| H.264 is the newest international video coding standard proposed by the Joint Video Team of ITU-T VCEG and ISO/IEC MPEG. H.264 also uses hybrid video coding scheme based on motion compensation and transform. In order to avoid the matching error in IDCT caused by DCT in prior video standards, H.264 utilizes a new approximate DCT transform-integer transform. H.264 uses intra prediction to promote the intra encoding efficiency. It adopts variable block sizes, high exact fractional motion estimation and compensation and multi-reference frame selection to describe exactly the actual motion vector of objects. In order to eliminate the marginal effect, H.264 uses adaptive in-loop deblocking filter. H.264 uses CABAC to decrease the bits needed for encoding and so on. H.264 can decrease about 50% percent of bitrates under the same reconstructed picture quality comparing with prior standards by using above advance in video coding technology. But the high efficiency leads to high complexity of implementation directly and this will limit the application of H.264 in real time video coding. So it is a very important task to implement the H.264 coder with low hardware complexity and high encoding efficiency.In order to achieve high encoding efficiency, H.264 uses a great deal of encoding algorithms with high complexity. These complexity algorithms will limit the application of H.264 in real time video coding. So we must find corresponding fast algorithms to replace those with high complexity in H.264 reference software. In the mean time, in order to accelerate H.264 encoder, those modules used repeatedly in H.264 should be optimized in multimedia instruction. Moreover the structure of program and data for H.264 encoder can be modified according to the actual needs. According to above ideas, this dissertation gives deep research on some key algorithms and the implementation of H.264 encoder under the reference software JM7.6 in PC with high speed and efficiency.The main creative works are as follows.·Solved the exact detection of zero-block for H.264 theoretically. According to the characteristic of integer transform in H.264, this dissertation gives deep research on the zero-block detection for it, and deducts the formula of threshold for zero-block detection in H.264.·Proposed three motion search patterns and two fast ingeger pixel motion estimation. In order to decrease the complexity of motion estimation, this dissertation proposed that TP, LP or SSP can be used in coarse location for big motion vector. Based on these search patterns, two fast motion estimation algorithms were proposed-one is the fast motion estimation algorithm based on diamond and triangle search patterns DTS (Diamond Triangle Search) and the other is the fast motion estimation algorithm based on diamond and simplified square search patterns DSSS (Diamond Simplified Square Search).·Proposed a fast multi-frame selection algorithm. In order to decrease the complexity of multi-frame selection, a fast multi-frame selection algorithm based on motion vector and the RDO monotic of the multi-frame was proposed to decrease the sum of reference-frames decision.·Proposed a new variable length coding scheme-quotient-residual coding. This coding scheme can decrease the H.264 encoding length of video efficiently.The above mentioned algorithms were embedded in the H.264 reference software JM7.6 and the structure of program and data are also optimized properly according to actual needs. Some key modules such as computing the sum of absolute difference SAD in integer pixel motion estimation, computing the Hadamard transform of difference matrix and computing the sum of absolute difference SATD, integer transform and its inverse transform, sub-pel interpolation were optimized with multimedia instructions of PC, in the end the H.264 encoder was optimized wholly and it achieved satisfied effects. The experimental results show that the H.264 encoder with whole structure optimization, algorithm optimization and platform optimization can encode real time QCIF video series on general PC platform.
|
PDF Full Text Request