JP2011234070A - Image coding apparatus, image coding method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an adaptive quantization technique capable of preventing propagation within a screen of an error generated in intra-prediction in a method called intra-prediction coding.SOLUTION: An image coding apparatus codes an input image using intra-prediction coding. The image coding apparatus comprises a saturation calculation section 130 for calculating saturation in a block to be coded; a DC calculation section 131 for calculating a DC value of a difference block calculated by performing a difference calculation between the block to be coded and a prediction block corresponding to the block to be coded; and a quantization parameter determination section 103 for determining a quantization parameter of a target block from the block saturation, the DC value of the difference block, and the picture quantization parameter determined in advance.

Description

  The present invention relates to a high-efficiency encoding technique. More specifically, H.264 of moving images. The present invention relates to an adaptive quantization technique that can prevent an error generated in intra prediction from being propagated to subsequent macroblocks in H.264 encoding.

  In recent years, digitalization of AV information has progressed and devices capable of digitizing and handling video signals are becoming widespread. Since a video signal has a large amount of information, it is common to encode while reducing the amount of information in consideration of recording capacity and transmission efficiency. As a video signal encoding technique, H.264 is used. An international standard called H.264 has been established. The H.264 standard provides approximately twice the compression efficiency compared to other existing coding technologies. Therefore, H.I. The H.264 standard is an encoding method that is very suitable for recorder recording, Internet streaming, and the like.

  H. In H.264, intra prediction that improves coding efficiency by reducing spatial redundancy in an image, and motion compensation that improves coding efficiency using temporal redundancy existing between consecutive pictures Has been introduced. A picture coded by intra prediction is called an I picture, and a picture coded by motion compensation is called a P picture or a B picture. FIG. 6 shows an example of the overall configuration of a typical moving image compression encoding apparatus based on intra prediction. The moving image compression encoding apparatus includes an input unit 300, a difference calculation unit 301, a conversion unit 302 that converts image data into a frequency domain, and a conversion coefficient obtained by the conversion unit 302 is quantized with a predetermined quantization parameter. A quantization unit 304; an entropy coding unit 305 that entropy-encodes the quantized transform coefficient; an output unit 306 for supplying compressed image data having a variable bit rate to a transmission line according to a transmission rate; a local decoder 310; An intra prediction mode detection unit 307 is provided.

  Intra prediction is an effective method that can significantly reduce the amount of information of a block to be encoded and can improve compression efficiency. However, there is a problem that an error generated in an arbitrary macro block is propagated in a screen in moving image compression encoding at a low bit rate. In intra-prediction encoding, a prediction error block obtained by a difference calculation between an encoding target block and an intra prediction block is an encoding target. That is, in the decoding process, a reproduction macroblock is obtained by adding an intra prediction block and a prediction error block. If Qp_MB is not small enough to reproduce the information included in the prediction error block at a low bit rate, the residual signal to be added to the intra prediction block becomes 0, and the reproduced image becomes the intra prediction block itself. . This phenomenon also occurs for the subsequent macroblocks, resulting in a large visual deterioration. In particular, this problem becomes significant in the color difference signal.

  The manner in which the intra prediction error propagates in the screen will be described with reference to FIGS. 7, 8, 9, 10, and 11. In FIG. 7, an encoding target image 400 is an image in which a red object is present on a white background. The encoding target image 400 is divided into slices 401, 402, 403, and 404. A red object 410 exists in the encoding target image 400. In FIG. 7, macroblocks 420, 430, and 440 are located at the boundary between the slice 402 and the slice 403. The macro block 420 includes only the object 410. Macroblocks 430 and 440 are composed only of the background. FIG. 8 shows pixel values in the original picture state of the Cr and Cb components of each macroblock. In FIG. 8A, the pixel 500 is the pixel value of the Cr component of the macroblock 420. In FIG. 8B, the pixel 501 is the pixel value of the Cb component of the macroblock 420. In FIG. 8C, the pixel 502 is a pixel value of the Cr component of the macro block 430. In FIG. 8D, the pixel 503 is the pixel value of the Cb component of the macroblock 430. In FIG. 8E, a pixel 504 is a pixel value of the Cr component of the macro block 440. In FIG. 8F, the pixel 505 is the pixel value of the Cb component of the macroblock 440. Macroblocks 420, 430, and 440 are encoded in this order.

  With reference to FIGS. 9 and 10, the encoding of the macroblock 430 will be described with a particular focus on the Cr component. In FIG. 9, a pixel 510 indicates a Cr component pixel of the macroblock 430. A peripheral pixel 511 indicates a peripheral pixel used for intra prediction, and is a reproduction pixel of the macroblock 420. Since the macroblock 430 is located at the boundary between the slice 402 and the slice 403, only the prediction mode 0 (DC prediction) and the prediction mode 1 (horizontal prediction) can be used as the color difference intra prediction modes. FIG. 10A shows a two-dimensional representation of the macroblock pixel 510 and the peripheral pixel 511 as a pixel waveform 431 and a pixel waveform 425, respectively. The pixel values of the pixel waveform 431 and the pixel waveform 425 are respectively shown. 512 and 576 in 10-bit representation. The vertical axis represents the pixel value, and the horizontal axis represents the horizontal position within the macroblock. FIG. 10B shows a predicted block pixel waveform 432 when the optimum prediction mode is selected as 0 (DC prediction). FIG. 10C shows a prediction error block pixel waveform 433 obtained by calculating a difference between the macroblock pixel waveform 431 and the intra prediction block pixel waveform 432. The prediction error block pixel waveform 433 is a target of encoding. When the quantization parameter increases at a low bit rate, the information amount of the prediction error block pixel waveform 433 is quantized as shown in FIG. The lost prediction error block pixel waveform 434 becomes 0. In this case, the decoded pixel waveform of the macroblock pixel waveform 431 is the predicted block pixel waveform 432. FIG. 10E shows a decoded pixel waveform 435 of the macroblock pixel waveform 431. From the figure, it can be seen that in the macroblock pixel waveform 431, the pixels of the object 410 are diffused in the portion (pixel waveform 436) that was the background in the original image state.

  Next, with reference to FIG. 11, the encoding of the macroblock 440 will be described focusing on the Cr component in the same manner as the macroblock 430. In FIG. 11A, a pixel waveform 441 indicates a Cr component pixel of the macroblock 440. Since the macroblock 440 is located at the boundary between the slice 402 and the slice 403, only the prediction modes 0 and 1 can be used. Therefore, the peripheral pixels used for the intra prediction of the macroblock 440 are the reproduced pixels of the macroblock 430, which is the pixel waveform 435 described with reference to FIG. FIG. 11B shows a prediction block pixel waveform 442 when the optimum prediction mode is selected as 0 (DC prediction). FIG. 11C shows a prediction error block pixel waveform 443 obtained by calculating a difference between the macroblock pixel waveform 441 and the intra prediction block pixel waveform 442. This prediction error block pixel waveform 443 is an object to be encoded. However, when the quantization parameter increases at a low bit rate, the prediction error block pixel waveform 443 is lost due to quantization as shown in FIG. Therefore, the decoded prediction error block pixel waveform 444 is also zero. In this case, the decoded pixel waveform of the macroblock pixel waveform 441 is a predicted block pixel waveform 442. FIG. 11E shows a decoded pixel waveform 445 of the macroblock pixel waveform 441. From the figure, it can be seen that the pixels of the object 410 are diffused in the portion of the macroblock pixel waveform 441 that was the background in the original image state (pixel waveform 446). Also in the macroblocks after the macroblock 440, as described above, the information amount of the prediction error block pixels becomes 0 due to quantization, so that the pixels of the object 410 included in the macroblock 420 are propagated.

  The root cause of intra-screen propagation of this subjectively undesirable intra prediction error is that the quantization parameter becomes large, so that the information of the prediction error block is lost due to quantization and becomes 0, and the shape of the intra prediction block Becomes a decoding block of the encoding target block as it is.

  In Patent Document 1, in order to avoid the above problem, distortion of the color difference signal is easily detected in a region where the luminance signal is flat. Therefore, in the macro block where the luminance signal is flat, the quantum difference of the color difference signal in the same macro block is detected. We propose a method to avoid intra prediction propagation of intra prediction errors by making the optimization parameter relatively small.

JP-A-2-105791

  However, although the above-described conventional technique can avoid intra prediction error propagation in a screen for a macroblock with a flat luminance signal, there is a problem that it cannot be applied to a block with a non-flat luminance signal. Intra prediction error propagation may occur even in a block where the luminance signal is not flat, whereas the conventional technique cannot cope with such a case.

  In order to solve the above-described conventional problems, an image encoding device according to the present invention is an image encoding device that encodes an input image using intra prediction encoding, and detects the visual sensitivity of an encoding target block. A visual sensitivity detector, a prediction transition detector for detecting a transition between the encoding target block and a prediction block corresponding to the encoding target block, the visual sensitivity, the transition, and a preset picture quantization A quantization parameter determining unit that determines a quantization parameter of the encoding target block according to the parameter.

  According to the image coding apparatus of the present invention, the degree of intra prediction propagation of intra prediction error is detected from the visual sensitivity, transition, and picture quantization parameter of the encoding target block, and it is determined that intra screen propagation occurs. By making the quantization parameter relatively small with respect to the generated block, it is possible to prevent intra prediction error propagation in the screen, and to improve the subjective image quality.

FIG. 1 is a block diagram showing a configuration of a moving image compression coding apparatus in which an adaptive processing apparatus according to Embodiment 1 is incorporated. The figure which shows the pixel value change of the macroblock in the intra prediction process in Embodiment 1. The figure which shows the pixel value change of the other macroblock in the intra prediction process in Embodiment 1. Explanatory drawing about nine types of intra prediction with respect to the luminance signal in Embodiment 1. Explanatory drawing about 4 types of intra prediction with respect to the color difference signal in Embodiment 1 The block diagram which shows the whole structure of the conventional image coding apparatus. The figure which shows the encoding object picture The figure which shows the pixel value of Cr component of the macroblock in Embodiment 1, and Cb component The figure which shows the positional relationship of the surrounding pixel used for object macroblock and intra prediction The figure which shows the pixel value change of the macroblock in the process of the conventional intra prediction process The figure which shows the pixel value change of the other macroblock in the process of the conventional intra prediction process

  Hereinafter, embodiments will be described with reference to the drawings.

(Embodiment 1)
With reference to FIG. 1, FIG. 2, and FIG. 3, the image coding apparatus according to Embodiment 1 will be described.

  FIG. 1 is a block diagram conceptually showing the overall configuration of a moving picture compression encoding apparatus including a motion vector detection apparatus in the first embodiment. The moving image compression encoding apparatus includes an input unit 100, a difference calculation unit 101, a conversion unit 102 that converts image data into a frequency domain, and a conversion coefficient obtained by the conversion unit 102 is quantized with a predetermined quantization parameter. A quantization unit 104; an entropy encoding unit 105 that entropy-encodes the quantized transform coefficient; an output unit 106 for supplying compressed image data having a variable bit rate to a transmission line according to a transmission rate; a local decoder 110; An intra prediction mode detection unit 107, a saturation calculation unit 130, a DC calculation unit 131, and a quantization parameter determination unit 103 are provided.

  The quantization unit 104 performs a process of rounding the result obtained by dividing the transform coefficient by the quantization step to an integer value. H. In H.264, a quantization step is derived from a parameter called Qp. As the value of Qp increases, the quantization step increases and the number of bits required for encoding decreases. On the contrary, when the value of Qp becomes small, the quantization step becomes small and the number of bits necessary for encoding increases. Therefore, by changing the value of Qp, the generated code amount in the encoding target image can be suppressed within a predetermined code amount corresponding to the recording medium or the transmission medium. The value of Qp can take a value from 0 to 51. When Qp is reduced by 6, there is a characteristic that the quantization step is halved.

  The difference calculation unit 101 calculates a difference value between the image data 120 and the intra predicted image 121. The intra prediction image 121 is obtained by decoding an already encoded image and performing intra prediction based on the intra prediction mode. This is performed by the local decoder 110 corresponding to image coding.

  The local decoder 110 performs the encoding procedure in reverse. That is, the local decoder 110 adds the intra prediction image generated by the buffer 114 and the intra prediction unit 115 to the difference data decoded by the inverse quantization unit 111 and the inverse transform unit 112, and an addition calculation unit 113. The same image as that obtained on the decoding side is generated.

  In intra prediction encoding, an intra prediction image of a target image is an intra between the target image and a decoded image (a decoded image stored in the buffer 114 and corresponding to a reference image). Based on the prediction, it is generated from the image data of the corresponding decoded image.

  The intra prediction mode detection unit 107 uses the spatial correlation between the input image data 120 and the encoded intra prediction mode detection target image input from the buffer 114 in the local decoder, to optimize the prediction. Intra prediction mode is detected. Usually, intra prediction is performed in units of blocks. That is, a prediction mode that generates an intra prediction block having the highest correlation with a block of a target image in the decoded image is detected as an optimal prediction mode. An intra prediction mode detection unit 107 and an intra prediction unit 115 that generates an intra prediction image from an image decoded in accordance with the optimal intra prediction mode are incorporated in the encoder.

  The saturation calculation unit 130 calculates the saturation for each of the Cr and Cb blocks of the image data 120 divided into macroblock units.

  The DC calculation unit 131 calculates a value indicating how much the DC value of the Cr component and the Cb component has fluctuated due to the intra prediction for each of the Cr and Cb blocks in the input intra prediction error block.

  The quantization parameter determination unit 103 determines whether intra prediction error propagation occurs in the macroblock based on the saturation calculated by the saturation calculation unit 130, the value calculated by the DC calculation unit 131, and Qp_pic. Determine the quantization parameters for the macroblock.

  Next, the operation of the moving image compression encoding apparatus shown in FIG. 1 will be described. In FIG. 1, first, the image data 120 input from the input unit 100 is divided into macroblocks (vertical 16 pixels × horizontal 16 pixels). Thereafter, encoding processing is performed in units of macroblocks. The difference calculation unit 101 performs a difference calculation with the intra predicted image 121 on the image data divided into macro blocks.

  The difference data subjected to the difference calculation is orthogonally transformed by the transform unit 102 and quantized by the quantization unit 104. The quantization unit 104 performs a quantization process using the quantization parameter determined by the quantization parameter determination unit 103. The orthogonal transform coefficient of the quantized difference data is entropy encoded by the entropy encoding unit 105 and output as a bit stream from the output unit 106. Further, the orthogonal transform coefficient of the quantized difference data is inversely quantized by the inverse quantization unit 111, inversely orthogonally transformed by the inverse transform unit 112, and decoded as difference data. Note that the difference data at this time is not the original difference data because irreversible processing is performed by quantization for reducing the amount of data.

  The decoded difference data is added by the addition operation unit 113 to the intra predicted image that has been intra predicted by the intra prediction unit 115. The added image is temporarily stored in the buffer 114 as a local decoded image in the encoder as a reference image for later intra prediction mode detection or intra prediction. The intra prediction mode detection unit 107 calculates an intra prediction mode between the image data divided into macroblocks from which the intra prediction mode is to be detected and the reference image stored in the buffer 114.

  The intra prediction mode is calculated as follows. For example, block matching is performed between the target block and a prediction block generated by all 9 modes, and the difference absolute value sum between the blocks is obtained, and the prediction when the difference absolute value sum is minimized The mode is determined as the optimum prediction mode in the target block. The optimal intra prediction mode calculated by the intra prediction mode detection unit 107 is supplied to the intra prediction unit 115. In the intra prediction unit 115, an intra predicted image is generated from the reference images stored in the buffer 114 using peripheral pixels corresponding to the optimal intra prediction mode.

  The calculation method of the intra prediction mode in the intra prediction mode detection unit 107 will be described in more detail. H. In H.264, intra prediction in each macroblock can be performed using different block sizes for the luminance signal. Individual intra prediction modes can be determined for a total of three types of blocks of 16 × 16, 8 × 8, and 4 × 4 pixels, which are block sizes for intra prediction. Hereinafter, description will be made assuming that the block size of intra prediction is 8 × 8.

  First, intra prediction for a luminance signal will be described. FIG. 4 is a diagram illustrating nine intra prediction modes used for intra prediction. Pixels A to Y are encoded peripheral pixels used for intra prediction. There are nine types of intra prediction modes that are candidates for each 8 × 8 block, corresponding to the prediction direction, and one of these is selected for encoding.

  In FIG. 4, mode 0 is defined as “vertical prediction”, and the luminance of pixels A to H of the encoded block adjacent to the upper part of the 8 × 8 block to be predicted is converted from the luminance of the pixel below vertically. Then, an absolute value sum of differences is obtained, and an absolute difference sum of luminance is calculated. That is, the brightness of pixel A is subtracted from the brightness of 8 pixels below pixel A, and the brightness of pixel B is subtracted from the brightness of 8 pixels below pixel B. Similarly, from the luminance of each of the eight pixels below the pixel C, pixel D, pixel E, pixel F, pixel G, and pixel H, the pixel C, pixel D, pixel E, pixel F, pixel G, and pixel, respectively. The luminance of H is subtracted. Thus, the sum of absolute difference values of luminance obtained as a subtraction result becomes a prediction error in prediction mode 0 of the 8 × 8 block to be predicted.

  Next, mode 1 is defined as “horizontal prediction”, and the luminance of the encoded pixels Q to X adjacent to the left part of the prediction target block is subtracted from the luminance of the pixels in the horizontal direction to calculate the absolute difference. A value sum is obtained, and an absolute difference value sum of luminance is calculated. The sum of absolute difference values of luminance obtained as a result of this subtraction becomes a prediction error in prediction mode 1 of the prediction target block.

  Next, mode 2 is defined as “DC prediction”, and the average values of the luminances of the encoded pixels A to H and the pixels Q to X adjacent to the upper and left parts of the prediction target block are set in the prediction block. By subtracting from the luminance of each pixel, the absolute value sum of the differences is obtained, and the absolute value sum of the luminance differences is calculated. The sum of absolute difference values of luminance obtained as a result of this subtraction becomes a prediction error in prediction mode 2 of the prediction target block.

  Thereafter, the same processing is performed for the modes 3 to 8, and the prediction error in each mode is calculated. After calculating the prediction error, the mode with the smallest prediction error among the nine modes is determined as the optimum prediction mode for the luminance signal of the prediction target block.

  Next, intra prediction for color difference signals will be described. For the color difference signal, one intra prediction mode is set for one macroblock. In the present invention, for convenience of explanation, the block size for intra prediction for color difference signals is assumed to be 8 × 8. FIG. 5 is a diagram illustrating four intra prediction modes used for intra prediction. Pixels A to H and Q to Y are encoded peripheral pixels and are used when performing intra prediction. There are four types of intra prediction modes that are candidates for each 8 × 8 block, corresponding to the prediction direction, and one of these is selected for encoding.

  In FIG. 5, mode 0 is defined as “DC prediction”, and the average value of the color differences between the encoded pixels A to H and the pixels Q to X adjacent to the prediction target block is calculated from the color difference of each pixel in the prediction block. Subtraction is performed to obtain the sum of absolute values of differences, and the sum of absolute values of differences of color differences is calculated. The absolute difference value sum of the color differences obtained as a result of this subtraction becomes a prediction error in the prediction mode 0 of the prediction target block.

  Next, mode 1 is defined as “horizontal prediction”, and the color difference between the pixels Q to X of the encoded block adjacent to the left part of the 8 × 8 block to be predicted is determined from the color difference between the pixels on the right side of the horizontal. Conversion is performed to obtain an absolute value sum of differences, and an absolute difference value sum of color differences is calculated. That is, the color difference of the pixel Q is subtracted from the color difference of 8 pixels on the right side of the pixel Q, and the color difference of the pixel R is subtracted from the color difference of 8 pixels on the right side of the pixel R. Similarly, pixel S, pixel T, pixel U, pixel V, pixel W and pixel S, pixel T, pixel U, pixel V, pixel W, and pixel X are obtained from the color difference of each of the eight pixels. The color difference of the pixel X is subtracted. The sum of absolute difference values of the color differences obtained as a result of the subtraction in this way becomes a prediction error in the prediction mode 1 of the 8 × 8 block to be predicted.

  Mode 2 is defined as “vertical prediction”, and the color difference between the pixels A to H of the encoded block adjacent to the upper part of the 8 × 8 block to be predicted is converted from the color difference of the pixel below vertically, and the difference Is obtained, and the sum of the absolute difference values of the color differences is calculated. The absolute difference value sum of the color differences obtained as a result of this subtraction becomes the prediction error in the prediction mode 2 of the prediction target block.

  Next, mode 3 is defined as “planar prediction”, and the prediction value calculated from the color difference between the encoded pixels A to H, the pixels Q to X, and the pixel Y adjacent to the prediction target block is set in the prediction block. The absolute value sum of the differences is obtained by subtracting from the color difference of each pixel, and the absolute value sum of the difference of the color differences is calculated. The absolute difference value sum of the color differences obtained as a result of this subtraction becomes the prediction error in the prediction mode 3 of the prediction target block.

  After calculating the prediction error, the mode with the smallest prediction error among the four modes is determined as the optimum prediction mode for the color difference signal of the prediction target block.

  The intra prediction unit 115 performs prediction based on the determined prediction mode based on the intra prediction mode detection result by the intra prediction mode detection unit 107. Information included in the prediction error block obtained from the predicted block is converted into a conversion coefficient in the conversion unit 102. As a conversion method, two-dimensional DCT (discrete cosine transform) is used. The obtained transform coefficient is quantized by a quantization unit 104 with a predetermined quantization parameter, and finally entropy coding unit 105 performs entropy coding. Note that the prediction mode detected by the intra prediction mode detection unit 107 is used for intra prediction, is incorporated into compressed image data via the entropy encoding unit 105, and is output from the output unit 106.

  The core technology in the present embodiment is to detect a block in which intra-propagation error propagation occurs, to reduce the quantization parameter of the block, and to improve reproducibility. Therefore, in this embodiment, the quantization parameter when quantizing the difference data orthogonally transformed by the quantization unit 104 is set for the saturation of the block, the DC value of the prediction error block, and the entire image. It is determined by the quantization parameter.

  First, the operation of the saturation detection unit 130 will be described. Saturation Sat_Cr and Sat_Cb are defined by the following equations for Cr and Cb blocks in the input macroblock, respectively. Note that the Cr and Cb block sizes in the present embodiment are 8 vertical pixels × 8 horizontal pixels.

Sat_Cr = (Σ | Pix_Cr (j, i) −GrayLevel |) / 64
Sat_Cb = (Σ | Pix_Cb (j, i) −GrayLevel |) / 64
In the above equation, Pix_Cr (j, i) represents a pixel value in the Cr block. j and i indicate pixel positions in the block. The upper left of the block is j = 0, i = 0, the lower right of the block is j = 7, and i = 7. GrayLevel indicates an achromatic pixel value, and when 10-bit representation (0 to 1023) is performed, GrayLevel = 512. Sat_Cr and Sat_Cb are quantified how far the Cr component and Cb component are away from Gray Level, which is an achromatic color. If this value is large, it indicates that the saturation is high, and conversely the value is Small values indicate low saturation.

  Next, the operation of the DC calculation unit 131 will be described. For each of the Cr and Cb blocks in the input intra prediction error block, diff_DC_Cr and diff_DC_Cb are defined by the following equations.

diff_DC_Cr = (Σ | diff_Cr (j, i) |) / 64
diff_DC_Cb = (Σ | diff_Cb (j, i) |) / 64
In the above equation, diff_Cr (j, i) represents a pixel value in the intra prediction error Cr block. j and i indicate pixel positions in the block. The upper left of the block is j = 0, i = 0, the lower right of the block is j = 7, and i = 7. diff_DC_Cr and diff_DC_Cb quantify how much the DC value of the Cr component and the Cb component fluctuates due to intra prediction. When this value is large, an error due to intra prediction is large. It shows that the error due to prediction is small.

  The quantization parameter determination unit 103 performs intra prediction in the macroblock when the following three conditions are simultaneously satisfied from the input saturation Sat_Cr, Sat_Cb, the DC value diff_DC_Cr, diff_DC_Cb, and Qp_pic of the intra prediction error block. It is determined that error propagation occurs.

(1) Sat_Cr ≦ thr_Sat and Sat_Cb ≦ thr_Sat
(2) diff_DC_Cr> thr_diff_DC or diff_DC_Cb> thr_diff_DC
(3) Qp_pic> thr_Qp
A quantization parameter Qp_MB for a macroblock determined to cause intra prediction error propagation is set as follows.

Qp_MB = Qp_pic-Qp_ofs
Here, thr_Sat and thr_diff_DC are predetermined threshold values, and Qp_ofs is a predetermined offset value.

  Using the test image of FIG. 7 used in the description of the conventional example, the saturation detection, DC detection, and quantization parameter determination means that are the core of the present embodiment will be specifically described.

  The encoding of the macroblock 430 in the present embodiment will be described using FIG. 2 and FIG. Note that the quantization parameter Qp_pic = 30 set in the picture 400 is set. As a result of examination using various test images, in the present embodiment, the thresholds related to intra prediction error propagation detection in the quantization parameter determination unit 103 are thr_Sat = 16, thr_diff_DC = 32, thr_Qp = 26, Qp_ofs = 6, respectively. Set to.

  First, the saturation calculation unit 130 calculates the saturation Sat_Cr and Sat_Cb of the macroblock 430. Since GrayLevel = 512, Sat_Cr = (64 × | 512−512 |) / 64 = 0, Sat_Cb = (64 × | 512−512) from the pixel values of the Cr and Cb components of the macroblock 430 shown in FIG. |) / 64 = 0.

  Next, the DC calculation unit 131 calculates DC values diff_DC_Cr and diff_DC_Cb of the intra prediction error block. The peripheral pixels used for the intra prediction of the macroblock 430 are reproduction pixels of the immediately preceding macroblock 420. Assuming that the Cr and Cb components of the macroblock 420 are reproduced without deterioration, the pixel values of the intra prediction block of the Cr component corresponding to the macroblock 430 are shown in FIG. 2A, and the Cb component of FIG. The pixel value of an intra prediction block is shown. From the drawing, diff_DC_Cr = (64 × | 576−512 |) / 64 = 64 and diff_DC_Cb = (64 × | 512−512 |) / 64 = 0.

  At this time, since the macroblock 430 satisfies the above three conditions for intra prediction error propagation simultaneously, Qp_MB = 30−6 = 24 is set in the macroblock. The change in the pixel value at this time will be described by focusing on the Cr component.

  In FIG. 2C, the pixel waveform 231 indicates the Cr component pixel of the macroblock 430. A peripheral pixel waveform 225 indicates peripheral pixels used for intra prediction, which are reproduced pixels of the macroblock 420. FIG. 2D shows a prediction block pixel waveform 232 when the optimum prediction mode is selected as 0 (DC prediction). FIG. 2E shows a prediction error block pixel waveform 233 obtained by calculating a difference between the macroblock pixel waveform 231 and the intra prediction block pixel waveform 232. This prediction error block pixel waveform 233 is an encoding target. The macro block 430 is determined to be an intra prediction error propagation block by the quantization parameter determination unit 103, and Qp_MB is set to 24. H. In H.264, when Qp is reduced by 6, the quantization step is halved. When Qp = 30, the data of 64 or less becomes 0 by quantization, but when Qp = 24, the quantization step is halved, so that data larger than 32 can be restored. As a result, as shown in FIG. 2 (f), it is possible to restore the information of the prediction error block that has been lost due to quantization in the conventional example (pixel waveform 234). FIG. 2G shows a decoded pixel waveform 235 of the macroblock pixel waveform 231. In the conventional example, the shape of the intra-prediction block appears as it is, and the pixels of the object 410 are diffused in the portion that was the background in the original image state. On the other hand, in the present embodiment, it can be seen that the background in the original image state is reproduced and diffusion of pixels of the object 410 is suppressed.

  Subsequently, encoding of the macroblock 440 in the present embodiment will be described.

  First, the saturation Sat_Cr and Sat_Cb of the macroblock 440 are calculated. Since GrayLevel = 512, Sat_Cr = (64 × | 512−512 |) / 64 = 0, Sat_Cb = (64 × | 512−) from the pixel values of the Cr and Cb components of the macroblock 440 shown in FIG. 512 |) / 64 = 0. Next, DC values diff_DC_Cr and diff_DC_Cb of the intra prediction error block are calculated. The peripheral pixels used for the intra prediction of the macroblock 440 are the reproduced pixels of the immediately preceding macroblock 430, which is the pixel waveform 235 described with reference to FIG. 3A and 3B show the pixel values of the intra prediction block corresponding to the macro block 440 for the Cr component and the Cb component, respectively. From the drawing, diff_DC_Cr = (64 × | 512−512 |) / 64 = 0 and diff_DC_Cb = (64 × | 512−512 |) / 64 = 0. At this time, since the macroblock 440 does not satisfy the above three conditions for the occurrence of intra prediction error propagation at the same time, Qp correction is not performed on the macroblock, and Qp_MB = 30 is set. The change in the pixel value at this time will be described by focusing on the Cr component.

  In FIG. 3C, the pixel waveform 241 indicates the Cr component pixel of the macroblock 440. The peripheral pixels are reproduction pixels of the macro block 430, which is the pixel waveform 235 described with reference to FIG. FIG. 3D shows an intra prediction block pixel waveform 242 when the optimum prediction mode is selected as 0 (DC prediction). FIG. 3E shows a prediction error block pixel waveform 243 obtained by calculating a difference between the macroblock pixel waveform 241 and the intra prediction block pixel waveform 242. This prediction error block pixel waveform 243 is a target of encoding, but since the prediction error is 0, it becomes 0 even if quantization processing is performed, and the pixel waveform 244 of the restored difference block is also shown in FIG. 0. FIG. 3G shows a decoded pixel waveform 245 of the macroblock pixel waveform 241. In the conventional example, the pixels of the object 410 are diffused in the portion that was the background in the original image state. On the other hand, in the present embodiment, it can be seen that the background in the original image state is reproduced and the diffusion of the pixels of the object 410 is suppressed.

  As described above, in the present embodiment, the degree of occurrence of intra prediction error propagation of the target macroblock is determined from the saturation detected for the color difference component of the target macroblock, the DC value of the intra prediction error, and the picture quantization parameter. By detecting and controlling the quantization parameter of the target macroblock according to the degree of occurrence, it is possible to suppress intra-screen propagation of intra prediction errors that are subjectively undesirable.

  In the present embodiment, saturation is used as visual sensitivity, but the present invention is not limited to this. For example, luminance may be used as visual sensitivity. Further, the saturation detection, DC value detection, and quantization parameter determination of the target macroblock may be realized by software.

  The moving image compression coding apparatus according to the present embodiment is the same as that of H.264. In the H.264 coding, intra prediction error propagation in a screen can be suppressed and subjective image quality can be improved. Therefore, it is useful for AV equipment adopting the same method, a soft encoder that operates on a nonlinear editing machine, and the like. In particular, it can be used for adaptive processing in low bit rate encoding such as Internet streaming.

100, 300 Input unit 101, 301 Difference calculation unit 102, 302 Transform unit 103 Quantization parameter determination unit 104, 304 Quantization unit 105, 305 Entropy encoding unit 106, 306 Output unit 107, 307 Intra prediction mode detection unit 110, 310 Local decoder 111, 311 Inverse quantization unit 112, 312 Inverse transformation unit 113, 313 Addition operation unit 114, 314 Buffer 115, 315 Intra prediction unit 120, 320 Input image 121, 321 Intra prediction image 130 Saturation calculation unit 131 DC Calculation unit 225, 231, 232, 233, 234, 235, 241, 242, 243, 244, 245, 425, 431, 432, 433, 434, 435, 436, 441, 442, 443, 444, 445, 446 pixels Waveform 4 Peripheral pixels used for 0 coded picture 401, 402, 403, 404 slice 410 objects 420,430,440 macroblock 500,501,502,503,504,505 macroblock pixel 510 the encoding target block 511 intraprediction

Claims (5)

An image encoding device that encodes an input image using intra prediction encoding,
A visual sensitivity detector that detects the visual sensitivity of the encoding target block;
A prediction transition detector that detects a transition between the encoding target block and a prediction block corresponding to the encoding target block;
A quantization parameter determining unit that determines a quantization parameter of the encoding target block based on the visual sensitivity, the transition, and a preset picture quantization parameter;
An image encoding device comprising: The visual sensitivity detector
Detecting the saturation of the encoding target block;
The predictive transition detector is
The image coding apparatus according to claim 1, wherein a difference operation between the encoding target block and the prediction block is performed, and a block DC value is calculated for the calculated difference block. The quantization parameter determination unit
The saturation detected by the visual sensitivity detector is smaller than a predetermined first threshold, and the block DC value detected by the predictive transition detector is larger than a predetermined second threshold, and is set in advance. The image encoding device according to claim 2, wherein when the picture quantization parameter is larger than a predetermined third threshold, the quantization parameter of the encoding target macroblock is made smaller than the picture quantization parameter. An image encoding method for encoding an input image using intra prediction encoding,
A visual sensitivity detection step for detecting the visual sensitivity of the block to be encoded;
A prediction transition detection step of detecting a transition between the encoding target block and a prediction block corresponding to the encoding target block;
A quantization parameter determining step for determining a quantization parameter of the encoding target block based on the visual sensitivity, the transition, and a preset picture quantization parameter;
An image encoding method including: A program for causing a computer to execute an image encoding method for encoding an input image using intra prediction encoding,
A visual sensitivity detection step for detecting the visual sensitivity of the block to be encoded;
A prediction transition detection step of detecting a transition between the encoding target block and a prediction block corresponding to the encoding target block;
A quantization parameter determining step for determining a quantization parameter of the encoding target block based on the visual sensitivity, the transition, and a preset picture quantization parameter;
Including programs.

Images