KR102050238B1 - System and method for effectively deciding intra prediction mode for hevc encoder - Google Patents

Abstract

The present invention relates to an efficient intra prediction prediction Angular mode determination system and method, wherein, in the intra prediction of video encoding, the pixels of each line in the horizontal and vertical directions of the original image pixel data are different. The amount of computation and the computation time can be reduced by predicting the directionality using the position where the difference is greatest and determining the optimal intra prediction Angular mode for each of the horizontal and vertical directions.

Description

Efficient Intra Prediction Mode Determination System and Method for HEVC Coder

The present invention relates to an intra prediction mode determination system and method for an HEVC encoder. More particularly, the present invention compares the difference value of an original image pixel to reduce the computation time and the high computation amount of the existing intra prediction mode determination of HEVC. An efficient intra prediction mode determination system and method for an HEVC encoder for efficiently determining intra prediction angular modes.

Recently, due to the development of various video devices that support ultra high definition (UHD) -class ultra high definition images, users' interest and demand for high resolution images have increased. For this reason, it is necessary to develop a new video compression technology standard to support high resolution video such as UHD video. In line with this trend, HEVC (High Efficiency Video Coding) was jointly developed by the Moving Picture Experts Group (MPEC) from ISO / IEC and the Video Coding Experts Group (VCEG) from ITU-T in January 2010. (Joint Collaborative Team on Video Coding) is a new generation of video compression standard. HEVC is able to efficiently manage ultra-high-definition video by showing coding efficiency (i.e., compression rate) more than twice that of the same picture quality compared to H.264 / AVC, which is the existing video compression standard.

In-picture prediction in HEVC is a method of predicting the current block with reference to samples already reconstructed around the current block. The intra prediction sequence is performed in the order of reference sample preparation, intra prediction image generation, and prediction mode encoding as shown in FIG. 1. The reference sample preparation step includes reference sample padding and reference sample filtering. First, in intra prediction, prediction is performed using surrounding reconstructed samples. Such a sample is called a reference sample. Depending on the mode, padding and filtering of reference samples are performed, and intra prediction is performed. Reference sample padding is a step of preparing a reference sample by performing padding on a case where no reference samples exist, and there are strong filtering and weak filtering. After that, transform and quantization are performed on the difference signal between the original image and the predicted image, and the optimal mode is selected by measuring the block reconstruction and the rate-distortion cost.

The intra prediction mode decision technique in HEVC is a good way to compress the video effectively, but it takes a very high time complexity because it takes up a small portion of the total coding time. In addition, depending on the conditions, the complexity can be increased up to 3 times compared to the existing H.264 / AVC. When comparing the intra prediction mode decision technique in HEVC with the conventional intra prediction mode determination technique in H.264 / AVC, the biggest difference is the size of the prediction unit (PU) and the number of prediction directions. . Conventional H.264 / AVC supports a total of nine prediction modes in 4 × 4 to 16 × 16 intra picture coded prediction blocks, while in HEVC, 4 × 4 to 64 × 64 picture intra coded predictions By supporting a total of 35 prediction modes in the block, the size of the PU is larger than before, the units of the PU are diverse, and more various directions can be calculated. In particular, the intra prediction mode in HEVC is composed of 1 DC mode, 1 planar mode, and 33 Angular modes, and the prediction is most similar to the current frame by determining the optimal mode among 35 modes. Encoding performance is improved by generating frames. Herein, mode 0 represents an Intra_Planar mode using a reference pixel value and position, mode 1 represents an Intra_DC mode using an average value of the reference pixel, and modes 2 through 34 represent an Intra_Angular mode using directionality of the reference pixel. As can be seen in FIG. 2, the intra prediction mode of HEVC supports various directions from the upper right direction to the lower left direction.

As such, since HEVC supports more prediction modes than H.264 / AVC, flexible and more accurate intra prediction is possible. However, because HEVC has to calculate the rate-distortion cost for all prediction modes in consideration of all 35 modes in PUs ranging from 4x4 to 64x64 in order to determine the optimal intra-picture prediction mode, And computation time are required.

Therefore, in the present invention, a system using an algorithm for efficiently determining an optimal intra prediction Angular mode by comparing the difference value of the original image pixels in order to reduce the computation time and the high computation amount of the conventional intra prediction mode determination method in HEVC And a method.

Republic of Korea Patent Publication No. 10-2014-0056599 (2014.05.12) Republic of Korea Patent Publication No. 10-1621358 (2016.05.17)

Accordingly, the present invention has been made to solve the above-mentioned problems of the prior art, and an object of the present invention is to select a directionality using the difference and position of the original image pixel data, thereby selecting a horizontal direction and a vertical direction among 33 Angular modes. The present invention provides an efficient intra-picture prediction Angular mode determination system and method for an HEVC encoder by applying an efficient intra-picture prediction mode selection algorithm that efficiently selects one mode for each vertical direction.

In the intra prediction Angular mode determination system of the present invention for achieving the above object, in the intra prediction of the video encoding, each line in the horizontal and vertical direction of the original image pixel data The optimal intra-picture prediction Angular mode is determined for each of the horizontal and vertical directions by predicting the directionality using the position where the difference between the pixels is the largest.

Preferably, the intra prediction Angular mode determination system of the present invention comprises a memory for storing the original image pixel data; And an Angular mode selection module for determining an optimal intra-picture prediction Angular mode of the original image pixel data, wherein the Angular mode selection module comprises: an input unit for receiving the original image pixel data from the memory and storing the Angular mode; A subtraction operation is performed on the previous image pixel data stored in the register and the original image pixel data newly input to the input unit, and the difference value and the position between the result pixels of the operation are stored, and the new original image pixel data is continuously transferred to the input unit. A comparison unit for comparing the stored difference value with the newly obtained difference value and updating the larger value and its position, and calculating a difference between the pointers of the positions having the largest difference value between pixels in the adjacent lines. To calculate the difference between the pointers for each value. And a mode selection unit for comparing the maximum counting number with a predetermined threshold value and selecting an optimal intra-picture prediction Angular mode for a specific direction in the prediction unit according to the comparison result.

More preferably, the memory is configured separately for each of the transverse and longitudinal directions.

Even more preferably, the intra prediction prediction Angular mode determination system of the present invention performs the intra prediction prediction Angular mode determination for each prediction unit of a predetermined size in parallel.

On the other hand, the intra prediction Angular mode determination method of the present invention, in the intra prediction of the video encoding, by using the position where the difference between the pixels of each line in the horizontal and vertical direction of the original image pixel data is the largest, Prediction determines the best intra prediction Angular mode for each of the horizontal and vertical directions.

Preferably, the intra prediction prediction Angular mode determination method of the present invention comprises: reading original image pixel data from a memory; Performing a subtraction operation on the previous original image pixel data and the newly read original image pixel data, and storing the difference value and the position between the result pixel of the operation; Subsequently, when new original image pixel data is input, comparing the stored difference value with the newly obtained difference value and updating a larger value and its position; Performing an operation for calculating a difference between pointers of positions having the largest difference value between pixels in respective adjacent lines, and counting the difference value between the pointers for each value as a result of the operation; And comparing the maximum counting number with a predetermined threshold value and selecting an optimal in-picture prediction Angular mode for a specific direction in the prediction unit according to the comparison result.

As described above, the efficient intra prediction prediction Angular mode determination system and method according to the present invention derives the position with the most difference through the pixel difference in the horizontal direction and the vertical direction, and then predicts the directionality to efficiently predict the Angular mode. By reducing the amount of computation and computation time.

1 is a flowchart schematically illustrating a general process of intra picture prediction of HEVC.
2 is a schematic diagram illustrating various directions supported by the Angular mode, which is one of the intra prediction modes of HEVC.
3 is a block diagram schematically showing the overall configuration of an intra prediction Angular mode determination system according to a preferred embodiment of the present invention.
4 is a schematic diagram illustrating a process of calculating original image pixel data for determining an intra prediction angular mode according to a preferred embodiment of the present invention.
FIG. 5 is a schematic diagram illustrating an example of a calculation process for determining an intra-picture prediction angular mode in a vertical direction for a prediction unit having a 5 × 5 block size according to an embodiment of the present invention.
6 is a schematic diagram showing parallel processing of original image pixel data for determining an intra prediction angular mode according to a preferred embodiment of the present invention.
7 is a flowchart schematically illustrating an entire process of an intra prediction angular mode determination method according to an exemplary embodiment of the present invention.

Advantages and features of the present invention and methods for achieving them will be apparent with reference to the embodiments described below in detail with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but may be embodied in various different forms, and only the embodiments are provided to make the disclosure of the present invention complete, and those of ordinary skill in the art to which the present invention belongs. It is provided to fully convey the scope of the invention to those skilled in the art, and the present invention is defined only by the scope of the claims. Like reference numerals refer to like elements throughout.

Hereinafter, an efficient intra prediction prediction Angular mode determination system of the present invention will be described in detail with reference to the accompanying drawings.

3 is a block diagram schematically showing the overall configuration of an intra prediction Angular mode determination system according to a preferred embodiment of the present invention.

As shown in FIG. 3, an intra prediction prediction Angular mode determination system according to a preferred embodiment of the present invention includes a memory Memory_ctrl 100 and an Angular mode selection module Intra_top 200.

The memory 100 stores original image pixel data in a horizontal direction and a vertical direction. At this time, the memory 100 is preferably configured separately from the memory (W_mem_ctrl, 110) for the horizontal direction and the memory (H_mem_ctrl, 120) for the vertical direction, which is to use the memory for both as one memory This is because the area is increased to maintain the original image pixel data when the mode decision on the vertical direction is performed. The original image pixel values in each memory 110 and 120 according to the preferred embodiment of the present invention are used as an input of the Angular mode selection module 200 described later.

Angular mode selection module 200 is an Angular mode selection module (W_ctrl, 210) for the horizontal direction and Angular mode selection module for the vertical direction to determine the best in-picture prediction Angular mode for each landscape and portrait direction. (H_ctrl, 220), each Angular mode selection module (210 and 220) is first to each of the horizontal memory 110 and vertical memory in which the original image pixel data for the horizontal and vertical directions are stored, respectively. After receiving the original image pixel data from 120, the received original image pixel data is applied to each of the received original image pixel data by applying the intra prediction Angular mode determination algorithm disclosed in the present invention, which will be described in detail with reference to FIGS. 4 and 5 below. It determines the best in-picture prediction Angular mode for the horizontal and vertical directions.

4 is a schematic diagram illustrating a process of calculating original image pixel data for determining an intra prediction angular mode according to a preferred embodiment of the present invention.

Referring to FIG. 4, first, the input unit 201 of the Angular mode selection module 200 receives original image pixel data Origin_pixel from the memory 100 for each direction, and converts the input original image pixel data into a pixel register. 202 can be stored. Subsequently, the comparison unit 203 performs a subtraction operation on the previous original image pixel data stored in the pixel register 202 and the newly input original adjacent image pixel data, and calculates the difference value and the position between adjacent pixels. The register 204 and the position register 205 are stored. In this case, it is preferable to use only one register using a comparator to use the minimized register. Thereafter, when the new original image pixel data is input to the input unit 201, the comparison unit 203 compares the difference value between the pixels stored in the value register 204 with the newly obtained pixel difference value and determines a larger value and its position. The values are stored in the value register 204 and the position register 205, respectively, and are updated. Accordingly, the operation of all pixel data values for one line in a specific direction in one prediction unit (PU) is completed. The value register 204 and the position register 205 store the difference value between the largest pixel in the line and its position.

Then the mode selector (not shown) of the Angular mode selection module 200 again points to a pointer at the location determined to have the largest pixel-to-pixel difference value (ie, all pixel data values for one line). After the operation is finished, the pointer of the position determined to have the difference value between the largest pixel in the line adjacent to the position stored in the position register 205 (that is, the operation is finished for all pixel data values for the line adjacent to it). Subsequently, a subtraction operation is performed on the pointer stored in the position register 205, and counting cases in which the difference value between the pointers of adjacent lines correspond to predetermined values are counted.

Finally, the mode selector of the Angular mode selection module 200 compares the maximum counting number among the counted numbers with a predetermined threshold value for each prediction mode determined by the experiment, and according to the result, each horizontal direction in one prediction unit And determine the intra prediction Angular mode for the vertical direction.

FIG. 5 is a schematic diagram illustrating an example of a calculation process for determining an intra-picture prediction angular mode in a vertical direction for a prediction unit having a 5 × 5 block size according to an embodiment of the present invention. This calculation process is described only in the vertical direction in the present application, but the same will be the same in the horizontal direction.

Referring to FIG. 5, assuming that lines in a vertical direction are divided in a PU having a 5 × 5 block size, the Angular mode selection module 200 of the present invention first divides the angles divided by (1) to (5) in FIG. 5. An operation for calculating the difference between adjacent pixels with respect to the vertical lines is performed to find a position having the largest difference value between pixels. As an example in FIG. 5, the position having the largest difference value between pixels in each vertical line is indicated by an arrow.

Then, the Angular mode selection module 200 again calculates a difference between pointers of positions having the largest pixel difference value in each adjacent vertical lines, so that the difference value between the pointers is '-2', The cases corresponding to '-1', '0,' '1', and '2' are counted for each value. For example, in FIG. 5, the pointer of the position having the largest pixel-to-pixel difference value in the vertical line 1 is 4, and the pointer of the position having the largest pixel-to-pixel difference value in the vertical line 2 adjacent thereto is 3. In this case, since the difference between the pointers is 1, the counting value is '1' and the counting number becomes 1. If the pointer of the position having the largest pixel-to-pixel difference value in the vertical line 2 is 3 and the pointer of the position having the largest pixel-to-pixel difference value in the vertical line 3 adjacent thereto is 3, the difference between the pointers is Since it counts as '0', the counting number is 1, and so on. In the example of FIG. 5, when the operation for calculating the difference between pointers for each vertical line is completed, the counting number for '0' is 1 and the counting number for the value of '1' will be 3.

Finally, the Angular mode selection module 200 compares the maximum counting number (counting number 3 for the value of '1' in this example) with the predetermined threshold value for each prediction mode determined by the experiment. In this case, the intra prediction Angular mode for the vertical direction in the corresponding PU having a 5 × 5 block size is determined.

As described above, the present invention can significantly reduce the amount of computation and the computation time compared to the prior art by determining the intra prediction angular mode for each of the horizontal and vertical directions through a process as shown in FIG. 5.

6 is a schematic diagram showing parallel processing of original image pixel data for determining an intra prediction angular mode according to a preferred embodiment of the present invention.

The intra prediction Angular mode determination system of the present invention supports all blocks ranging from 4 × 4 size blocks to 64 × 64 size blocks and is divided by the operation of the original image pixel data described with reference to FIGS. 4 and 5. Angular mode decisions for each of them can be processed in parallel.

Referring to FIG. 6, when the original image pixel data is processed up to a 64 × 4 sized prediction block indicated in red, an angular mode of a 4 × 4 block of one line is output. When image pixel data processing is performed up to a 64 × 8 prediction block, the angular modes of 4 × 4 and 8 × 8 blocks are determined in parallel. In this way, when the last original image pixel data is input and the angular mode is determined, the angular mode is determined in parallel from the 4 × 4 prediction block to the 64 × 64 prediction block.

Block size Number of Angular Modes in a Line sum 4 × 4 16 256 8 × 8 8 64 16 × 16 4 16 32 × 32 2 4 64 × 64 One One

Table 1 shows the number of Angular modes according to the block size, and the number of output modes in the horizontal and vertical directions are the same. Taking a 4x4 block as an example, the number of outputs of Angular modes for one line in a 64x4 block is 16. Processing all 64x64 blocks results in 256 total modes for a 4x4 block size. The number of modes for the remaining block sizes is shown in the table above.

Next, an efficient in-picture prediction Angular mode determination method of the present invention using the system configured as described above will be described. This will be described in detail as follows.

7 is a flowchart schematically illustrating an entire process of an intra prediction angular mode determination method according to an exemplary embodiment of the present invention.

Referring to FIG. 7, first, the input unit 201 of the Angular mode selection module 200 reads original image pixel data Origin_pixel from memory 110 and 120 for each direction (S710). In this case, the input unit 201 may store the read original image pixel data in the pixel register 202.

Then, the comparing unit 203 of the Angular mode selection module 200 performs a subtraction operation on the previous original image pixel data stored in the pixel register 202 and the original input image pixel data newly input thereto, and the resultant operation results in a neighboring pixel. The difference value and its position are stored in the value register 204 and the position register 205 (S720).

After that, when new original image pixel data is continuously input to the input unit 201 and steps S710 and S720 are performed, the comparison unit 203 compares the difference value between the pixels stored in the value register 204 and the newly obtained pixel difference value. Compare and store the larger value and its position in the value register 204 and the position register 205 and update them (S730). Accordingly, if steps S710 to S730 are performed for all pixel data values for one line in a specific direction in one PU, the value register 204 and the position register 205 are between the largest pixel in the corresponding line. The largest difference value and its location will be stored.

Then, the mode selection unit (not shown) of the Angular mode selection module 200 performs steps S710 to S730 on all pixel data values for one line, so that the pointer at the determined position and all the pixel data for the line adjacent thereto are determined. Steps S710 to S730 are performed on the value to perform a subtraction operation on the pointer at the determined position, and counting cases in which the difference value between the pointers adjacent to each other line corresponds to predetermined values as a result of the operation (S740).

In addition, the mode selection unit of the Angular mode selection module 200 compares the maximum counting number among the counted numbers with a predetermined threshold value for each prediction mode determined by the experiment (S750), and then in one prediction unit according to the result. Intra-prediction Angular mode for each of the horizontal and vertical directions is determined (S760).

The existing intra prediction mode determination algorithm determines the mode by considering all 35 intra prediction modes, whereas the intra prediction Angular mode determination system proposed in the present application predicts the direction through simple calculation as described above. Deciding the mode allows faster mode decision. In addition, the memory area is reduced by storing the original image pixels in each direction by making the memory for the horizontal and vertical directions of the original image separately for efficient memory management. In addition, since the Angular mode determination is performed in parallel from the 4 × 4 block size to the 64 × 64 prediction block, prediction mode determination for all block sizes is performed when the last original image pixel data is processed.

Experimental Example

Table 2 shows the performance comparison of the efficient intra prediction mode selection algorithm of the present invention applied to the HM-16.9 standard software and the proposed hardware. On average, BDPSNR increased by 0.035, BDBitrate decreased by 0.623, and encoding time by 11.389%.

In this experimental example, an efficient in-picture prediction Angular mode decision algorithm with good results compared to standard software was applied to the hardware design. The algorithm that minimizes the angular mode by predicting the directionality through simple calculation rather than the existing intra prediction algorithm is applied to the hardware. As a result, the hardware area is minimized through the use of a minimized calculator. The hardware structure proposed in this experiment was designed with Verilog HDL and used 65nm process. Synopsys synthesized by Design Compiler, Table 3 shows the synthesis results and comparison of the hardware structure proposed in the experimental example. The synthesis results in a gate count of 14.9K and a maximum operating frequency of 2GHz. The hardware structure of this experiment is compared with the existing software which has the best results in terms of the number of gates and the maximum operating frequency among the standard software. In comparison, the gate count is reduced by 75% and the maximum operating frequency is increased from 622MHz to 2GHz.

Class resolution Experimental Example BDPSNR BDBitrate △ Bitrate (%) △ PSNR (%) △ TS (%) 4k 3840x2160 0.016 -0.613 -0.663 0.007 7.181 Class A 2560x1600 0.031 -0.518 -0.366 0.023 6.714 Class B 1920x1080 0.027 -0.642 -0.543 0.013 6.112 Class C 832 x 480 0.039 -0.620 -0.469 0.023 6.043 Class D 416 x 240 0.042 -0.672 -0.678 0.012 11.389 Class E 1280 x 720 0.046 -0.848 -0.726 0.019 5.829 Class F 832 x 480 0.047 -0.449 -0.391 0.016 5.064 Average - 0.035 -0.623 -0.548 0.016 6.905

Experimental Example Comparative Example 1 Comparative Example 2 Technology (nm) 65 65 90 Number of gates
(Nand gate) 14.9K 59.5K 214.1K Support blocks All blocks All blocks 64x64 no Cycles / 64x64 4,100 2,539 15,908 Max Frequency (MHz) 2,000 622 357 Throughput 8K @ 60fps 8K @ 30fps 1K @ 44fps

Although the present invention has been described in more detail with reference to some embodiments, the present invention is not necessarily limited to these embodiments, and various modifications can be made without departing from the spirit of the present invention.

100: memory
200: Angular mode selection module
201: input unit
202: pixel register
203: comparison unit
204: value register
205: location register

Claims (6)

In the intra prediction of the video encoding, the directionality is predicted by using the position where the difference between the pixels of each line in the horizontal direction and the vertical direction of the original image pixel data is the largest, and the respective horizontal and An in-screen prediction Angular mode determination system that determines an optimal in-picture prediction Angular mode for portrait orientation. In claim 1,
A memory storing the original image pixel data; And
And an Angular mode selection module for determining an optimal intra-picture prediction Angular mode of the original image pixel data.
The Angular mode selection module is:
An input unit which receives the original image pixel data from the memory and stores the original image pixel data;
A subtraction operation is performed on the previous original image pixel data stored in the pixel register and the original image pixel data newly input to the input unit, the difference value and the position between the resultant pixels are stored, and the new original image is subsequently added to the input unit. A comparison unit for comparing the stored difference value with the newly obtained difference value and updating a larger value and its position when pixel data is input, and
The data difference between the pointers of the positions having the largest difference value between pixels in each adjacent line is calculated to count the data difference value between the pointers for each value, and the maximum counting number and the predetermined threshold value are calculated. And a mode selector to determine an optimal intra-picture prediction angular mode with respect to the horizontal or vertical direction in the prediction unit according to the comparison result. In claim 2,
And the memory is configured separately for each of the landscape and portrait orientations. The method according to any one of claims 1 to 3,
An in-picture prediction Angular mode determination system for performing in-picture prediction Angular mode determination in parallel for respective prediction units of a predetermined size. In the intra prediction of the video encoding, the directionality is estimated using the position where the difference between the pixels of each line in the horizontal direction and the vertical direction of the original image pixel data is the largest, and is optimal for the respective horizontal and vertical directions. How to determine intra prediction Angular mode. In claim 5,
Reading original image pixel data from memory;
Performing a subtraction operation on the previous original image pixel data and the newly read original image pixel data, and storing the difference value and the position between the result pixel of the operation;
Subsequently, when new original image pixel data is input, comparing the stored difference value with the newly obtained difference value and updating a larger value and its position;
Performing an operation for calculating a data difference between pointers at positions having the largest pixel difference value in each adjacent line, and counting the data difference value between the pointers for each value as a result of the operation;
And comparing the maximum counting number with a predetermined threshold value and determining an optimal in-picture prediction Angular mode for the horizontal direction or the vertical direction in the prediction unit according to the comparison result.

Images