H.264 video encoder in DSP realization and optimization (1)

Summary: in DM642 EVM platform enables H.264 video encoder, and memory allocation, Cache optimization, code optimization, and assembler-level optimization, and other aspects are optimized for the encoder.

Experimental results show that the optimized encoders can maintain a higher image quality and compression efficiency and has better real-time performance.

1 Introduction

H.264/AVC is ITU-T video coding experts group and ISO/IEC moving picture experts group joint submission of a new generation of video coding standard.

Under the same conditions, and MPEG-1, MPEG-2, H.263, MPEG-4 standard H.264/AVC, compared to reduce 50% of the bit stream. However, the H.264 high coding efficiency is high computation and high complexity.

This article adopts high operation speed and strong data processing capabilities of DM642 EVM as H.264 video encoder implementation and optimization of DSP hardware platform, enables H.264 video encoding algorithm and algorithm program conducted comprehensive optimizer.

Experimental results show that optimized H.264 video encoder can maintain a higher image quality and compression efficiency and has better real-time performance.

2 H.264 video encoding technology and DM642 EVM development platform

2.1 H264 video encoding technology

H.264 compression algorithm used with the H.263 and MPEG-4 similar coding based on block mixing method using frames (Intra) and inter-frame (Inter) two encoding modes.

In order to improve the coding efficiency, compression and image quality, H.264 uses many new encoding technology, these technologies include:

⑴ H.264 standard compression system consists of video encoding layer (VCL, Video Coding Layer) and the network abstraction layer (NAL, Network Abstraction Layer) made up of two parts.

⑵ H.264 uses intra prediction, Yun to minimize redundant information in image space.

⑶ h.264 intra-frame prediction using multiple frames of reference frames (reference number from 1 to 5), high accuracy of interpolation algorithms (including 1/4 and 1/8 precision), a variety of new block of deformation, to a large extent, increase the motion estimation and compensation efficiency.

⑷ 1/4, 1/8-pixel accuracy of sub-pixel motion estimation: for QCIF video format uses 1/4 pixel accuracy prediction method for CIF video format uses a 1/8-pixel accuracy prediction methods.

⑸ residual image 4x4 integer DCT techniques, the inverse process does not have a matching error problem.

⑹ new loop filter technology and entropy coding, etc.

2.2 DM642 DM642 EVM development platform

EVM is TI launched a designed for multimedia application development platform, on-board resources including: DM642 CPU chip, 4M×64bit synchronous dynamic RAM (SDRAM), FLASH memory and all the way 4M×8bit video encoding and two channel video decoder, its structure as shown in Figure 1.

DM642 C64X kernel, based on the frequency up to 600MHz, adopting VLIW (VLIW) structure, each instruction cycle can parallel processing 8 32-bit processing capabilities of the directive, up to 4800MIPS; on-chip memory with L2 cache Cache structure, L1 data Cache by 16KB L1D and the procedure Cache L1P 16KB, 256KB L2 SRAM can be configured to Cache, or significantly improve the operational performance of the program; on-chip 64-bit EMIF (External MemoryInterface) interface to SDRAM and Flash, and storage devices to connect seamlessly and greatly facilitates the removal of large amounts of data; DM642 including three dedicated video port (VP0 ~ VP2) for receiving, processing, video data, improves the performance of the entire system; DM642 the EMAC mouth as well as from the EMIF extended out of the mouth of the ATA, also when processing is complete of massive data provides storage channel. Therefore, the realization of H.264 video algorithm, high-performance DM642 EVM is an ideal hardware platform.

3 H.264 video encoder implementation and optimization

3.1 encoder implementations

H.264 video encoder implementations there are a variety of ways, but mostly for transplantation, optimized operations.

H.264 code to DSP software platform runs under CCS environment, you need to be aware of several issues, such as configuration files, library files, change, data types of adjustment, assembler of processing, memory, end mode adjustment.

H.264 encoding using transform and prediction of mixed encoding methods, its principle is shown in Figure 2.

Enter the frame or field Fn to macro blocks are processed, the image encoder into sub image block to sub image block as an encoding element. When using intra prediction coding, P is the predicted value from the current movie encoded reference image after motion compensation (MC), which refer to images with F1n-1 said; in order to improve forecast accuracy, thereby increasing the compression ratio, the actual reference image may have been in the past or future encoding and decoding reconstruction and filter frame. Forecast values P and subtract the current block, produce a residual block Dn, block transform, quantized results in a set of quantized coefficient of variation of X, then after entropy coding, and decode the required some side information (such as prediction model quantization parameter, Motion vectors, etc.) together make up a compressed stream, after NAL for transmission and storage.

[1] [2] [3]