Video Coding Motion Estimation Algorithm

With the rapid development of multimedia techniques, many video applications such as Video on Demand, Video Conference, Mobile Video need the video compression system to be a real-time one. The key point to solve this problem is the fast and efficient compression algorithm. As a vital technique of video compression, motion estimation(ME) is also the most complex and time-consuming module. Some researches show that motion estimation takes 70% of the total computation for one reference frame case, and yet for multi-reference frames the proportion can rise up to 90%. Therefore fast and effective motion estimation algorithms are the most common search area for industrial and academic researchers. In a variety of motion estimation methods, due to the excellence of simpleness and hardware implementation of block matching algorithm(BMA) the relative fast algorithms have extensive researches and applications. The thesis focuses on researches about fast block matching motion estimation algorithms based on single and multi-reference frame. The main achievements are as follows.For one reference frame case, according to the distributional characteristics of motion vectors, a new fast block matching algorithm named ACSS is proposed which is based on the detailed analysis of various classical motion estimation algorithm. The new algorithm exploits a series of techniques such as multiple reference motion vector prediction, adaptive termination criteria and multiple search templates. According to test on H.264 verification model JM 10.2, with the similar Peak Signal to Noise Ratio(PSNR) and bit rate the new algorithm has the minimum computational complexity. That demonstrates the new algorithm is most effective compared with others.With regard to the characteristics that H.264 can support for multi-reference frame and multi-model motion estimation, a fast multi-reference frame motion estimation algorithm for H.264 is proposed according to the statistical characteristics of motion vector and different effects for coding efficiency enhancement of various reference frames and block patterns. The algorithm is simple and easy to implement and meanwhile it takes fully into account the spatio-temporal correlation of image sequences and combines motion vector reuse with a variety of termination conditions. Experiments show that at low bit-rate of image sequence transmission, thus multi-reference frame search algorithm has some advantages in matching accuracy and speed of operation.