More detailed content of this thesis can be described as follows: Firstly，the sub-pixel motion estimation techniques are deeply studied. The sub-pixel motion estimation can be divided into two steps: sub-pixel interpolation and sub-pixel matching. In sub-pixel interpolation, the applications of multi-tap filter and the effect of the taps of the filter to the precision and complexity of the sub-pixel interpolation algorithm are studied. In sub-pixel matching, a fast quarter-pixel motion estimation algorithm is proposed, based on the statistical analysis of the distribution of quarter-pixel precision motion vectors. The algorithm reduces the complexity of quarter-pixel motion estimation greatly. Secondly, the multi-block mode motion estimation algorithm of very low bit-rate video sequences is studied. The inter-frame motion of the very low bit-rate video sequences is much less compared with that of the high bit-rate video sequences, thus the bit-rate is much lower. Through analyzing the characteristic of the very low bit-rate video sequences and the correlation of the multi-block modes, a fast quarter-pixel motion estimation algorithm is proposed, which offers the computational complexity and performance trade-offs in a very low bit-rate video coding scheme. Finally, the multi-block mode motion estimation is studied. Based on the statistical characteristic of the mode distribution, two multi-block mode motion estimation algorithms are put forward, in which we define a threshold to judge whether it is necessary to continue the motion estimation for the smaller block modes. Experimental results show that the proposed techniques provide competitive performance compared with full search (FS), that is, highly reduced computational complexity, very similar image quality and bit-rate.
修改评论