CN100592799C

CN100592799C - Rapid reading method of motion compensating data based on H.264 standard

Info

Publication number: CN100592799C
Application number: CN 200810302116
Authority: CN
Inventors: 鲁国宁; 袁梓瑾; 莫启会; 毛夏飞
Original assignee: Sichuan Hongwei Technology Co Ltd
Current assignee: Sichuan Hongwei Technology Co Ltd
Priority date: 2008-06-12
Filing date: 2008-06-12
Publication date: 2010-02-24
Anticipated expiration: 2028-06-12
Also published as: CN101287126A

Abstract

The invention relates to the technology of video coding processing and provides a fast data reading method based on H.264 standard motion compensation. The technical proposal of the method is that: the fast data reading method based on the H.264 standard motion compensation comprises the steps as follows: a. a sub-block segmentation type and a corresponding reference frame data matrix comparison list are generated when a system is started; b. the code of the current macroblock type is analyzed and the type of the segmentation of the current sub-block is obtained; c. the reference frame data matrix scale that needs to be read is obtained after the list is checked, and a DMA controller is informed; d. the DMA controller transmits the reference frame data into the in-chip cache from an out-of-chip memory cell; e. a motion compensation unit carries out coding motion compensation calculation by visiting the reference frame data in the in-chip cache. The method of the invention has the advantages of greatly improving data access efficiency and reducing bus pressure and is suitable for a video decoding system that adopts H.264 standard.

Description

Data rapidly-reading method based on the compensation of standard movement H.264

Technical field

The present invention relates to a kind of video coding treatment technology, particularly the video decode treatment technology of standard H.264.

Background technology

Along with Digital Television, network video stream and DVD broad application, video compression has become the element of broadcasting and entertainment medium.The success of Digital Television and DVD is that it is very practical that this standard once was proved to be on the Moving Picture Experts Group-2 that was based upon before more than ten years, but it seems now, and is somewhat out-of-date.With more efficient, more practical, and can well utilize trend that the technology of stronger disposal ability replaces MPEG-2 clearly.Relatively good recently is international standard MPEG-4 and H.264.H.264 standard is devoted to the narrower tradition of with respect to MPEG-4 standard scope and is used, focus on by the efficient compression of the captured rectangle video image of common camera and can be on network the video flowing of reliable transmission.H.264 standard technique is to be that cost improves its compression efficiency with the implementation complexity that increases it.As everyone knows, numerous main flow video compression technologies all adopt motion compensation technique, and motion compensating system can be removed the redundant information on the time dimension between frame of video very efficiently.H.264 equally also used motion compensation technique.In order to improve the compression performance of motion compensation, the smallest match piece of motion compensation is brightness 4 * 4 sub-block of pixels, colourity 2 * 2 sub-block of pixels, and the maximal accuracy at the motion vector of reference frame matched position is brightness 1/4 pixel, colourity 1/8 pixel, then decoder is when one 4 * 4 sub-block of pixels of every decoding is cut apart type, all to carry out interpolation arithmetic to the sub-piece in the corresponding reference frame, and reference frame data is because excessive, generally be placed on chip external memory, according to conventional algorithm, 4 * 4 sub-block of pixels of every pair of reference frame are carried out interpolation, just need read the data of the sub-block of pixels of corresponding 9 * 9 sizes in this reference frame, this is because are the interpolation filters that adopted six rank at 1/2nd interpolation arithmetics H.264, when therefore 4 * 4 sub-block of pixels being carried out 1/2nd interpolation, need read in the reference frame data of the sub-block of pixels of 9 * 9 sizes, therefore cutting apart when the sub-piece that needs interpolation is 4 * 8 types, interpolation algorithm according to routine, will read in two 9 * 9 sub-block of pixels at twice, this will have lot of data and be repeated to read in, this is unnecessary consumption, can calculate so, when the macro block of one 16 * 16 size is carried out interpolation, need to divide and read in 16 9 * 9 sub-block of pixels 16 times, the reference frame data scale that then needs to read in will be (during the yuv420 form): 9*9*16*8*1.5=15552bit.Therefore if will accomplish the real-time decoding of high-definition digital video, will bring huge pressure to the outer transfer of data of sheet to the data transmission bus in the sheet to being responsible for.

Summary of the invention

Technical problem to be solved by this invention, a large amount of reference frame datas repeats to read in during exactly for fear of the motion compensated interpolation computing, and a kind of data rapidly-reading method based on the compensation of standard movement H.264 is provided.

The present invention solves its technical problem, and the technical scheme of employing is, based on the data rapidly-reading method of standard movement compensation H.264, it is characterized in that, may further comprise the steps:

A. generate sub-piece during system start-up and cut apart type and should the corresponding reference frame data matrix size table of comparisons that reads, and deposit in the system;

B. go out the code of current macro type by the code stream analyzing unit resolves of standard H.264,, obtain the type that current sub-block is cut apart according to the grammer of standard code H.264;

C. cut apart according to current sub-block and obtain the reference frame data matrix size that read behind the type check table, and inform dma controller;

The d.DMA controller is transferred to the reference frame data of corresponding scale in the sheet in the buffer memory by sheet external memory unit;

E. motion compensation units is by visiting the motion compensation computing that the reference frame data in the buffer memory is decoded in the sheet.

Further, the described table of comparisons of step c such as following table:

The sub-piece of table 1 is cut apart the type and the reference frame data matrix table of comparisons

Current sub-block is cut apart type	The reference frame data matrix
Current sub-block is cut apart type	The reference frame data matrix	4×4	9×9
4×8	9×13	4×4	9×9
4×8	9×13	8×4	13×9
8×8	13×13	8×4	13×9
8×8	13×13	8×16	13×21
16×8	21×13	8×16	13×21
16×8	21×13	16×16	21×21

The invention has the beneficial effects as follows, by from the code stream analyzing unit, obtaining the categorical data that current sub-block is cut apart, and establishment look-up table, obtain the reference frame data scale that to read in then by tabling look-up, rather than as conventional way, no matter current sub-block is cut apart is for which kind of pattern, reads in reference data with 9 * 9 scale without exception.The present invention can make that the process that reads of reference frame data scale becomes more flexibly and have a preliminary self-adaptive processing ability, though increased the work of a part of computing of tabling look-up, but saved the pressure of data/address bus greatly, because in whole system, the right to use of data/address bus is a scarce resource, so the present invention can improve decoding efficiency greatly.

Description of drawings

Fig. 1 is the system flow chart of embodiment;

Fig. 2 is cut apart the schematic diagram that type need be read in pixel for embodiment 4 * 8 sub-pieces;

Wherein 1 for to need 4 * 8 sub-pieces of interpolation to cut apart pixel, and 2 are the pixel that need additionally read in, 3 pixels that need repeat to read in for conventional algorithm.

Embodiment

Below in conjunction with drawings and Examples, describe technical scheme of the present invention in detail.

The invention provides a kind of establishment and search the table of comparisons, from the code stream analyzing unit, obtain the categorical data that current sub-block is cut apart again, obtain the reference frame data scale that to read in then by tabling look-up, carry out the motion compensation computing at last again, solved the problem that reference frame data repeats to read in, and can improve data access efficiency greatly, reduce total linear pressure.

Embodiment

This example is divided into example with 4 * 8 sub-pieces, sets forth technical scheme of the present invention, its system flow such as Fig. 1, and 4 * 8 sub-pieces are cut apart schematic diagram such as Fig. 2 that type need be read in pixel.

At first when system start-up, generating sub-piece cuts apart type and answers the corresponding reference frame data matrix table of comparisons that reads, as table 1, and be stored in the system, so that search use later on, when receiving to carry out the macro block of motion compensation computing the time, cut apart type by the code stream analyzing element analysis current sub-block of standard H.264, if it is 4 * 8 forms that the current sub-piece that needs interpolation is cut apart type, 1 can get through tabling look-up, the reference frame data scale that need read in is 9 * 13 matrix, therefore the dma controller pixel number that will disposablely read in the integer pixel positions of 9 * 13 matrixes is sent in the sheet in the buffer memory according to (comprising the pixel 2 that 4 * 8 sub-pieces that need interpolation are cut apart pixel 1 and need additionally be read in) and with it, rather than as conventional way, the pixel number of the integer pixel positions of 9 * 13 matrixes 9 * 9 is read in twice for unit according to this, to reduce like this needs 45 pixel numbers repeating to read in according to 3 work in the conventional way, carry out the motion compensation computing by the reference frame data in the buffer memory in the motion compensation arithmetic element visit sheet more at last.

Claims

1. based on the data rapidly-reading method of the compensation of standard movement H.264, it is characterized in that, may further comprise the steps:

2. according to the described data rapidly-reading method of claim 1, it is characterized in that the described table of comparisons of step c is described below based on the compensation of standard movement H.264:

It is 4 * 4 o'clock that sub-piece is cut apart type, and the reference frame data matrix size is 9 * 9;

It is 4 * 8 o'clock that sub-piece is cut apart type, and the reference frame data matrix size is 9 * 13;

It is 8 * 4 o'clock that sub-piece is cut apart type, and the reference frame data matrix size is 13 * 9;

It is 8 * 8 o'clock that sub-piece is cut apart type, and the reference frame data matrix size is 13 * 13;

It is 8 * 16 o'clock that sub-piece is cut apart type, and the reference frame data matrix size is 13 * 21;

It is 16 * 8 o'clock that sub-piece is cut apart type, and the reference frame data matrix size is 21 * 13;

It is 16 * 16 o'clock that sub-piece is cut apart type, and the reference frame data matrix size is 21 * 21.