transmission.

  FIG. 4 is a block diagram showing a decoding apparatus that decodes the encoded data D4 encoded in this way. In the decoding device 20, the accumulation buffer 21 temporarily accumulates and outputs the encoded data D4 input via the transmission path. The lossless decoding circuit 22 decodes the output data of the storage buffer 21 by variable length decoding, arithmetic decoding, etc., and reproduces the input data of the lossless encoding circuit 10 in the encoding device 1. At this time, if the output data is intra-coded, the intra-prediction mode information stored in the header is decoded and transmitted to the intra-prediction circuit 23, whereas the output data is inter-coded. If it is, the information on the motion vector stored in the header is decoded and transferred to the motion prediction / compensation circuit 24.

  The inverse quantization circuit 25 performs inverse quantization processing on the output data of the lossless decoding circuit 22, thereby reproducing the transform coefficient data D <b> 3 input to the quantization circuit 8 of the encoding device 1. The inverse orthogonal transform circuit 26 receives the transform coefficient data output from the inverse quantization circuit 25 and executes a fourth-order inverse orthogonal transform process, whereby the difference input to the orthogonal transform circuit 7 of the encoding device 1. Data D2 is reproduced.

  The adder 27 receives the difference data D2 output from the inverse orthogonal transform circuit 26, adds the predicted value based on the prediction image generated by the intra prediction circuit 23 in intra coding, and outputs the result. In the conversion, the predicted value based on the predicted image output from the motion prediction / compensation circuit 24 is added and output. As a result, the adder 27 reproduces the input data of the subtraction circuit 4 in the encoding device 1.

  The deblock filter 28 removes block distortion from the output data of the adder 27 and outputs the result. The screen rearrangement buffer 29 arranges the frames of the image data output from the deblock filter 28 according to the GOP structure. Change the output. A digital / analog conversion circuit (D / A) 30 performs a digital / analog conversion process on the output data of the screen rearrangement buffer 29 and outputs the result.

  The frame memory 31 records and holds the output data of the deblock filter 28 as reference image information. In inter coding, the motion prediction / compensation circuit 24 performs motion compensation on the reference image information held in the frame memory 31 based on the motion vector information notified from the lossless decoding circuit 22, and generates a predicted value based on the predicted image. The predicted value is output to the adder 27. Also, the intra prediction circuit 23 generates a prediction value based on the prediction image from the reference image information held in the frame memory 31 by the intra prediction mode notified from the lossless decoding circuit 22 in the intra coding, and adds this prediction value. To the device 27.

  In the AVC encoding process by such a series of processes, as shown in FIG. 5, one macro block is formed by 16 × 16 pixels in the luminance signal Y, whereas the color difference signals Cr, Cb is formed by 8 × 8 pixels, and each macroblock is processed as a unit. That is, these macroblocks are divided into small blocks of 4 × 4 pixels indicated by numerals 0 to 25, and difference data D2 is subjected to orthogonal transform processing and quantization processing for each small block.

  In this process, the chrominance signals Cr and Cb are obtained by collecting DC components for each macroblock from the coefficients obtained by the orthogonal transform process to form a 2 × 2 matrix. Quantized. In addition, in the case of the intra 16 × 16 prediction mode described later, the luminance signal Y is obtained by collecting DC components for each macroblock from coefficients obtained by orthogonal transform processing to form a 4 × 4 matrix. After Hadamard transform processing, quantization processing is performed.

  For intra coding related to such coding processing, an intra 4 × 4 prediction mode and an intra 16 × 16 prediction mode are prepared for luminance signal processing. In the AVC, as described above, the difference data D2 is subjected to orthogonal transform processing in units of 4 × 4 pixels, and the intra 4 × 4 prediction mode generates prediction values related to intra prediction in units of blocks of the orthogonal transform processing. It is a mode to do. On the other hand, the intra 16 × 16 prediction mode is a mode for generating a prediction value related to intra prediction in units of a plurality of blocks of the orthogonal transform process, and the plurality of 4 × 4 in the horizontal direction and the vertical direction, respectively. Set to

  Among them, in the intra 4 × 4 prediction mode, as shown in FIG. 6, a part of the neighboring 13 pixels A to M has a predicted value for a block of 4 × 4 pixels a to p that generate a predicted value. It is set to a prediction pixel to be used for generation, and a prediction value is generated from this prediction pixel. Here, the thirteen pixels A to M are the four pixels A to D adjacent in the vertical direction in the scanning start side of the block, and the pixels on the scanning end side of the four pixels A to D. Among four pixels E to F following D, four pixels I to L adjacent in the horizontal direction on the scanning start end side of this block, and four pixels I to L adjacent in the horizontal direction The pixel M is located above the pixel I on the scanning start end side.

  In the intra 4 × 4 prediction mode, as shown in FIG. 7 and FIG. 8, depending on the relative relationship between the 13 predicted pixels A to M and the 4 × 4 pixels a to p used for generating a predicted value. In addition, prediction modes of mode 0 to mode 8 are defined. That is, as shown in FIG. 6, for example, in modes 0 and 1, among the 13 prediction pixels A to M used to generate a prediction value, prediction pixels A to D and I to I that are adjacent in the vertical direction and the horizontal direction, respectively. A predicted value is generated by L.

  More specifically, as indicated by an arrow in FIG. 9A, mode 0 is a mode for generating a prediction value from prediction pixels A to D adjacent in the vertical direction. Among the 4 × 4 pixels a to p that generate values, the pixels a, e, i, and m in the first column that are continuous in the vertical direction have the pixel A in the upper direction set as a predicted pixel. In the subsequent pixels b, f, j, and n in the second column, the pixel B in the upper direction is set as a predicted pixel, and the pixels c, g, k, o, and d, h in the subsequent third and fourth columns are set. , L, and p, the upper pixels C and D are set as predicted pixels, and the pixel values of the predicted pixels A to D are set to the predicted values of the corresponding pixels a to p, respectively. Note that mode 0 is applied only when the prediction pixels A to D in this mode are significant.

  As shown in FIG. 9B, mode 1 is a mode for generating a prediction value from the prediction pixels I to L adjacent in the horizontal direction. As shown by the following equation, 4 × 4 for generating a prediction value. Among the pixels a to p, in the pixels a to d on one line continuous in the horizontal direction, the pixel I on the left side is set as the prediction pixel. The pixels e to h in the subsequent second line are set to the pixel J on the left side, and the pixels i to l and mp in the third line and the fourth line are respectively set to the left pixel K. And L are set as prediction pixels, and the pixel values of these prediction pixels I to L are set to the prediction values of the corresponding pixels a to p, respectively. Note that mode 1 is applied only when the prediction pixels I to L in this mode are significant.

  On the other hand, mode 2 is predicted from pixels A to D and I to L adjacent in the vertical and horizontal directions of the block among the 13 prediction pixels A to M, as shown in FIG. This is a mode for generating values, and when these pixels A to D and I to L are all significant, predicted values of the pixels a to p are generated according to the following equations.

  In mode 2, when all of the pixels A to D are not significant, the predicted value is generated by the equation (4). When all of the pixels I to L are not significant, the predicted value is generated by the equation (5). If the pixels A to D and I to L are not all significant, the predicted value is set to the value 128.

  On the other hand, as shown in FIG. 9D, mode 3 is a mode in which predicted values are generated from pixels A to H that are continuous in the horizontal direction among the 13 predicted pixels A to M. This is applied only when the pixels A to D and the pixels I to M among the pixels A to H are significant, and the predicted values of the pixels a to p are generated by the following equation.

  On the other hand, in the mode 4, as shown in FIG. 9E, among the 13 prediction pixels A to M, the pixels A to D and I to N adjacent to the block of 4 × 4 pixels ap are used. This is a mode for generating a predicted value by M, and is applied only when these pixels A to D and I to M are all significant, and the predicted value of each pixel a to p is generated by the following equation.

  On the other hand, in the mode 5, as shown in FIG. 9F, among the 13 prediction pixels A to M, the pixels A to D and I to N adjacent to the block of 4 × 4 pixels ap are used. This is a mode for generating predicted values by K and M, and is applied only when the predicted pixels A to D and I to M are all significant, and the predicted values of the pixels a to p are generated by the following equations.

  On the other hand, in the mode 6, as shown in FIG. 9G, among the 13 prediction pixels A to M, the pixels A to C, I to C adjacent to the block of 4 × 4 pixels a to p are arranged. M is a mode for generating a prediction value, and is applied only when the prediction pixels A to D and I to M are all significant, and the prediction value of each pixel a to p is generated by the following equation.

  On the other hand, in the mode 7, as shown in FIG. 9H, among the 13 prediction pixels A to M, four pixels A adjacent above the block of 4 × 4 pixels ap. To D and the four pixels E to G following the four pixels A to D are modes for generating a prediction value, and the pixels A to D and the pixels I to M are all significant. The prediction value of each pixel ap is generated by the following equation.

  On the other hand, in the mode 8, as shown in FIG. 9I, four pixels I to L adjacent to the left of the block of 4 × 4 pixels among the 13 prediction pixels A to M. Is a mode for generating predicted values, and is applied only when the pixels A to D and the pixels I to M are all significant, and the predicted values of the pixels a to p are generated by the following equations.

  In such an AVC, in the encoding process using the intra 4 × 4 prediction mode, the prediction mode is notified to the transmission target by effectively using the process based on the raster scanning order. That is, in the case of intra-encoding in the 4 × 4 prediction mode, as shown in FIG. The prediction modes Intra 4x4 pred mode A and Intra 4x4 pred mode B of adjacent blocks A and B have high correlation. Thus, the most probable prediction mode Most Probable Mode is defined by the following equation using the prediction modes Intra 4x4 pred mode A and Intra 4x4 pred mode B of the adjacent blocks A and B. The determination of min in this equation (12) is executed by the code mode number used for transmission of these prediction modes, and the prediction mode with the smaller value of the code mode number is set to the most likely prediction mode Most Probable Mode. To do.

In addition, in the bitstream, as a parameter related to the block of 4 × 4 pixels, a flag prev intra 4x4 pred mode flag [luma 4x4 BlkIdx] indicating whether or not the prediction mode is transmitted and a prediction mode rem intra 4x4 pred mode [luma 4x4 BlkIdx] is defined, and the decoding side processes these two parameters as shown in FIG. 11 according to the description in C language, and detects the prediction mode Intra 4x4 pred mode C of the block C to be processed. Here [luma 4x4
BlkIdx] is a block number that identifies a target block related to luminance data.

  That is, in this case, when the flag prev intra 4x4 pred mode flag [luma 4x4BlkIdx] indicating whether or not the prediction mode is transmitted is set, the prediction modes Intra 4x4 pred mode A and Intra 4x4 pred mode B of the adjacent blocks A and B are set. By using the equation (12), the most probable prediction mode Most Probable Mode detected on the decoding side is set as the prediction mode of the block C to be processed. Also, when this flag prev intra 4x4 pred mode flag [luma 4x4 BlkIdx] is not set, the prediction mode transmitted from the most probable prediction mode Most Probable Mode rem intra 4x4 pred mode [luma 4x4 BlkIdx] When the code mode number of is smaller, the transmitted prediction mode rem intra 4x4 pred mode [luma 4x4 BlkIdx] is set as the prediction mode of the processing target block C. Also, when the flag prev intra 4x4 pred mode flag [luma4x4 BlkIdx] is not set, the prediction mode transmitted from the most probable prediction mode Most Probable Mode rem intra 4x4 pred mode [luma 4x4 BlkIdx] code If the mode number is not small, the prediction mode of the code mode number obtained by adding the value 1 to the code mode number of the transmitted prediction mode rem intra 4x4 pred mode [luma 4x4 BlkIdx] is set as the prediction mode of the processing target block C.

  Accordingly, when the most probable prediction mode Most Probable Mode matches the prediction mode of the processing target block C, the encoding device 1 uses the flag prev intra 4x4 pred mode flag [luma 4x4 BlkIdx ] To stop transmission of the prediction mode rem intra 4x4 pred mode [luma 4x4 BlkIdx] and reduce the amount of data used for transmission.

  On the other hand, in the intra 16 × 16 prediction mode, as shown in FIG. 12, for a block B made up of 16 × 16 pixels P (0, 15) to P (15, 15) for generating a prediction value, Pixels P (0,15) to P (15,15) constituting this block, and pixels P (0, -1) to P (15, -1) and P adjacent above and to the left of this block B (-1, 0) to P (-1, 15) are set as prediction pixels, and a prediction value is generated by these prediction pixels.

  In the intra 16 × 16 prediction mode, as shown in FIG. 13, prediction modes of mode 0 to mode 3 are defined. Of these, mode 0 is a pixel P (0, −1) adjacent above the processing target block B. -P (15, -1) (P (x, -1); x, y = -1 to 15) is applied only when it is significant, and as shown by the following equation, each pixel constituting the block B Predicted values of P (0,15) to P (15,15) are generated. As a result, as shown in FIG. 14A, prediction of each pixel continuous in the vertical direction of the block B is performed by the pixel values of the pixels P (0, −1) to P (15, −1) adjacent to the block B. A value is generated.

  On the other hand, in the mode 1, the pixels P (−1, 0) to P (−1, 15) adjacent to the left side of the block B (P (−1, y); x, y = −1 to 15). Is applied only when is significant, and predicted values of the pixels P (0,15) to P (15,15) constituting the block B are generated as shown by the following equation, and FIG. ), Predicted values of pixels continuous in the horizontal direction of the block B are generated by the pixel values of the pixels P (−1, 0) to P (−1, 15) adjacent to the block B.

  On the other hand, in the mode 2, the pixels P (0, −1) to P (15, −1) and P (−1, 0) to P (−1, 15) adjacent to the upper side and the left side of the block B are used. Are all significant, a predicted value is obtained by the following equation, and as shown in FIG. 14C, these pixels P (0, -1) to P (15, -1) and P ( A predicted value of each pixel constituting the block B is generated based on the average value of the pixel values of (−1, 0) to P (−1, 15).

  In mode 2, pixels P (0, -1) to P (15, -1) and P (-1, 0) to P (-1, 15) adjacent to the upper and left sides of the block B are used. Among them, when the pixels P (−1, 0) to P (−1, 15) adjacent to the upper side are not significant, the predicted value of each pixel is calculated based on the average value of the adjacent pixels on the significant side by applying the equation (16). Is generated. Further, when the pixels P (−1, 0) to P (−1, 15) adjacent to the left are not significant, the equation (17) is applied. In this case as well, the block B is determined by the average value of the adjacent pixels on the significant side. The predicted value of each pixel that constitutes is generated. If all of the pixels P (0, −1) to P (15, −1) and P (−1, 0) to P (−1, 15) adjacent to the upper and left sides of the block B are not significant, A predicted value is set to the value 128.

  On the other hand, in the mode 3, the pixels P (0, −1) to P (15, −1) and P (−1, 0) to P (−1, 15) adjacent to the upper side and the left side of the block B are used. Is applied only when all are significant, and a predicted value is obtained by the following equation. As a result, as shown in FIG. 14D, a predicted value of each pixel is generated by a calculation process in an oblique direction.

  For various intra prediction modes related to such a luminance signal, the prediction mode is set for the color difference signal in the same manner as the intra 16 × 16 prediction mode for the luminance signal. However, while the intra 16 × 16 prediction mode is a macro block of 16 × 16 pixels, the intra prediction mode for the color difference signal is a macro block of 8 × 8 pixels, and is shown in FIG. As described above, the mode number and the corresponding prediction mode are different from those in the case of the luminance signal. The prediction mode is set independently for the luminance signal and the color difference signal.

  That is, in mode 0, when the pixel P (x, −1) and the pixel P (−1, y) are significant, a predicted value is generated by the following equation.

  When the pixel P (−1, y) is not significant, the predicted value is generated according to the equation (21) when the pixel P (x, −1) is not significant according to the equation (20).

  In mode 1, it is applied only when the pixel P (-1, y) is significant, and a predicted value is generated by the following equation.

  In mode 2, it is applied only when P (x, -1) is significant, and a predicted value is generated by the following equation.

  In mode 3, when the pixel P (x, -1) and the pixel P (-1, y) are significant, a predicted value is generated by the following equation.

  On the other hand, in inter coding, multiple reference frames are set so that motion compensation can be performed by selecting any of a plurality of reference frames Ref for the processing target frame Org as shown in FIG. Thus, even when the part corresponding to the motion compensation block is hidden in the immediately preceding frame, or even when the entire pixel value temporarily changes in the immediately preceding frame due to flash or the like, it is possible to achieve high accuracy. Data compensation efficiency is improved by motion compensation.

In the block relating to motion compensation, as shown in FIG. 17A1, motion compensation is performed with reference to a block of 16 pixels × 16 pixels.
MC-block size supports tree-structured motion compensation. As shown in FIGS. 17 (A2) to (A4), a block of 16 pixels × 16 pixels is divided into two in the horizontal direction and / or the vertical direction. , 16 pixels × 8 pixels, 8 pixels × 16 pixels, and 8 pixels × 8 pixels are set so that motion compensation can be performed by setting a motion vector and a reference frame independently. In addition, as shown in FIGS. 17B1 to 17B4, the sub-macro block having 8 pixels × 8 pixels has 8 pixels × 8 pixels, 8 pixels × 4 pixels, 4 pixels × 8 pixels, 4 pixels × 4 pixels. Are further divided into sub-macroblocks, and motion vectors and reference frames are set independently so that motion compensation can be performed.

  The motion compensation is set such that motion compensation can be performed with a 1/4 pixel accuracy using a 6-tap FIR filter. Accordingly, in FIG. 18, motion prediction is performed so that a pixel value of 1 pixel accuracy is indicated by the symbol A, a pixel value of 1/2 pixel accuracy is indicated by the symbols b to d, and a pixel value of 1/4 pixel accuracy is indicated by the symbols e1 to e3. The compensation circuit 6 performs horizontal calculation or vertical calculation by weighting each tap input of the 6-tap FIR filter with the values 1, -5, 20, 20, -5, and 1 and executing the following arithmetic processing. A pixel value b or d with 1/2 pixel accuracy is calculated between successive pixels.

  Also, using the pixel value b or d with 1/2 pixel accuracy calculated in this way, each tap input of the 6-tap FIR filter is weighted by the values 1, -5, 20, 20, -5, and 1. By executing the arithmetic processing of the following equation, a pixel value c with a ½ pixel accuracy between consecutive pixels in the horizontal direction and the vertical direction is calculated.

  Further, the pixel values e1 to e3 with the ¼ pixel accuracy are calculated by executing the following arithmetic processing by linear interpolation using the pixel values b to d with the ½ pixel accuracy calculated in this way. . Note that in the normalization processing related to the weighted addition of the equations (25) and (26), all the interpolation processes in the vertical direction and the horizontal direction are completed and executed.

  In contrast to such motion compensation processing for luminance signals, motion compensation for color difference signals is executed by linear interpolation. That is, as shown in FIG. 19, the adjacent pixels A to D with the pixel pitch s are set to sampling points with internal division ratios dx, s-dx, dy, and s-dy in the horizontal and vertical directions, respectively. The pixel value ν is expressed by the following equation.

  In AVC, even for a motion vector that is coding information related to such inter prediction, the data transmission amount is reduced by effectively using the correlation between successive macroblocks and sub-macroblocks. That is, in AVC coding, one macroblock can be divided into a plurality of sub-macroblocks and motion compensation can be performed, thereby increasing the amount of code used for motion vector transmission. Therefore, a motion vector prediction value pmv is generated for each horizontal component and vertical component for each block, and the motion vector prediction value pmv and the actual motion vector mv are calculated by the arithmetic processing represented by the following equation. The difference vector motion vector information MVD (Motion Vector Data) is encoded and transmitted.

  However, as shown in FIG. 20A, the block related to the motion vector mv is the right sub-macroblock C of the two sub-macroblocks formed by dividing one macroblock into two in the horizontal direction. If the reference frame refIdxE related to the detection of the motion vector prediction value mv is equal to the reference frame refIdxA of the remaining left submacroblock A, the submacroblock adjacent to the left side as shown by the following equation The motion vector mvA detected in A is set as the motion vector prediction value pmv.

  On the other hand, when the block related to the motion vector mv is the left sub-macroblock A, the reference frame refIdxE related to the detection of the motion vector prediction value mv is adjacent to the remaining left sub-macroblock C. If it is equal to the reference frame refIdxC, the motion vector mvC detected in the sub-macroblock C adjacent to the right side is set to the motion vector prediction value pmv as shown by the following equation.

  Further, as shown in FIG. 20B, the block related to the motion vector mv is an upper sub-macroblock C of two sub-macroblocks formed by dividing one macroblock into two in the vertical direction. When the reference frame refIdxE related to the detection of the motion vector prediction value mv is equal to the reference frame refIdxA of the remaining lower submacroblock B, as shown by the following equation, The motion vector mvB detected in the block B is set to the motion vector prediction value pmv.

  On the contrary, when the block related to the motion vector mv is the lower sub-macroblock B, the reference frame refIdxE related to the detection of the motion vector prediction value mv is the remaining adjacent sub-macroblock. When equal to the reference frame refIdxA of A, as shown by the following equation, the motion vector mvA detected in the sub macroblock A adjacent on the lower side is set to the motion vector prediction value pmv.

  In other cases, as shown in FIG. 21A, a motion vector prediction value pmv is generated from a motion vector detected in an adjacent block for a block E related to motion correction. Here, this adjacent block includes a block A adjacent to the horizontal scanning start side according to the raster scanning order, a block B adjacent to the vertical scanning start side according to the raster scanning order, and left and right blocks C of this block, D. Note that the motion vector prediction value pmv by these adjacent blocks is also applied to the motion vector detected by the sub-macroblock belonging to this adjacent block, as shown in FIG.

  Specifically, when there is an adjacent block whose reference frame matches with the block E related to motion correction based on the values of the reference frame indexes refIdxA, refIdxB, and refIdxC related to detection of each adjacent block, A motion vector mvN by an adjacent block (N = A or B or C) having a matching reference frame is set as a motion vector prediction value pmv.

  In other cases, for each component in the vertical direction and the horizontal direction, the component based on the processing result by the median filter is set as each component of the motion vector prediction value pmv by the following equation.

  However, when either the block B adjacent in the vertical direction or the block C following the block B is not significant and the block A adjacent in the horizontal direction is significant, the adjacent block B in the vertical direction The motion vector mvA and the reference frame index refIdx of the block A are substituted for the motion vector mv and the reference frame index refIdx of C and C, as shown by the following equation.

  In the AVC, the B picture has a direct mode of a temporal (temporal) direct mode and a spatial (spatial) direct mode. In this direct mode, transmission of information on motion vectors is stopped to improve encoding efficiency. improves.

That is, in the spatial direct mode, the decoding process is executed with the prediction vector pmv set as a motion vector. On the other hand, in the temporal direct mode, assuming that the motion is linear, as shown in FIG. 22, the motion vector mvcol of the corresponding block (Co-Located Block) of the prediction frame L1 that has completed the encoding process is obtained. by linear interpolation using, is to create a motion vector MV _l0 and MV _l1 according to B-picture to be processed. In the AVC image compression information, since the parameter TD relating to the time information between these pictures L0 and L1 does not exist, POC (Picture Order Count) is used instead.

  With respect to the prediction modes related to intra and inter prediction, AVC uses High Model Mode assuming multi-pass encoding and Low Complexity Mode assuming one-pass encoding by the Joint Model (AVC reference encoding method) related to AVC. Are defined, and an optimum mode is selected according to these definitions, and the encoding process is executed. Of these modes, in Low Complexity Mode, a cost function indicating encoding efficiency is defined by the following equation, and an optimal mode is detected by comparing cost values Cost (Mode) obtained by the cost function.

  Here, SA (T) D is an error value between the original image and the predicted image, and an absolute value error sum of pixel value difference values between the original image and the predicted image is applied. SA (T) D0 is a header bit and a cost as a weight for mode determination, an offset value given to the error value SA (T) D, and is used for transmission of additional information such as a motion vector. The amount of data is shown.

  Specifically, the absolute value error sum SAD is expressed by the following equation for each macroblock, and the difference value between the original image and the predicted image in each prediction mode Mode is applied.

  Here, instead of the absolute value error sum SAD according to the equation (38), a difference addition value SATD (Mode) obtained by the following equation may be used.

Hadamard () is a Hadamard transform operation for multiplying the target matrix by a Hadamard transform matrix, as shown by the following equation. Note Hadamard transform matrix is expressed by equation (41), H ^T is a transposed matrix of the Hadamard transform matrix.

  The offset value SA (T) D0 is represented by the following equation in the previous prediction mode. Here, QP0 (QP) is a function for converting the quantization parameter QP into a quantization scale, MFDFW is a motion vector related to the previous prediction, and Bit to code is a bit stream related to this motion vector. Code amount.

  The offset value SA (T) D0 is expressed by the following equation in the post-prediction mode. Here, MVDBW is a motion vector related to post prediction.

  The offset value SA (T) D0 is expressed by the following equation in the Bi-Predictive prediction mode. Here, Bit to code forward Blk size and Bit to code backward Blk size are code amounts on the bitstream necessary for transmission of information related to the motion compensation block related to the forward prediction and backward prediction, respectively.

  In the direct mode, the offset value SA (T) D0 is obtained by the following equation.

  In the intra 4 × 4 prediction mode, the offset value SA (T) D0 is obtained by the following equation.

  Incidentally, this cost function is also applied to a search for a motion vector, and a motion vector that minimizes the cost value Cost is detected as shown by the following equation.

  Accordingly, when the optimum mode is detected in the Low Complexity Mode, the encoding device 1 uses the luminance signal in the intra prediction circuit 5 and the motion prediction / compensation circuit 6, respectively, to perform all of intra coding and inter coding. The cost value Cost of the prediction mode is calculated, the prediction mode having the smallest cost value Cost is selected, and the optimum mode for intra coding and the optimum mode for inter coding are detected. In addition, by comparing the cost value Cost in the optimum mode for intra coding and the optimum mode for inter coding, intra coding and inter coding are selected, and the optimum mode of the luminance signal is detected. When intra coding is selected in this way, the cost value of each intra prediction mode is calculated for the color difference signal, and the intra prediction mode having the smallest value is set as the optimum mode of the color difference signal by comparing the cost values. When inter coding is selected, a predicted value of the color difference signal is generated by a reference frame related to the luminance signal, a motion vector, and a motion compensation block corresponding to the luminance signal.

  As a result, in AVC, the optimal mode is detected for each macroblock from a plurality of intra prediction modes and a plurality of inter prediction modes, and image data is encoded by this optimal mode, thereby efficiently encoding the image data. .

  In AVC, the deblocking filters 15 and 28 are applied to remove block distortion in a decoded image and prevent propagation of block distortion due to motion compensation processing, and are defined as follows. Here, as the quantization parameter QP, QPY is applied in the luminance signal processing, and QPC is applied in the color difference signal processing. In the deblocking filter processing, regarding adjacent pixels, even if pixel values belonging to different slices belong to the same picture, the processing is executed.

  Here, as shown in FIG. 23, for the pixels that are continuous with the block boundary in between, the pixel values before processing by the deblocking filter are p0 to p3 and q0 to q3, and the pixel values after processing are p0 ′ to p3. ', Q0' to q3 '. With respect to the pixel values to be processed, as shown in FIG. 24, block boundary strength values (Bs: Boundary Strength) are defined depending on whether or not each pixel belongs to an intra macroblock.

  On the premise of this definition, the deblocking filter process is executed when the relational expression expressed by the following expression holds.

  Here, the constants α and β are set by default as shown in FIG. 25 by the quantization parameter QP as shown by the following expression, but are included in the slice header in the image compression information as shown by the arrow A. The usage can be adjusted by the slice alpha c0 offset div2 and the slice beta offset div2. Note that FIG. 26 is a chart showing the settings of α and β. In FIG. 26, indexA and indexB are defined by the following equations, and the offset values Filter OffsetA and Filter OffsetB correspond to adjustments by the user.

  In AVC, when the block boundary intensity value Bs is 4 or less, pixel values p0 ′ and q0 ′ are set as shown by the following equations.

  Here, tc is set to a value according to equation (51) when the chroma edge flag is 0, and is set to a value according to equation (52) otherwise. Tco is defined by indexA, indexB, and Bs as shown in FIG.

  Ap and aq are expressed by the following equations.

  On the other hand, the pixel value p1 ′ is set to a value according to the equation (54) when the chroma edge flag is 0 and the value of ap is less than or equal to b, otherwise, The value is set according to equation (55).

  The pixel value q1 ′ is set to a value according to the expression (56) when the chroma edge flag is 0 and the value of aq is less than or equal to b, and in other cases, (57) Set to an expression value.

  The pixel values p2 ′ and q2 ′ are set to the pixel values p2 and q2 before processing, as shown by the following equation.

  On the other hand, when the block boundary intensity value Bs is 4, the processed pixel value pi ′ (i = 0 to 2) is the case where the chroma edge flag is 0, and When the relational expression is established, it is set to a value indicated by the expression (60).

  If this condition is not met, the value is set according to the following equation.

  When the block boundary intensity value Bs is 4, the processed pixel value qi ′ (i = 0 to 2) is a case where the chroma edge flag is 0, and the relationship of the following equation: When the formula is established, the value is set to a value indicated by the formula (63).

  If this condition is not met, the value is set according to the following equation.

  In the AVC encoding device 1 and decoding device 20, the deblocking filter 28 switches the characteristics accordingly to prevent the occurrence of block distortion.

  On the other hand, in rate control, for example, a technique based on TM5 (MPEG-2 Test Model 5) is applied. Here, the rate control by TM5 has three steps: a bit allocation step for setting a target code amount to each picture, a rate control step using virtual buffer control, and an adaptive quantization step considering visual characteristics. Consists of hierarchies.

  Of these steps, in the bit allocation step, the target code amount for a picture that has not yet been encoded is calculated from the bit amount allocated to 1 GOP and the code amount generated so far, and based on the following two assumptions: Thus, the code amount allocation amount for each picture is calculated.

  Here, the first assumption is that the product of the average quantization scale used when encoding each picture and the generated code amount is a constant value for each picture type unless the screen changes. . Thus, in this rate control, after encoding each picture, parameters Xi, Xp, and Xb (global complexity measure) representing the complexity of the screen are updated by the following equation for each picture type. Thereby, in rate control by TM5, the relationship between the quantization scale code and the generated code amount when the next picture is encoded is estimated by using these parameters Xi, Xp, and Xb.

  Here, the subscript of each variable in the expression (65) is a subscript indicating an I picture, a P picture, and a B picture, respectively. Si, Sp, and Sb are generated code bit amounts by the encoding process of each picture, and Qi, Qp, and Qb are average quantization scale codes at the time of encoding of each picture. The initial values of the parameters Xi, Xp, and Xb are given by the following equation using the target code amount bit rate [bit / sec].

  The second assumption is that the ratio Kp of the quantization scale code of the P picture to the quantization scale of the I picture and the ratio Kb of the quantization scale code of the B picture to the quantization scale of the I picture are held in the relationship of the following equations: It is assumed that the overall image quality is always best when

  In other words, this assumption means that the overall picture quality is best when the quantization scale of the B picture is always set to 1.4 times the quantization scale of the I picture and the P picture. Compared to pictures and P pictures, the B picture is roughly quantized to save the code amount assigned to the B picture, and accordingly, a large amount of code amount is allocated to the I picture and P picture to improve the image quality of the I picture and P picture. In addition to the improvement, the image quality of the B picture referring to the I picture and the P picture is also improved, and the overall image quality is thereby optimized.

  As a result, TM5 calculates the allocated bit amounts Ti, Tp, and Tb for each picture by the following arithmetic processing. Here, Np and Nb are the number of P pictures and B pictures that have not been encoded in the GOP to be processed.

  Thereby, TM5 estimates the generated code amount of each picture based on the above two assumptions. At this time, for a picture of a picture type different from the code allocation target, it is estimated how many times the code amount generated by the picture is larger than the generated code amount of the allocation target picture under the image quality optimization condition. Also, by this estimation, it is estimated how many pictures in the picture type to be code assigned correspond to the uncoded pictures in the GOP, and the allocated bit amount to each picture is calculated from this estimation result. In this case, the rate control circuit 9 considers the fixedly required code amount such as a header and sets the lower limit to the value and calculates the allocated bit amount.

  On the other hand, in the subsequent rate control step, the allocated bit amounts Ti, Tp, Tb obtained in the bit allocation step and the actual generated code amount are matched with each other, so that each picture type is independent. The three types of virtual buffers are set in the virtual buffer, and the quantization scale of the quantization circuit 8 is calculated by feedback control in units of macroblocks based on the capacity of the virtual buffer.

  First, the occupation ratios of these three types of virtual buffers are calculated by the following formula. Here, d0i, d0p, and d0b are initial occupancy amounts of each virtual buffer, Bj is the generated bit amount from the beginning of the picture to the jth macroblock, and MB_cnt is the number of macroblocks in one picture.

  The quantization scale for the jth macroblock is calculated by the following equation based on the calculation result by the equation (69).

  Here, r is a reaction parameter, which is a parameter for controlling a feedback response. In TM5, the reaction parameter r and the initial values d0i, d0p, d0b are given by the following equations.

  The initial value of the virtual buffer at the beginning of the sequence is given by the following equation.

  In the subsequent adaptive quantization step, the quantization scale calculated in the rate control step is corrected in consideration of the visual characteristics, thereby executing an optimal quantization process in consideration of the visual characteristics. Here, in this optimal quantization process, in order to quantize more finely in the flat part where deterioration is visually noticeable, and coarser in the complicated part of the pattern where deterioration is relatively inconspicuous, The quantization scale is corrected by the activity indicating the flatness of each macroblock.

  Here, for each macroblock having a size of 16 × 16 pixels, the activity is divided into four blocks in the frame DCT mode and four blocks in the field DCT mode for four blocks of 8 × 8 pixels constituting the macroblock. The pixel values of a total of eight blocks are calculated using the following equation, thereby indicating the smoothness of the luminance level in the corresponding macroblock.

  Here, Pk is the pixel value in the luminance signal block of the original image. The reason why the minimum value is taken in the equation (73) is to prevent the image quality deterioration by making the quantization step finer when only a part of the macroblock has a flat part.

  In TM5, the activity obtained by this calculation formula is normalized by the following formula, thereby obtaining a normalized activity Nactj having a value in the range of 0.5-2. Here, avg_act is an average value of activity actj in the picture encoded immediately before.

  In addition, the normalization activity Nactj is used to execute the following arithmetic processing to correct the quantization scale Qj calculated in the rate control step.

  Thus, in the encoding apparatus 1, the rate control circuit 9 executes the rate control processing related to TM5 and sequentially encodes the image data D1.

  With regard to such an encoding apparatus, for example, various ideas for convenience of decoding processing and the like have been proposed in Japanese Patent Application Laid-Open No. 2004-56827.

  By the way, in the AVC, the error value SA (T) D due to the intra prediction mode is small in a region where the activity is relatively low. The cost value Cost (Mode) may be smaller in the prediction mode. Accordingly, in AVC, when an area with relatively low activity occupies a large portion in the second half of the image, such as the ground, the intra prediction mode is easily selected in the area with low activity even in the inter slice.

  However, in such an encoding process using the intra prediction mode in the low activity area, the coding is performed by switching the prediction mode in the nine types of intra 4 × 4 prediction modes described above with reference to FIGS. In this case, the switching of the prediction mode appears as if it fluctuates and appears in the decoded image. Here, such fluttering is easily seen by the viewer by being seen like flicker, and in this case, there is a problem that image quality is impaired in the conventional AVC.

Furthermore, in the AVC, as shown in FIG. 24, when the block boundary strength value Bs is increased in an intra macroblock, when a region with low activity is encoded in the intra prediction mode, a deblock filter is used. This excessively suppresses block boundary distortion. Here, in the suppression of the excessive block boundary distortion, the viewer recognizes that the resolution is remarkably lowered by losing the local change in the flat portion. This also has the problem that the image quality is impaired in AVC.
JP 2004-56827 A

  The present invention has been made in consideration of the above points. When image data is encoded by selecting an optimal mode from an intra prediction mode and an inter prediction mode by a cost function, image quality degradation in a low activity area is reduced. It is an object of the present invention to propose an encoding apparatus, an encoding method, an encoding method program, and a recording medium on which the encoding method program can be prevented.

  In order to solve such a problem, in the invention of claim 1, an optimal mode is detected for each macroblock from a plurality of intra prediction modes and a plurality of inter prediction modes by comparing cost values with a cost function indicating coding efficiency. Applied to an encoding device that encodes image data in the optimum mode, activity calculating means for calculating an activity indicating the flatness of the image by the image data for each macroblock, and the cost value by the activity And an optimum mode detecting means for detecting the optimum mode.

  In the invention of claim 11, an optimum mode is detected for each macroblock from a plurality of intra prediction modes and a plurality of inter prediction modes by comparing cost values with a cost function indicating coding efficiency, and an image is obtained by the optimum mode. Applying to an encoding method for encoding data, an activity calculation step for calculating an activity indicating the flatness of the image based on the image data for each macroblock, and correcting the cost value by the activity An optimum mode detecting step for detecting the optimum mode.

  Further, in the invention of claim 12, an optimum mode detection step of detecting an optimum mode for each macroblock from a plurality of intra prediction modes and a plurality of inter prediction modes by comparing cost values with a cost function indicating coding efficiency; And an encoding method program including an encoding process step for encoding image data in the optimum mode, and calculating an activity indicating the flatness of the image based on the image data for each macroblock. A step of calculating an activity, and the step of detecting the optimum mode detects the optimum mode by correcting the cost value by the activity.

  According to a thirteenth aspect of the present invention, the program of the encoding method is applied to a recording medium on which a program of the encoding method executed by the arithmetic processing means is recorded. By comparison, a plurality of intra prediction modes, a step of optimal mode detection for detecting an optimal mode for each macroblock from a plurality of inter prediction modes, a step of encoding processing for encoding image data in the optimal mode, An activity calculation step of calculating an activity indicating the flatness of the image based on the image data for each macroblock, and the step of detecting the optimum mode corrects the cost value by the activity and sets the optimum mode. To detect.

  According to the configuration of claim 1, an optimum mode is detected for each macroblock from a plurality of intra prediction modes and a plurality of inter prediction modes by comparing cost values using a cost function indicating coding efficiency, and image data is detected by the optimum mode. Applying to an encoding apparatus that performs encoding processing, activity calculating means for calculating an activity indicating the flatness of the image by the image data for each macroblock, and correcting the cost value by the activity to correct the optimal mode If an optimum mode detecting means for detecting the cost is provided, the cost value can be set so that the intra prediction mode is not selected when the activity is low due to the correction of the cost value. As a result, suppression of excessive block boundary distortion can be prevented to prevent resolution degradation, and image quality deterioration due to switching of a plurality of intra prediction modes can be prevented. When image data is encoded by selecting the optimum mode from the inter prediction mode, it is possible to prevent image quality deterioration in a low activity area.

  Thus, according to the configurations of claims 11, 12, and 13, when the optimal mode is selected from the intra prediction mode and the inter prediction mode by the cost function and the image data is encoded, the region with low activity Encoding method, encoding method program, and recording medium recording the encoding method program can be provided.

  According to the present invention, when image data is encoded by selecting an optimal mode from an intra prediction mode and an inter prediction mode by a cost function, it is possible to prevent image quality deterioration in a low activity area.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings as appropriate.

(1) Configuration of Embodiment FIG. 1 is a block diagram showing an AVC encoding apparatus according to an embodiment of the present invention. In this encoding device 41, the same components as those of the encoding device 1 described above with reference to FIG. 3 are denoted by the corresponding reference numerals, and redundant description is omitted. As a result, the encoding device 41 performs analog-digital conversion processing on the sequentially input video signal S1 and converts it into image data D1, and then selects an optimum mode from the intra prediction mode and the inter prediction mode to select the image data D1. Encode.

  In this encoding device 41, the activity calculation circuit 42 calculates a parameter indicating the flatness of the image based on the image data D1 for each macroblock of 16 × 16 pixels for the image data D1 to be processed. Now the activity is applied to this parameter. As a result, the activity calculation circuit 42 calculates the activity act by calculating the variance value of the pixel value for each macroblock by executing the arithmetic processing of the following equation. The activity calculation circuit 42 sets this activity act to the activity MB act of the macro block and outputs it.

  The rate control circuit 43 uses the activity MB act obtained from the activity calculation circuit 42 to execute rate control processing by the above-described TM5 method.

  The motion prediction / compensation circuit 44 calculates the cost value Cost (Mode) of the equation (37) for all prediction modes of the luminance signal related to inter coding under the control of the intra / inter determination circuit 45, and determines each prediction mode. The prediction mode with the smallest value is detected by comparing the cost value Cost (Mode). Thereby, the motion prediction / compensation circuit 44 detects the optimum mode from the inter prediction mode, and notifies the intra / inter determination circuit 45 of the cost value Cost (Mode) of this optimum mode. In addition, according to an instruction from the intra / inter determination circuit 45 obtained by this notification, in the case of inter coding processing, a prediction value in the optimum mode is generated for the luminance signal and the color difference signal and output to the subtraction circuit 4.

  Similarly, the intra prediction circuit 46 controls the error value SA between the original image and the prediction image described above with respect to the expression (37) for all prediction modes of the luminance signal related to intra coding under the control of the intra / inter determination circuit 45. (T) D and the offset value SA (T) D0 are calculated and notified to the intra / inter determination circuit 45. In addition, according to an instruction from the intra / inter determination circuit 45 obtained by this notification, a corresponding predicted value is output to the subtraction circuit 4 in the case of intra coding processing. In this case, the cost value is calculated for the color difference signal to detect the optimum mode, and the predicted value in the optimum mode is output to the subtraction circuit 4.

  The intra / inter determination circuit 45 calculates the cost value Cost (Mode) for all intra prediction modes using the error value SA (T) D and the offset value SA (T) D0 notified from the intra prediction circuit 46, The calculated cost value Cost (Mode) is compared with the cost value Cost (Mode) notified from the motion prediction / compensation circuit 44, and the prediction mode with the smallest cost value Cost (Mode) is detected. The intra / inter determination circuit 45 selects intra prediction and inter prediction for each macroblock by detecting the prediction mode, and detects an optimum mode for each macroblock from a plurality of intra prediction modes and a plurality of inter prediction modes. . The output of the predicted value in the detected optimum mode is instructed to the intra prediction circuit 46 and the motion prediction / compensation circuit 44.

  In this series of processing, the intra / inter determination circuit 45 corrects the cost value Cost (Mode) related to the intra 4 × 4 prediction mode based on the activity MB act obtained from the activity calculation circuit 42.

That is, as described above with respect to the equation (46), conventionally, the Low Complexity.
In Mode, the cost value Cost (Mode) in the intra 4 × 4 prediction mode is expressed by the following equation, and a conversion function QP0 (QP) from a quantization parameter related to additional information to a quantization value is added to a constant of value 24. An offset value SA (T) D0 is defined by multiplication, and the cost value Cost (Mode) is offset by this offset value SA (T) D0, thereby reducing the occurrence of intra macroblocks in the inter slice. Here, this value 24 is a value based on experience values.

  With this setting, when an intra macroblock is selected in a region with high activity, an intra prediction mode in a specific direction corresponding to a picture is appropriately selected, thereby ensuring coding efficiency, Image quality degradation can be effectively avoided. However, in the low activity region, the prediction mode is switched variously due to the influence of noise, and the block boundary distortion is excessively suppressed by the deblocking filter.

  Thereby, the intra / inter determination circuit 45 sets the offset value SA (T) D0 by the calculation processing of the following equation instead of the equation (77). Here, f (MB act) is a function having the activity MB act as a variable. Thereby, the intra / inter determination circuit 45 corrects the cost value so that it is difficult to select the intra prediction mode when the high-frequency component is small in the macro block by the activity MB act. More specifically, in this embodiment, a monotonically decreasing function is applied to the function f (MB act) having the activity MB act as a variable, thereby setting a larger offset value in a lower activity region. Therefore, it is difficult to select an intra macroblock. On the other hand, a small offset value is applied to a region with high activity to facilitate selection of an intra macroblock. For example, a function having a binary output value, a linear function, and various functions can be widely applied to the monotonically decreasing function.

  FIG. 2 is a flowchart showing a processing procedure of the encoding device 41 according to the optimum prediction mode. In the encoding device 41, this processing procedure is executed for each macroblock, the motion prediction process related to the inter prediction mode is executed in step SP1, and the cost value of each prediction mode is calculated in the subsequent step SP2. . In the subsequent step SP3, an optimum inter prediction mode is detected by comparing the calculated cost values.

  In addition, after the activity is calculated in step SP4 by the processes of the activity calculation circuit 42, the intra prediction circuit 46, and the intra / inter determination circuit 45 that are simultaneously and parallel to the process related to the motion prediction / compensation circuit 44, the following steps are performed. In SP5, an offset value is calculated by this activity. In subsequent step SP6, a cost value related to each intra prediction mode is calculated based on the calculated offset value. In step SP7, the calculated cost value is compared with the cost value calculated in step SP3. In the following step SP8, the optimum mode is detected.

(2) Operation of the embodiment In the above configuration, in the encoding device 41 (FIG. 1), the video signal S1 sequentially input is converted into the image data D1 by the analog-digital conversion circuit 2, and the image data D1 is converted into the image data D1. The data is rearranged in the processing order by the screen rearrangement buffer 3 and input to the subtraction circuit 4. Here, the image data D1 is subtracted between the prediction values obtained by intra prediction and inter prediction to generate subtraction data D2, and the subtraction data D2 is generated by the orthogonal transform circuit 7, the quantization circuit 8, and the lossless encoding circuit 10. It is sequentially processed and converted into encoded data D4, and this encoded data D4 is recorded on a recording medium by a recording system, for example. Further, the output data of the quantization circuit 8 is decoded into image data and recorded as a reference image in the frame memory 16, and based on this reference image, the motion prediction / compensation circuit 44 and the intra prediction circuit 46 predict inter prediction and intra prediction prediction values. Is generated.

  In these series of processing, the image data D1 is obtained in the motion prediction / compensation circuit 44 and the intra prediction circuit 46 for the cost value indicating the coding efficiency for each prediction mode of inter prediction and intra prediction, respectively. For inter prediction, an optimum mode most suitable for encoding processing is detected by comparing cost values in the motion prediction / compensation circuit 44. The intra / inter determination circuit 45 detects the optimal prediction mode by comparing the cost value of each prediction mode of intra prediction with the cost value of the optimal mode related to inter prediction detected by the motion prediction / compensation circuit 44. Is done. As a result, the encoding device 41 determines whether to perform the encoding process by intra prediction or inter prediction based on the optimal prediction mode. In the case of intra prediction, the intra prediction circuit 46 performs prediction based on the optimal mode. A value is generated and output to the subtraction circuit 4. In the case of inter prediction, the motion prediction / compensation circuit 44 generates a prediction value in the optimum mode and outputs it to the subtraction circuit 4. Accordingly, the encoding device 41 detects the optimum mode for each macroblock from the plurality of intra prediction modes and the plurality of inter prediction modes by comparing the cost values with the cost function indicating the coding efficiency. D1 is sequentially encoded.

  Therefore, among the cost values in each of these prediction modes, the cost value in the intra 4 × 4 prediction mode has conventionally been changed from a quantization parameter related to additional information to a quantized value based on a constant of value 24 as shown in Equation (77). An intra macro block in the inter slice is calculated by multiplying the conversion function QP0 (QP) into the offset value SA (T) D0 and offsetting the cost value Cost (Mode) by the offset value SA (T) D0. Is set so as to reduce the occurrence of.

  As a result, in areas with high activity, when an intra macroblock is selected, an intra prediction mode in a specific direction according to the design is appropriately selected, thereby ensuring image coding efficiency and effective image quality degradation. Can be avoided. However, in the low activity region, the prediction mode is switched variously due to the influence of noise, and the block boundary distortion is excessively suppressed by the deblocking filter.

  For this reason, in this encoding device 41, the activity of the image data D1 is calculated by the activity calculation circuit 42 as a parameter indicating the flatness of the image, and the intra / inter determination circuit 45 uses this activity in the intra 4 × 4 prediction mode. After the cost value is corrected, the optimum mode is detected. As a result, the encoding device 41 is configured to control the selection of the optimal mode according to the activity. With this configuration, it is possible to appropriately select the optimal mode and prevent image quality deterioration in a low activity area. Become.

  That is, in this encoding device 41, the correction of the cost value is set so that the intra prediction mode becomes difficult to be selected when there are few high-frequency components in the macroblock, so that the intra 4 × 4 prediction in the low activity region is performed. A fluttering feeling of the decoded image due to frequent switching of modes is prevented, and image quality deterioration such as flicker is prevented. Furthermore, it is possible to prevent excessive block boundary distortion suppression by the deblocking filter in the low activity region as described above, thereby preventing the apparent resolution from deteriorating and preventing the image quality from deteriorating.

More specifically, in this embodiment, the intra 4 × 4 prediction mode is defined by a function that gives an offset value SA (T) D0 to the error value SA (T) D between the original image and the predicted image. For the cost function, the multiplication value of the function f (MB act) having the activity MB act as a variable and the conversion function QP0 (QP) from the quantization parameter to the quantization value regarding the additional information is set as the offset value SA (T) D0. Is set to correct the cost value according to the activity, and this activity MB
By setting the function f (MB act) with act as a variable, it is possible to correct the cost value with various characteristics as necessary, thereby improving the image quality in a simple and reliable manner.

  Further, in this embodiment, the function f (MB act) having the activity MB act as a variable is set as a monotonically decreasing function, and thereby it is possible to reliably prevent image quality deterioration in a low activity area.

  In addition, the activity that is the processing standard is calculated based on the variance value of the pixel values based on the image data D1, so that the selection of the intra prediction mode can be appropriately controlled in a region where such image quality degradation is easily perceived. As a result, the image quality can be further improved as compared with the prior art.

(3) Effects of the embodiment According to the above configuration, the optimum mode is detected for each macroblock from the plurality of intra prediction modes and the plurality of inter prediction modes by comparing the cost values with the cost function indicating the coding efficiency. When image data is encoded, the cost value is corrected by activity and the optimal mode is detected, and the optimal mode is selected from the intra prediction mode and the inter prediction mode by the cost function to encode the image data. In this case, it is possible to prevent image quality deterioration in a low activity area.

  Also, at this time, the correction of the cost value is a correction of the cost value that makes it difficult to select the intra prediction mode when there are few high-frequency components in the macroblock. Can be prevented.

  In addition, since the activity that is the processing standard is a dispersion value of pixel values based on the image data D1, it is possible to appropriately control the selection of the intra prediction mode in an area where image quality degradation is easily perceived. Thus, the image quality can be further improved.

  Also, for the cost value by the cost function that gives an offset value to the error value between the original image and the predicted image, the product of the function with the activity as a variable and the conversion function from the quantization parameter to the quantization value for the additional information Is set as an offset value, and the cost value is corrected by the activity, so that the cost value can be corrected by various characteristics as required by setting the function using this activity as a variable. Certainly, the image quality can be improved in various ways.

  More specifically, by applying a monotone decreasing function to a function having this activity as a variable, it is possible to reliably prevent image quality deterioration in a low activity area.

  In this embodiment, instead of directly calculating the macroblock activity MB act in units of macroblocks of 16 × 16 pixels described above for the first embodiment, the macroblock is divided into four in the horizontal and vertical directions, respectively. The macro block activity MB act is calculated in units of blocks of 4 × 4 pixels. The encoding apparatus according to the present embodiment is configured in the same manner as the encoding apparatus 41 described above with respect to the first embodiment except that the configuration of the activity calculation circuit related to the detection of this activity is different.

  That is, in this embodiment, the activity calculation circuit executes the following arithmetic processing for each block of 4 × 4 pixels obtained by dividing the macroblock into 4 parts in the horizontal direction and the vertical direction, and thereby the 4 × 4 pixels are used. The activity act of each 4 × 4 pixel block is detected by block distribution.

  Further, the activity calculation circuit calculates the activity MB act of the macro block by grouping the activity act of the various blocks calculated in this way by the macro block. Specifically, the activity calculation circuit calculates the minimum value of the activity MB act of the macro block by calculating the minimum value from the activity act of the 4 × 4 pixel block constituting the macro block by the arithmetic processing of the following equation.

  As in this embodiment, after calculating the dispersion value of the pixel value based on the image data D1 for each block obtained by subdividing the macroblock, the minimum value may be detected and set in the activity. Similar effects can be obtained.

  In this embodiment, instead of the direct calculation of the activity MBact of the macroblock in units of blocks of 4 × 4 pixels described above for the second embodiment, the macroblock is divided into two in the horizontal direction and the vertical direction, respectively. The activity MB act of the macroblock is calculated in units of blocks of x8 pixels. Note that the encoding apparatus according to this embodiment is configured in the same manner as the encoding apparatus 41 described above with respect to the second embodiment, except that the configuration of the activity calculation circuit for detecting this activity is different.

  As in this embodiment, a block obtained by subdividing a macroblock is set to a block of 8 × 8 pixels, and a variance value of pixel values based on image data D1 is calculated for each block, and then a minimum value is detected. Even if the activity is set, the same effect as in the second embodiment can be obtained.

  In this embodiment, instead of detecting the activity MB act by calculating the variance value based on the average value, the activity MB act is calculated by Hadamard transform processing. The encoding apparatus according to the present embodiment is configured in the same manner as the encoding apparatus described above with respect to the second embodiment, except that the configuration of the activity calculation circuit for detecting the activity is different.

  In other words, in this embodiment, the activity calculation circuit performs the following arithmetic processing for each block of 4 × 4 pixels obtained by dividing the macroblock into 4 parts in the horizontal direction and the vertical direction, thereby performing 4 × by Hadamard transform processing. The activity act of the 4-pixel block is detected.

Here, H ₄ is a fourth-order Hadamard matrix expressed by the equation (82). Further, the absolute value sum excluding the DC component is obtained from the equation (83) for the matrix obtained by the arithmetic processing of the equation (81), and this is defined as the activity act of the 4 × 4 pixel block. Instead of such a 4 × 4 pixel block process using a 4th order Hadamard matrix, the activity is detected by an 8 × 8 pixel block process or a 16 × 16 pixel block process using an 8th order or 16th order Hadamard matrix. May be.

  In addition, the activity act based on the 4 × 4 pixel block calculated in this way is collected in a macro block by the arithmetic processing of the expression (80) to detect the activity MB act.

  Even if the activity MB act is detected by Hadamard transform processing as in this embodiment, the same effect as in the first embodiment can be obtained.

  In this embodiment, the cost function is corrected by the activity even in the detection of the optimal mode related to the inter prediction. The encoding apparatus according to the present embodiment is configured in the same manner as the encoding apparatus 41 described above with respect to the first embodiment except that the configuration of the motion prediction / compensation circuit according to the inter prediction process is different.

  That is, in this embodiment, the motion prediction / compensation circuit generates an offset value based on the activity MB act and calculates the cost value based on this offset value in calculating the cost value when detecting the optimum mode from the prediction mode. Thus, the motion prediction / compensation circuit controls the selection of the size of the motion compensation block, the pre-prediction, the post-prediction, and the bi-directional prediction according to the activity.

  As a result, in this encoding apparatus, for example, when the activity is low, frequent switching of the size of the motion compensation block, frequent switching of the forward prediction, the post prediction, and the bidirectional prediction is prevented. The flickering feeling caused by such switching is prevented and the image quality is further improved.

  According to this embodiment, it is possible to further improve the image quality by further controlling the selection of the size of the motion compensation block, the pre-prediction, the post-prediction, and the bi-directional prediction by the activity.

  In the above-described embodiments, the case where the present invention is applied to the Low Complexity Mode in AVC has been described. However, the present invention is not limited to this, and may be applied to the High Complexity Mode.

  In the above-described embodiments, the case where the present invention is applied to an AVC encoding apparatus has been described. However, the present invention is not limited to this, and a plurality of intra-frames can be obtained by comparing cost values using cost functions indicating encoding efficiency. The present invention can be widely applied to a case where an optimal mode is detected for each macroblock from a prediction mode and a plurality of inter prediction modes and image data is encoded.

  In the above-described embodiments, the case where the present invention is applied to the hardware configuration has been described. However, the present invention is not limited to this, and can also be applied to the case where image data is processed by software. Note that the encoding processing and decoding processing programs related to such software are widely used when provided by a network such as the Internet, when provided by various recording media such as an optical disk, a magnetic disk, and a memory card. Can be applied.

  The present invention relates to an encoding apparatus, an encoding method, an encoding method program, and a recording medium on which the encoding method program is recorded, and is applied to a video camera, an electronic still camera, a monitoring apparatus, and the like that record imaging results of moving images. be able to.

It is a block diagram which shows the encoding apparatus which concerns on Example 1 of this invention. It is a flowchart which shows the process sequence of the optimal mode detection in the encoding apparatus of FIG. 1 is a block diagram illustrating an AVC encoding apparatus. FIG. It is a block diagram which shows the decoding apparatus of an AVC system. It is a basic diagram with which it uses for description of the process of the coefficient data by an AVC system. It is a basic diagram with which it uses for description of the setting of the prediction pixel in the intra 4 * 4 prediction mode of AVC system. It is a basic diagram with which it uses for description of intra 4x4 prediction mode. It is a chart which shows intra 4x4 prediction mode. It is a basic diagram with which it uses for description of each mode of intra 4x4 prediction mode. It is a basic diagram with which it uses for description of transmission of prediction mode. It is a graph which shows the decoding process of prediction mode by the description of C language. It is a basic diagram with which it uses for description of the prediction pixel of intra 16x16 prediction mode. It is a graph which shows intra 16x16 prediction mode. It is a basic diagram with which it uses for description of intra 16x16 prediction mode. It is a graph with which it uses for description of the intra prediction mode which concerns on a color difference signal. It is a basic diagram with which it uses for description of the reference frame of an AVC system. It is a basic diagram with which it uses for description of the motion compensation of an AVC system. It is a basic diagram with which it uses for description of the motion compensation precision of an AVC system. It is an approximate line figure used for explanation of motion compensation of a color difference signal. It is a basic diagram with which it uses for description of the predicted value of the motion vector which concerns on a submacroblock. It is a basic diagram with which it uses for description of the predicted value of the motion vector by another example. It is an approximate line figure used for explanation of temporal direct mode. It is an approximate line figure used for explanation of processing of a deblocking filter. It is a table | surface used for description of the intensity | strength of a block boundary. It is a characteristic curve figure with which it uses for description of adjustment | control of the intensity | strength of a deblocking filter. 6 is a chart showing parameters α and β related to setting of characteristics of a deblocking filter. It is a graph which shows the parameter tco which concerns on the setting of the characteristic of a deblocking filter.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Coding device 5, 23, 46 ... Intra prediction circuit, 6, 24, 44 ... Motion prediction / compensation circuit, 15, 28 ... Deblock filter, 42 ... Activity calculation circuit, Intra / inter Judgment circuit

Claims (9)

An encoding device that detects an optimal mode for each macroblock from a plurality of intra prediction modes and a plurality of inter prediction modes by comparing cost values with cost functions indicating encoding efficiency, and encodes image data in the optimal mode In
Activity calculating means for calculating an activity indicating the flatness of the image by the image data for each macroblock;
Optimum mode detection means for correcting the cost value by the activity and detecting the optimum mode;
Equipped with a,
The encoding apparatus, wherein the correction of the cost value by the optimum mode detection means is a correction of a cost value that makes it difficult to select the intra prediction mode when there are few high-frequency components in the macroblock.   The encoding apparatus according to claim 1, wherein the activity is a variance value of pixel values based on the image data. The activity calculating means calculates a variance value of pixel values based on the image data for each block obtained by subdividing the macroblock, and detects a minimum value from the variance value of the subdivided block for each macroblock. The encoding apparatus according to claim 2 , wherein the activity is set to the activity. The activity calculation means includes:
Performing Hadamard transform for each block obtained by subdividing the macroblock,
By calculating the absolute value sum by removing the coefficient of the DC component from the coefficient data by the processing result, the activity is calculated for each of the subdivided blocks,
The encoding apparatus according to claim 1 , wherein a minimum value is detected from the activity of the subdivided block and set as the activity of the macroblock for each macroblock. The cost function is a function that gives an offset value with respect to an error value between an original image and a predicted image,
The activity calculation means sets the cost value by the activity by setting a multiplication value of a function having the activity as a variable and a conversion function from a quantization parameter to a quantization value regarding additional information to the offset value. The encoding device according to claim 1, wherein correction is performed. The encoding apparatus according to claim 5 , wherein the function having the activity as a variable is a monotone decreasing function. An encoding method for detecting an optimal mode for each macroblock from a plurality of intra prediction modes and a plurality of inter prediction modes by comparing cost values using a cost function indicating encoding efficiency, and encoding the image data in the optimal mode In
An activity calculating step for calculating an activity indicating the flatness of the image by the image data for each macroblock;
An optimal mode detection step of detecting the optimal mode by correcting the cost value by the activity;
I have a,
The encoding method, wherein the correction of the cost value in the step of detecting the optimum mode is a correction of a cost value that makes it difficult to select the intra prediction mode when there are few high-frequency components in the macroblock . An optimum mode detection step for detecting an optimum mode for each macroblock from a plurality of intra prediction modes and a plurality of inter prediction modes by comparing cost values using a cost function indicating coding efficiency, and encoding the image data by the optimum mode An encoding method program comprising:
An activity calculating step for calculating an activity indicating flatness of an image based on the image data for each macroblock;
The step of detecting the optimum mode is a step of detecting the optimum mode by correcting the cost value by the activity ,
The program of the encoding method , wherein the correction of the cost value in the optimum mode detection step is a correction of a cost value that makes it difficult to select the intra prediction mode when there are few high-frequency components in the macroblock . In a recording medium on which a program of an encoding method executed by arithmetic processing means is recorded,
The encoding method program is:
Optimal mode detection step for detecting an optimal mode for each macroblock from a plurality of intra-prediction modes and a plurality of inter-prediction modes by comparing cost values with a cost function indicating coding efficiency;
A step of encoding processing for encoding image data in the optimum mode;
An activity calculating step for calculating an activity indicating flatness of an image based on the image data for each macroblock;
The step of detecting the optimum mode is a step of detecting the optimum mode by correcting the cost value by the activity ,
A program of an encoding method , wherein the correction of the cost value in the optimal mode detection step is a correction of a cost value that makes it difficult to select the intra prediction mode when there are few high-frequency components in the macroblock. Recorded recording medium.

Images