JP4995789B2 - Intra-screen predictive encoding method, intra-screen predictive decoding method, these devices, their programs, and recording media recording the programs - Google Patents

Description

  The present invention relates to an improvement in compression efficiency of intra prediction in image / video coding. Intra-frame predictive coding that performs prediction within the same screen cannot achieve greater compression efficiency than inter-frame predictive coding that performs prediction between different frames. Therefore, an intra-screen predictive coding method with higher compression efficiency is desired.

  Intra-screen prediction in image / video coding is the video coding international standard H.264. H.264 / MPEG-4 AVC (see Non-Patent Document 1), this method is used to remove high-frequency components that are likely to occur at the edge using the correlation between adjacent pixels, The amplitude of the difference can be efficiently suppressed. In-screen prediction is performed in units of blocks in which several pixels are combined. In H.264 / MPEG-4 AVC FRExt (refer to Non-Patent Document 2; hereinafter H.264), three types of blocks of 4 × 4, 8 × 8, and 16 × 16 are included in a 16 × 16 macroblock for a luminance signal. Divide and predict by size.

  For these luminance signals, 9 types of prediction modes can be applied to 4 × 4, 8 × 8 blocks, and 4 types of prediction modes can be applied to 16 × 16 blocks. A prediction mode and a block size with high coding efficiency are determined. In the 4 × 4, 8 × 8 block, each decoded pixel value that intersects with a straight line generated when one type of DC component and a predicted angle of 45 ° to 206.57 ° are divided into eight types at unequal angles. -Generation of a predicted value is realized by referring to the center value between pixels.

  FIG. 20 shows nine types of prediction modes 0 to 8 applicable to a 4 × 4 block used for encoding a luminance signal and their prediction angles.

  Here, in the prediction modes 0 to 2, the values of the pixels A, B, C, and D of the block located on the block to be encoded defined in FIG. 21A and the values defined in FIG. Using the values of the pixels I, J, K, and L of the block located to the left of the encoding target block, the values of the pixels a to p of the encoding target block are predicted.

  In prediction modes 3 to 8, the values of these pixels A, B, C, D, I, J, K, and L and the block located at the upper left of the encoding target block defined in FIG. Based on the value of the pixel X possessed and the values of the pixels E, F, G, and H of the block located at the upper right of the encoding target block defined in FIG. Further, values X ′ and A ′ to L ′ associated with the values of the pixels X and A to L are calculated, and the block located at the lower left of the encoding target block defined in FIG. A value M ′ associated with the value of the pixel M is calculated, and further, values A ″ to F ″ and I ″ to L ″ associated with the center between pixels are calculated, and these values X ′, A ′ to M ′, A ″ to F ″, I ″ to L ″ are used to predict the values of the pixels a to p included in the encoding target block.

  That is, as shown in FIG. 20, in the prediction mode 0, the pixels a to p of the encoding target block are copied by copying the values of the pixels A, B, C, and D in the direction (vertical direction) of the prediction angle 90 °. In the prediction mode 1, the values of the pixels a to p of the encoding target block are obtained by copying the values of the pixels I, J, K, and L in the direction of the prediction angle 180 ° (horizontal direction). In prediction mode 2, the values of pixels a to p of the encoding target block are predicted according to the value of (A + B + C + D + I + J + K + L + 4) / 8.

  In prediction mode 3, the reference values (B ′ to H ′) calculated in association with the pixel values are copied in the direction of the prediction angle 45 ° to predict the values of the pixels a to p of the block to be encoded. In the prediction mode 4, the reference values calculated in association with the pixel values are copied in the direction of the prediction angle 135 °, thereby predicting the values of the pixels a to p of the encoding target block.

  In prediction mode 5, the reference value calculated in association with the pixel value and the reference value calculated in association with the center between the pixels are copied in the direction of the prediction angle 116.57 ° to have the encoding target block. The values of the pixels a to p are predicted, and in the prediction mode 6, the reference value calculated in association with the pixel value and the reference value calculated in association with the center between the pixels are copied in the direction of the prediction angle 153.43 °. Thus, the values of the pixels a to p of the block to be encoded are predicted. In the prediction mode 7, the reference value calculated in the form corresponding to the pixel value and the reference value calculated in the form corresponding to the center between the pixels are set to the prediction angle 63. The value of the pixels a to p of the encoding target block is predicted by copying in the direction of .43 °, and in the prediction mode 8, the reference value calculated in association with the pixel value is associated with the center between the pixels. The calculated reference values to predict the value of a pixel a~p with the encoding target block by copying in the direction of the prediction angle 206.57 °.

  In this way, in the prior art, when performing intra prediction encoding, in the 4 × 4 and 8 × 8 blocks, in addition to generating a prediction signal by referring to one type of DC component, , 45 ° to 206.57 ° prediction angle range can be generated by referring to each decoded pixel value / inter-pixel center value that intersects with a straight line generated when the prediction angle range is divided into eight types at unequal angles. Realized.

  FIG. 22 shows a flowchart of the encoding process in the conventional video encoding method. Hereinafter, the conventional video encoding process will be described with reference to FIG.

First, in step S801, the Lagrangian cost minimum value L _min is initialized to + ∞, and the bit string R _best when the Lagrangian cost takes the minimum value is initialized to 0. Is predicted (hereinafter referred to as Pred_PredMode). Next, in step S803, a loop for detecting the prediction mode number n that minimizes the Lagrangian cost is performed (since the in-screen prediction of the conventional method has 9 patterns, n∈ [0, 1,. 8] loop).

  In this loop, first, in step S804, the bit string R generated in the prediction mode n is initialized with 0, and in step S805, conditional branching is performed by comparing Pred_PredMode with n. If they match, a 1-bit prediction flag is added to R in step S806. If they do not match, 1-bit prediction flag is added to R and 3 bits indicating the n prediction direction in step S807. Add a correction value.

Subsequently, in step S808, a prediction value of each pixel in the encoding target block is generated to obtain a prediction residual, and in step S809, the prediction residual is subjected to DCT, quantization, and encoding, and the generated bit is converted to R Append to In step S810, a Lagrangian cost L _cost (see Non-Patent Documents 3 and 4) is calculated from the error between the number of R bits and the original image.

Next, the comparison of the Lagrangian cost minimum L _min and L _cost in step S811, if L _cost is smaller than L _min, the L _min to L _cost at step S812, updates the R _best to R. The processes of steps S804 to S812 are repeated for each prediction mode n from n = 0 to 8.

  In the conventional method, in step S802, as shown in FIG. 23, the prediction mode of the block adjacent to the encoding target block (hereinafter referred to as Block A) is PredModeA, and the prediction mode of the block adjacent to the left (hereinafter Block B) is PredModeB. In other words, Pred_PredMode is determined by the following expression 1 (described in Non-Patent Document 1, p. 102, Expression 8-42).

Pred_PredMode
= Min (PredModeA, PredModeB) (Formula 1)
That is, the smaller one of the prediction modes of Block A and Block B is used as the prediction value (Pred_PredMode) of the prediction mode.

Further, a 1-bit flag is used to predict the prediction mode of the encoding target block from the prediction mode in the encoded block adjacent to the encoding target block. If the prediction mode in the encoding target block matches Pred_PredMode, it can be described with only a 1-bit flag as shown in step S806, and if different, in addition to the 1-bit flag as shown in step S807. In order to perform fixed-length encoding using a code indicating any of the remaining 8 types as a 3-bit correction value, the code is described in a total of 4 bits.
ITU-T Rec. H.264 | ISO / IEC 14496-10 version 1 (2003) ITU-T Rec. H.264 | ISO / IEC 14496-10 version 4 (2005) H.264 Reference Software JM ver. 13.2 Encoder, http://iphome.hhi.de/suehring/tml/, 2008 H.264 Reference Software KTA ver.1.9 Encorder, http://iphome.hhi.de/suehring/tml/, 2008

  In the conventional method, since the prediction value of the prediction mode in the intra prediction is generated using Equation 1, for example, the prediction in the vertical direction (prediction mode 0: Intra4 × 4PredMode0) illustrated in FIG. 20 is performed in either PredMode A or B. If present, the vertical prediction is most preferentially applied as Pred_PredMode.

  24-26 is a figure for demonstrating the comparison of the prediction residual by the prediction modes 0 and 7 applied. 24 shows the pixel value of the original image, FIG. 25 shows the prediction pixel value and the prediction residual when the encoding target block is intra-coded in prediction mode 0, and FIG. 26 shows the encoding target block in the prediction mode. 7 shows a predicted pixel value and a predicted residual when intra-screen coding is performed.

  Now, consider the situation where the lower 4 × 4 block of the current image in FIG. 24 is encoded. It is assumed that PredModeA is predicted in prediction mode 7 and PredModeB is predicted in prediction mode 0. At this time, in the conventional method, the prediction mode 0 is selected for Pred_PredMode. However, when focusing on the edge of the image, the encoding target block does not require a 3-bit correction value when the prediction mode 7 is Pred_PredMode. Can be easily predicted.

  As described above, in the prediction value generation method of the prediction mode in the intra-screen prediction of the conventional method, there are many cases in which the prediction is out of order. Since description is required, there is a high possibility that the code amount will increase. As a result, the code amount increases.

  An object of the present invention is to provide an effective prediction mode prediction method in intra-screen prediction while suppressing the above-described situation in which prediction is lost without overhead.

  As a method for suppressing the case where the prediction mode is not correctly predicted, the present invention pays attention to the prediction vector determined by the prediction mode in the encoded block adjacent to the encoding target block. The presence or absence of an inflow vector is used as a selection key.

  Here, the prediction vector is a vector representing the prediction direction determined from the prediction mode in the intra prediction, and is determined from the prediction mode in the intra prediction as shown in the example of the case of 4 × 4 blocks in FIG. Means a direction corresponding to the predicted angle (see FIG. 20).

  Further, when the prediction vector determined from the prediction mode of the encoded block is directed toward the center of the block to be encoded, the prediction vector is referred to as an inflow vector. Whether or not the prediction vector is directed toward the center of the block to be encoded is determined by referring to a prediction angle determined from the prediction mode and comparing with a threshold value described in an embodiment described later.

In-screen prediction is a technique that effectively suppresses the amplitude of the difference by removing the edge mainly by suppressing the prediction residual, thereby effectively removing the high-frequency components generated around the edge. Therefore, assuming that the prediction residual in the already encoded block has been successfully suppressed by intra prediction, it is assumed that the encoded block is likely to have an edge along its prediction vector. . In the present invention, when generating a prediction value of a prediction mode, when there are a plurality of encoded blocks with an inflow vector, an edge in a reference pixel in the encoded block is detected, and a prediction of a direction in which the edge appears most strongly is performed. The prediction mode of the encoded block having the mode is set as the prediction value of the prediction mode of the encoding target block.

For example, when the prediction angle θ _{A in} PredMode _A exists around 90 ° (when there is an inflow vector toward the center of the block to be encoded), reference pixels A to D of the block to be encoded (see FIGS. 20 and 21). There is a high possibility that an edge will appear on top. Therefore, considering the continuity of pixels in a natural image, it is highly possible that edges are continuous in the direction of θ _A even in the encoding target block, and it is considered more appropriate to apply PredModeA as Pred_PredMode. If PredModeB is predicted in the prediction mode 0 (Intra4 × 4PredMode0) and there is no inflow vector in the direction of the encoding target block, an edge on the reference pixels I to L (see FIGS. 20 and 21) of the encoding target block. Since PredModeB is applied as Pred_PredMode, there is a high possibility that prediction mode will be unpredictable.

  In this way, it is possible to determine the presence of an edge by determining the presence or absence of an inflow vector toward the center of the encoding target block, and to realize an efficient prediction mode generation by predicting an appropriate prediction mode. To do.

  As described above, by using the present invention, by determining whether there is an inflow vector toward the center of the block to be encoded, which is not considered in the prediction value generation method of the prediction mode in the conventional intra prediction, the edge is determined. It is possible to realize prediction value generation in a prediction mode more efficiently than a conventional method in a moving image in which the images are continuous.

  As a result, the probability that a block that requires 4 bits for determining the prediction mode for intra prediction in the conventional method can be represented by 1 bit increases, leading to a reduction in the amount of codes and the encoding efficiency of intra prediction. Will improve. In particular, in the low bit rate encoding, the effect of 3 bits reduction is significant because it appears more conspicuously than the high bit rate.

  Further, according to the present invention, since the prediction of the prediction mode of the intra-screen prediction does not depend on the prediction mode number as in the conventional method, it depends on the prediction angle of the prediction vector determined from the prediction mode number. It has high affinity for prediction modes other than the prediction value generation method for the prediction mode of the prediction mode for intra-screen prediction based on H.264. Furthermore, it can be used not only as a 4 × 4 block but also as a prediction value generation method for a prediction mode in an 8 × 8 block.

  Prior to describing embodiments of the present invention in detail, an outline thereof will be described in comparison with the prior art.

  As described above, in the prediction value generation method of the prediction mode in the conventional intra-screen prediction, as shown in FIG. 23, the smaller prediction mode number of the prediction mode in the upper and left adjacent blocks is used as the prediction value of the prediction mode. Applicable.

  H. In H.264, nine patterns shown in FIG. 20 are predicted for a 4 × 4 encoding target block, and a flag (1) is used to encode a prediction mode that matches the prediction value of the prediction mode. Bit) + prediction residual DCT code data, and when encoding a prediction mode different from the prediction value, flag (1 bit) + prediction mode correction value (3 bits) + prediction residual The result is expressed as DCT code data. Therefore, the code amount increases when the prediction value of the prediction mode deviates.

  Therefore, in order to improve the accuracy of prediction value generation in the prediction mode, the present invention focuses on the prediction vector determined by the prediction mode in the encoded block adjacent to the encoding target block, and moves toward the encoding target block center direction. The presence / absence of a vector (inflow vector) is determined, and the prediction value of the prediction mode of the encoding target block is selected by selecting one of the prediction modes of one or more encoding target blocks determined to have the inflow vector. Generate.

As a process for realizing this, it has a new processing mechanism in the following two stages.
[Process 1] Extract encoded blocks with inflow vectors and list them.
[Process 2] A priority order is set for the listed encoded blocks, and the prediction mode of the block with the highest priority order is applied as the prediction value of the prediction mode of the encoding target block.

  In the extraction of an encoded block with an inflow vector in [Process 1] above, it is given from the outside which block to determine the presence or absence of an inflow vector among blocks adjacent to an encoding target block, or It can be changed by a preset value (referred to as SwKey).

For example, there are the following patterns according to SwKey.
When SwKey = 1, prediction mode prediction is performed using the Block A and B encoded blocks shown in FIG. 3 (described as the first embodiment).
When SwKey = 2, the prediction mode is predicted using the encoded blocks of Block A, B, and F shown in FIG. 3 (described as Example 2).
When SwKey = 3, prediction mode prediction is performed using one or both of the blocks A, B, F, E, and G shown in FIG. 3 (described as Example 3).
When SwKey = 4, the prediction mode is predicted using the encoded blocks of Block A to H shown in FIG. 3 (described as Example 4).

As a method for selecting the encoded block listed in [Process 2] according to the priority order, for example, the following method can be used.
[Selection method by determining priority order by block distance (described as Example 5)]
Priorities are set in advance in adjacent blocks so that the blocks closer to the encoding target block are preferentially selected, and the block with the highest priority among the listed encoded blocks Select.
[Selection Method by Priority Order Determination by Edge Detection of Reference Pixel (described as Example 6)]
The edge strength at the reference pixel is calculated, and the direction with the strongest edge is selected.

  The accuracy of prediction value generation in the prediction mode is improved by the above processing, so that it is possible to increase the number of blocks in which the prediction mode can be expressed with the 1-bit flag described in the related art. Therefore, the correction value (3 bits) added when the prediction value is different from the prediction value in the prediction mode can be reduced. Note that the accuracy of predicted value generation is improved. This means that the probability that the prediction mode number selected by the Lagrangian cost calculation formula used in H.264 matches the prediction value of the prediction mode is increased. Thus, by improving the accuracy of prediction value generation in the prediction mode, it is possible to reduce the amount of code under the condition of constant image quality, improve the image quality under the condition of constant code amount, and improve the coding efficiency. Is possible.

  The significance of the above [Process 1] will be supplementarily described. In the conventional method, the prediction value of the prediction mode is generated according to the prediction mode number of intra prediction, but the continuity of the edges that can be estimated from the prediction mode number is not taken into consideration. In the above [Process 1], it is determined whether or not there is an inflow vector from the adjacent encoded block toward the center of the encoding target block, and one of the prediction modes of one or more encoding target blocks determined to have an inflow vector. Since one is selected from the above, it is possible to generate a prediction value of the prediction mode by focusing on the continuity of the edges that can be estimated from the prediction mode number, and it is possible to generate a prediction value of the prediction mode with higher accuracy Become.

  Further, by determining the encoded block for determining the presence or absence of the inflow vector according to the set value of SwKey, an edge may enter the encoding target block center direction from a plurality of adjacent encoded blocks. A search for a high-encoded block can be performed efficiently.

  In [Process 2], the priority order is set for the blocks listed in [Process 1], and the prediction mode of the block with the highest priority order is applied as the prediction value of the prediction mode of the encoding target block. . This is because [Process 1] alone selects one of two or more prediction modes when there are two or more blocks that are judged to have “inflow vectors” from adjacent coded blocks. This is because it is necessary to use the prediction value of the prediction mode of the encoding target block. By applying [Process 2], one prediction mode is applied as a prediction value of the prediction mode of the encoding target block from among a plurality of encoded blocks that are highly likely to have edges entering the encoding target block center direction. Is possible.

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

  FIG. 1 is a block diagram showing an embodiment of an intra prediction encoding apparatus to which the present invention is applied. The encoding device 10 that uses the prediction value generation method of the prediction mode in the intra prediction according to the present embodiment includes the Lagrange cost minimum value and block encoded data initialization unit 11 and an encoded block adjacent to the encoding target block. Are composed of a prediction mode prediction unit 12 for predicting a prediction mode and a prediction mode determination loop unit 13 for determining a prediction mode.

  The prediction unit 12 in the prediction mode includes a block extraction unit 121 with an inflow vector and a block selection / prediction value generation unit 122.

  The prediction mode determination loop unit 13 includes a prediction mode encoding unit 131, an intra prediction unit 132 for generating a prediction residual by intra prediction, an integer precision DCT / quantization / encoding unit 133 for the prediction residual, It comprises a Lagrangian cost determination unit 134 that performs minimum value determination of a Lagrangian cost, and a loop end determination unit 135.

  FIG. 2 shows a flowchart of the encoding process in the intra-screen encoding system of one embodiment of the present invention. In the present embodiment, the step of determining Pred_PredMode in the conventional encoding algorithm (step S802 in FIG. 22) is replaced with steps S102 and S103 in FIG.

First, in step S101, as in step S801, the initial value of the Lagrangian cost minimum value L _min is set to + ∞, and the initial value of the bit string R _best when the Lagrangian cost takes the minimum value is set to 0. In step S102, a processing branch coefficient SwKey is acquired. This SwKey is a coefficient that is arbitrarily set by the user in the range of 1-4.

  In this example, four types of processing methods for determining Pred_PredMode are classified according to a preset SwKey value. However, the present invention is limited to one of the four types of processing methods described below. In this case, the processing method selection process based on the SwKey determination can be omitted.

  In step S103, the value of SwKey is determined, processing by a processing algorithm corresponding to SwKey is executed, and Pred_PredMode is determined. Details of the process of determining Pred_PredMode in accordance with the SwKey will be described later as Embodiments 1 to 4.

  Next, the processing in steps S105 to S113 is repeated for each prediction mode from prediction mode 0 to (MaxPred-1) by the loop processing in step S104. Note that MaxPred in step S104 represents the maximum value in the prediction direction. The processing after step S104 is the same as the processing after step S803 in FIG.

  In the present embodiment, it is assumed that prediction modes of up to eight encoded blocks adjacent to the encoding target block are referred to in step S103. Here, the blocks adjacent to the encoding target block are set as shown in FIG.

[Example 1]
The first embodiment is an example of processing (prediction value generation processing) for determining Pred_PredMode when SwKey = 1. In the case where prediction mode prediction is performed using the Block A and B encoded blocks shown in FIG. It is an example.

FIG. 4 shows an angle threshold used for determining the presence or absence of an inflow vector from an adjacent block. Here, θ _A and θ _B represent prediction angles determined from prediction modes in Block _A and Block _B , respectively. That is, for the angle threshold (45 °, 135 °, 225 °),
45 ° ≦ θ _A <135 °
135 ° ≦ θ _B <225 °
Whether or not there is an inflow vector is determined depending on whether or not

  FIG. 5 shows a flowchart of the predicted value generation process of the first embodiment. First, in step S201, the prediction value Pred_PredMode of the prediction mode is initialized with a DC component, the buffer BlockBuf that stores the block number in which the inflow vector exists is initialized with 0, and the number MaxElem with the inflow vector exists with 0.

  Next, in Step S202, a loop of TargetBlockε [BlockA, B] adjacent to the encoding target block is performed, and in Step S203, it is branched whether or not TargetBlock is an encoded block. For only the true TargetBlock in this branch, the predicted angle θ is calculated from the PredMode of the TargetBlock in Step S204.

  Subsequently, in step S205, branch processing is performed to determine whether θ is within the threshold in FIG. 4, and in step S206, TargetBlock is additionally stored in BlockBuf, and MaxElem is incremented by one for true TargetBlock. When the extraction of the block in which the inflow vector exists is completed in the loop of steps S202 to S206, one optimum prediction mode is selected from BlockBuf in step S207 (the example of the selection method is “Examples 5 and 6”). ”), And applies to Pred_PredMode.

[Example 2]
The second embodiment is an example of processing (prediction value generation processing) for determining Pred_PredMode when SwKey = 2, and prediction mode prediction is performed using the encoded blocks of Block A, B, and F shown in FIG. This is an example.

FIG. 6 shows an angle threshold used for determining the presence or absence of an inflow vector from an adjacent block. Here, θ _A , θ _B , and θ _F represent prediction angles determined from the prediction modes in Block A, B, and F, respectively. That is, for the angle threshold (45 °, 112.5 °, 157.5 °, 225 °),
45 ° ≦ θ _A <112.5 °
112.5 ° ≦ θ _F <157.5 °
157.5 ° ≦ θ _B <225 °
Whether or not there is an inflow vector is determined depending on whether or not

  FIG. 7 shows a flowchart of predicted value generation processing according to the second embodiment. First, in step S301, the prediction value Pred_PredMode of the prediction mode is initialized with the DC component, the buffer BlockBuf that stores the block number in which the inflow vector exists is initialized to 0, and the number MaxElem in which the inflow vector exists is initialized to 0.

  Next, in Step S302, a loop of TargetBlockε [BlockA, B, F] adjacent to the encoding target block is performed, and branching is performed in Step S303 as to whether or not the TargetBlock is an encoded block. For only the true TargetBlock in this branch, the predicted angle θ is calculated from the PredMode of the TargetBlock in Step S304.

  Subsequently, in step S305, branch processing is performed to determine whether θ is within the threshold in FIG. 6, and TargetBlock is additionally stored in BlockBuf in step S306 with respect to the true TargetBlock, and MaxElem is incremented by one. When the extraction of the block in which the inflow vector exists is completed in the loop of steps S302 to S306, one optimum prediction mode is selected from BlockBuf in step S307 (the example of the selection method is “Examples 5 and 6”). ”), And applies to Pred_predMode.

Example 3
The third embodiment is an example of processing (predicted value generation processing) for determining Pred_PredMode when SwKey = 3, and one or both of the blocks A, B, F and E, G shown in FIG. It is an example in the case of performing prediction of prediction mode using.

FIG. 8 shows the angle threshold used to determine the presence or absence of an inflow vector from an adjacent block using Blocks A, B, E, F, and G. FIG. 9 shows the angle threshold value from the adjacent block using Blocks A, B, E, and F. FIG. 10 shows angle thresholds used to determine the presence or absence of an inflow vector from adjacent blocks using Blocks A, B, F, and G. FIG. Here, θ _A , θ _B , θ _E , θ _F , and θ _G in FIGS. 8, 9, and 10 represent prediction angles determined from the prediction modes in Block A, B, E, F, and G, respectively.

In the example of FIG. 8, with respect to the angle threshold (22.5 °, 67.5 °, 112.5 °, 157.5 °, 202.5 °, 247.5 °),
22.5 ° ≦ θ _E <67.5 °
67.5 ° ≦ θ _A <112.5 °
112.5 ° ≦ θ _F <157.5 °
157.5 ° ≦ θ _B <202.5 °
202.5 ° ≦ θ _G <247.5 °
Whether or not there is an inflow vector is determined depending on whether or not

Moreover, in the example of FIG. 9, with respect to the angle threshold (22.5 °, 67.5 °, 112.5 °, 157.5 °, 225 °),
22.5 ° ≦ θ _E <67.5 °
67.5 ° ≦ θ _A <112.5 °
112.5 ° ≦ θ _F <157.5 °
157.5 ° ≦ θ _B <225 °
Whether or not there is an inflow vector is determined depending on whether or not

In the example of FIG. 10, for the angle threshold values (45 °, 112.5 °, 157.5 °, 202.5 °, 247.5 °),
45 ° ≦ θ _A <112.5 °
112.5 ° ≦ θ _F <157.5 °
157.5 ° ≦ θ _B <202.5 °
202.5 ° ≦ θ _G <247.5 °
Whether or not there is an inflow vector is determined depending on whether or not

  FIG. 11 shows a flowchart of a predicted value generation process when prediction mode prediction is performed using Block A, B, E, F, and G as an example.

  First, in step S401, the prediction value Pred_PredMode of the prediction mode is initialized with the DC component, the buffer BlockBuf that stores the block number in which the inflow vector exists, and the number MaxElem in which the inflow vector exists is set to 0.

  Next, in Step S402, a loop of TargetBlockε [Block A, B, E, F, G] adjacent to the encoding target block is performed, and branching is performed in Step S403 as to whether or not the TargetBlock is an encoded block. In this branch, only the true TargetBlock is calculated, and the predicted angle θ is calculated from the PredMode of the TargetBlock in Step S404.

  Subsequently, in step S405, branch processing is performed to determine whether θ is within the threshold value in FIG. 8, and in step S406, TargetBlock is additionally stored in BlockBuf, and MaxElem is incremented by 1 for true TargetBlock. When the extraction of the block having the inflow vector is completed in the loop of steps S402 to S406, one optimum prediction mode is selected from BlockBuf in step S407 (the example of the selection method is “Examples 5 and 6”). ”), And applies to Pred_PredMode.

Example 4
The fourth embodiment is an example of processing (prediction value generation processing) for determining Pred_PredMode when SwKey = 4, and when prediction mode prediction is adaptively performed using all the blocks A to H shown in FIG. It is an example.

FIG. 12 shows angle thresholds used to determine the presence or absence of an inflow vector from an adjacent block using Blocks A to H. Here, θ _{A to} θ _H represent prediction angles determined from prediction modes in Blocks _{A to H} , respectively.

In the example of FIG. 12, the angle threshold (22.5 °, 67.5 °, 112.5 °, 157.5 °, 202.5 °, 247.5 °, 292.5 °, 337.5 °) is set. for,
22.5 ° ≦ θ _E <67.5 °
67.5 ° ≦ θ _A <112.5 °
112.5 ° ≦ θ _F <157.5 °
157.5 ° ≦ θ _B <202.5 °
202.5 ° ≦ θ _G <247.5 °
247.5 ° ≦ θ _C <292.5 °
292.5 ° ≦ θ _H <337.5 °
337.5 ° ≦ θ _D <22.5 °
Whether or not there is an inflow vector is determined depending on whether or not

  FIG. 13 shows a flowchart of predicted value generation processing of this embodiment. First, in step S501, the prediction value Pred_PredMode of the prediction mode is initialized with a DC component, the buffer BlockBuf that stores the block number in which the inflow vector exists, 0, and the number MaxElem in which the inflow vector exists is initialized with 0. Next, in Step S502, a loop of TargetBlockε [BlockA, B,..., H] adjacent to the encoding target block is performed, and in Step S503, a branch process is performed to determine whether or not the TargetBlock is an encoded block. .

  Subsequently, in step S504, the predicted angle θ is calculated from the PredMode of TargetBlock, and in step S505, branch processing is performed to determine whether θ is within the threshold in FIG. In step S506, TargetBlock is additionally stored in BlockBuf, and MaxElm is incremented by 1, for both true SBlocks in steps S503 and S505. When the extraction of the block in which the inflow vector exists is completed in the loop of steps S502 to S506, one optimal prediction mode is selected from BlockBuf in step S507 (the example of the selection method is “Examples 5 and 6”). ”), And applies to Pred_PredMode.

[Example 5 (selection method based on priority order determination by block distance)]
As an example, a method for generating a prediction value of a prediction mode based on a block distance when Blocks A, B, E, and F (see FIG. 3) adjacent to a 4 × 4 encoding target block are stored in BlockBuf is described. .

  The blocks closest to the encoding target block are Block A and B, and Block E and F are far apart from Block A and B. Therefore, it can be determined that Block A and B have higher inter-pixel correlation. Since the reference pixels (RefPixels) E to H in Block E are far from the encoding target block, it is possible to determine that the inter-pixel correlation is lower than that in Block F. Although the block distances of Block A and B are equal, for the sake of convenience, it is assumed that Block A has a higher priority, and the priority order of Block A, B, E, and F is defined by the following inequality.

Block A> Block B> Block F> Block E (Formula 2)
When the priority order defined above is followed, BlockA's PredMode is applied as BestPredMode. Further, assuming that BlockB and E are stored in BlockBuf, the PredMode of BlockB is applied as BestPredMode. By setting the priority order in advance to adjacent blocks in this way, it becomes possible to select Pred_PredMode as one even in an encoding target block having a plurality of inflow vectors.

[Embodiment 6 (Selection Method by Priority Order Determination by Edge Detection of Reference Pixel)]
As an example of predicting a prediction mode that emphasizes edge strength, edge detection is performed when Block A, B, E, and F (see FIG. 3) adjacent to a 4 × 4 encoding target block are stored in BlockBuf. FIG. 14 is a flowchart showing an example of processing of a method for generating a prediction value of the prediction mode by the method.

First, in step S601, BlockBuf, MaxElem, receive Pred_PredMode, initialized to 0 the maximum edge intensity E _max at step S602. Next, in step S603, the edge strength EdgePower∈ [A ″ to L ″, X ″] is calculated for the same pixel as the reference pixel ∈ [A to L, X] in FIG. Use an extraction filter.

  An example of the contour extraction filter is shown in Equation 3.

EdgePower = | i−x | + | j−x | (Formula 3)
Here, x represents a pixel value in a pixel for obtaining EdgePower, and i and j represent pixel values in adjacent pixels (reference pixels ∈ [A to L, X]) on the left, right, and top and bottom of x. When calculating EdgePower, it is desirable to use values obtained by removing noise for i, j, and x. However, since the reference pixel used in the intra prediction is the pixel value before applying the deblocking filter, it is particularly referred to. In pixels A, D, E, I, and X, an edge is easily detected more than necessary. Therefore, strong noise reduction is applied to these pixels, especially at low bit rates.

In step S604, a loop of the blocks accumulated in BlockBuf is performed. In step S605, the PredMode of TargetBlock stored in the IDth block in BlockBuf is acquired. In step S606, the edge strength E _block in the encoding target block is initialized to zero.

  Next, in step S607, a loop is performed for all prediction pixels (hereinafter, PredPixel) in the encoding target block. In step S608, EdgePower existing at the reference pixel position of PredMode is referred to, and EdgePower's influence EdgeInfluence at the coordinate value of PredPixel is calculated. It is assumed that EdgeInfluence is inversely proportional to the square of the distance with respect to the reference pixel, and the distance between the pixels is set to 1 and calculation is performed using the following Expression 4.

EdgeInfluence
= | EdgePower | / RefLength ² (Formula 4)
Here, RefLength represents the distance between the PredPixel and the reference pixel as shown in FIG.

In step S609, EdgeInfluence is added to E _block . It performs a comparison of the E _max and E _block in step S610, the larger the better of E _block, to update the E _max in E _block at step S611, to update the Pred_PredMode in PredMode.

  As a result, the direction with the strongest edge can be selected as Pred_PredMode even in the encoding target block in which a plurality of inflow vectors exist for the encoding target block.

  FIG. 16 is a block diagram showing an embodiment of the intra prediction decoding apparatus to which the present invention is applied. The decoding apparatus 20 that uses the prediction value generation method of the prediction mode in the intra prediction according to the present embodiment includes a prediction mode prediction unit 21 for predicting a prediction mode from a decoded block adjacent to a decoding target block, and encoding. A prediction mode determination unit 22 that determines a prediction mode from data, a residual signal decoding unit 23 that decodes a prediction residual from encoded data, and a decoded image generation unit 24 are configured.

  The prediction unit 21 in the prediction mode includes a block extraction unit 211 having an inflow vector and a block selection / prediction value generation unit 212.

  FIG. 17 shows a flowchart of the decoding process in the video decoding system according to the embodiment of the present invention.

  First, in step S701, a 1-bit prediction flag F indicating whether the prediction values in the prediction mode match is acquired. Next, in step S702, the processing branch coefficient SwKey is acquired in the same manner as in step S102 at the time of encoding in FIG. 2, and then in step S703, the value of SwKey is determined in the same manner as in step S103 in the encoding in FIG. And processing by the processing algorithm (the first to fourth embodiments described above) according to SwKey is performed to determine Pred_PredMode.

  Subsequently, in step S704, a conditional branch is performed to determine whether or not the prediction flag F is predicted. If the prediction is correct, PredMode is set to Pred_PredMode in step S705, and if not, the correction value of the prediction mode is acquired and applied to PredMode in step S706.

  Subsequently, a prediction value is generated from PredMode in step S707, and prediction residual code data D is acquired in step S708. In step S709, the prediction residual is decoded from the prediction residual code data D. Finally, in step S710, a decoded signal is generated by adding the prediction value and the prediction residual.

[Experimental example]
FIG. 18 shows an RD curve (average value of 60 frames) in the test sequence. The embodiment according to the proposed method of the present invention is compared with JM ver. 11.0 KTA1.9 (see Non-Patent Document 4) and the results of performance comparison between the conventional method and the present invention are as shown in FIG. 18 (all the compared data are only luminance signals). . The data is an average value when all 60 frames are encoded by intra prediction. The sequence was “FOREMAN” and the resolution was CIF (352 × 288).

  FIG. 19 shows the predictive predictability applied as the prediction mode in which the prediction value of the prediction mode has the lowest Lagrangian cost in the experiment of FIG. The higher the predictive predictive value, the more the number represented by a 1-bit flag increases, leading to a reduction in code amount.

  In this experiment, the method described as Example 2 and Example 6 was used, and the proposed method was implemented not only for 4 × 4 block but also for 8 × 8 block prediction mode generation. As a result, in the above experiment, a code amount reduction of about 2.93% was confirmed when PSNR = 35 [dB], and a code amount reduction of about 5.01% was confirmed when PSNR = 32 [dB].

  The intra-screen prediction process in the above embodiment can be realized by a computer and a software program, and the program can be recorded on a computer-readable recording medium or provided through a network.

It is a block diagram which shows one Embodiment of the prediction encoding apparatus in a screen. It is a flowchart of an encoding process. It is a figure explaining the setting of the block number of an adjacent block. It is a figure which shows the example of the angle threshold value used for determination of the presence or absence of the inflow vector which made BlockA and B object. It is a flowchart of the prediction value generation process of prediction mode. It is a figure which shows the example of the angle threshold value used for judgment of the presence or absence of the inflow vector which made BlockA, B, and F object. It is a flowchart of the prediction value generation process of prediction mode. It is a figure which shows the example of the angle threshold value used for determination of the presence or absence of the inflow vector for BlockA, B, E, F, G. It is a figure which shows the example of the angle threshold value used for determination of the presence or absence of the inflow vector which made BlockA, B, E, and F object. It is a figure which shows the example of the angle threshold value used for determination of the presence or absence of the inflow vector which made BlockA, B, F, and G object. It is a flowchart of the prediction value generation process of prediction mode. It is a figure which shows the example of the angle threshold value used for the judgment of the presence or absence of the inflow vector which made BlockAH object. It is a flowchart of the prediction value generation process of prediction mode. The flowchart of the prediction value production | generation process of the prediction mode by edge detection. It is explanatory drawing of the distance between pixels. It is a block diagram which shows one Embodiment of the prediction decoding apparatus in a screen. It is a flowchart of a decoding process. It is a figure which shows the RD curve (60 frame average value) in a test sequence. It is a figure which shows the predictive probability applied as a prediction mode with the prediction value of a prediction mode having the lowest Lagrange cost. It is explanatory drawing of the prediction angle used with the conventional intra-screen prediction method. It is explanatory drawing of the pixel value of the reference pixel used with the conventional prediction method in a screen. It is a flowchart of the encoding process in a conventional method. It is a figure explaining the adjacent block used for prediction of the prediction mode in a conventional method. It is a figure which shows the pixel value of an original image for demonstrating the comparison of the prediction modes 0 and 7 applied. It is a figure which shows the example of a prediction pixel value when a coding object block is intra-coded in prediction mode 0, and a prediction residual. It is a figure which shows the example of a prediction pixel value when a coding object block is intra-coded in the prediction mode 7, and a prediction residual.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Encoding apparatus 11 Lagrange cost minimum value and block coding data initialization part 12 Prediction mode prediction part 121 Block extraction part with inflow vector 122 Block selection and prediction value generation part 13 Prediction mode determination loop part 131 Prediction mode code Encoding unit 132 intra prediction unit 133 integer precision DCT / quantization / coding unit 134 Lagrangian cost determination unit 135 loop end determination unit 20 decoding device 21 prediction mode prediction unit 211 block extraction unit with inflow vector 212 block selection and Prediction value generation unit 22 Prediction mode determination unit 23 Residual signal decoding unit 24 Decoded image generation unit

Claims (10)

An intra-screen predictive encoding method for encoding an image using intra-screen prediction,
By comparing the prediction angle determined from the prediction mode in the encoded block adjacent to the encoding target block with a predetermined threshold value, it is determined whether or not there is an inflow vector in which the prediction vector indicating the prediction direction is directed toward the center of the encoding target block. Extracting the encoded block with the inflow vector,
Possess a process of generating a prediction value of a prediction mode of the encoding target block by selecting one of the prediction modes of encoded blocks of there inflow vector is the extraction,
In the process of generating the prediction value of the prediction mode,
When there are a plurality of encoded blocks with the inflow vector, an edge of a reference pixel in the encoded block is detected, and a prediction mode of the encoded block having a prediction mode in a direction in which the edge appears most strongly, An intra-screen predictive encoding method, wherein a prediction value of a prediction mode of an encoding target block is used. In the intra prediction encoding method according to claim 1,
In the process of extracting the encoded block with the inflow vector,
In accordance with a predetermined setting value or a setting value given from the outside, an encoded block to be determined as to whether or not the inflow vector exists is selected, and the presence or absence of the inflow vector is determined only for the selected encoded block. An intra-screen predictive encoding method characterized by determining. An intra-screen prediction decoding method for decoding an image using intra-screen prediction,
By comparing the prediction angle determined from the prediction mode in the decoded block adjacent to the decoding target block with a predetermined threshold, it is determined whether or not there is an inflow vector in which the prediction vector indicating the prediction direction is directed to the decoding target block center direction. Extracting a decoded block with a vector;
Possess a process of generating a prediction value of a prediction mode of the block to be decoded by selecting one of the prediction mode of the decoded blocks there inflow vector is the extraction,
In the process of generating the prediction value of the prediction mode,
When there are a plurality of decoded blocks having the inflow vector, an edge of a reference pixel in the decoded block is detected, and a prediction mode of a decoded block having a prediction mode in a direction in which the edge appears most strongly is determined as a decoding target block An intra-screen predictive decoding method, characterized in that the prediction value of the prediction mode is used. In the intra prediction decoding method according to claim 3 ,
In the process of extracting the decoded block with the inflow vector,
In accordance with a predetermined setting value or a setting value given from the outside, a decoded block to be determined as to whether or not there is an inflow vector is selected, and the presence or absence of an inflow vector is determined only for the selected decoded block An intra-screen predictive decoding method characterized by An intra-screen predictive encoding device that encodes an image using intra-screen prediction,
By comparing the prediction angle determined from the prediction mode in the encoded block adjacent to the encoding target block with a predetermined threshold value, it is determined whether or not there is an inflow vector in which the prediction vector indicating the prediction direction is directed toward the center of the encoding target block. A means for extracting a coded block with an inflow vector;
E Bei and means for generating a prediction value of a prediction mode of the encoding target block by selecting one of the prediction modes of encoded blocks of there inflow vector is the extraction,
The means for generating the prediction value of the prediction mode is:
When there are a plurality of encoded blocks with the inflow vector, an edge of a reference pixel in the encoded block is detected, and a prediction mode of the encoded block having a prediction mode in a direction in which the edge appears most strongly, An intra-screen predictive coding apparatus characterized by using a prediction value of a prediction mode of an encoding target block . An intra-screen predictive decoding device that decodes an image using intra-screen prediction,
By comparing the prediction angle determined from the prediction mode in the decoded block adjacent to the decoding target block with a predetermined threshold, it is determined whether or not there is an inflow vector in which the prediction vector indicating the prediction direction is directed to the decoding target block center direction. Means for extracting decoded blocks with vectors;
E Bei and means for generating a prediction value of a prediction mode of the block to be decoded by selecting one of the prediction mode of the decoded blocks there inflow vector is the extraction,
The means for generating the prediction value of the prediction mode is:
When there are a plurality of decoded blocks having the inflow vector, an edge of a reference pixel in the decoded block is detected, and a prediction mode of a decoded block having a prediction mode in a direction in which the edge appears most strongly is determined as a decoding target block An intra-screen predictive decoding device characterized in that the prediction value is a prediction value of the prediction mode . An intra-screen predictive encoding program for causing a computer to execute the intra-screen predictive encoding method according to claim 1 . A computer-readable recording medium on which the in-screen predictive encoding program according to claim 7 is recorded. An intra-screen predictive decoding program for causing a computer to execute the intra-screen predictive decoding method according to claim 3 or 4 . A computer-readable recording medium on which the in-screen predictive decoding program according to claim 9 is recorded.

Images