Research On The Motion Estimation And Transform Based On The H.264 Framework

With the development of technologies in storage, communication and processor, the demands for high quality video service increase. H.264 is the newest international video coding standard, which achieves higher rate distortion performance than previous video coding standards, but the complexity increases a lot. So how to reduce the encoder complexity and further improve its rate distortion performance are two research focuses. Supported by the Project of National High Technique Research and Development (863), some researches are done on fast motion estimation, mode selection, search range adjustment, integer transform and quantization.Based on the DC transform coefficients gotten in the modified successive elimination algorithm and the theoretical model for transform coefficients, an adaptive threshold is chosen to detect if all the transform coefficients will be quantized into zeros. It not only can speed the motion estimation, but also reduce the complexity of transform, inverse transform, quantization and inverse quantization. In order to increase the speed of motion estimation further, a fast fractional motion estimation based on the group error is presented too. Simulation results show that these algorithms can respectively reduce the encoding computation complexity up to 31% and 17% with very small rate distortion performance loss.In order to get the best macroblock mode in the variable block size algorithm, the encoder generally searches all modes, so the computation burden is very high. In addition, the maximal searching range in motion estimation is fixed. This would waste computation in searching many unnecessary candidate points. Bases on the spatial-temporal correlation of coding modes and best searching range, a fast mode selecting algorithm and an adaptive searching range algorithm are presented in this paper. So the computation of mode selection and searching points are saved. Simulation results show that these algorithms can respectively save the encoding computation and search points up to 83% and 96% with small rate distortion performance loss.The discrete cosine transform (DCT) was widely used in previous image and video coding standards. A disadvantage of DCT is that the elements in the transform matrix are irrational numbers, so the decoded data in the encoder maybe mismatch with that in the decoder. In order to solve this problem a new integer transform is presented in this paper. Its transform matrix is more approximate to DCT matrix than that in H.264, and can achieve better decorrelation and compaction performance. In addition, the quantization, inverse quantization and the quantization parameter are also analyzed.In conclusion, the fast motion estimation, mode selection, search range adjustment, integer transform and quantization are investigated in this paper. Experimental results reveal that the algorithms optimized for speed can reduce the computation complexity efficiently and keep approximate rate distortion performance, while the new transform and quantization method can improve rate distortion performance.