Study On Motion Estimation Algorithm Based On MPEG-4

With the development and popularization of computer and network technology, video technology has been widely applied to various applications and plays a more and more important role nowadays. MPEG-4 is a popular video compression coding standard, which is applied to plenty of relative areas such as internet multimedia, real-time visual communication and etc. because of its powerful performance and high compression efficiency.Among the factors which influence the coding quality and efficiency for MPEG-4, motion estimation technology is the most important one. It is known that the accuracy of motion estimation usually determines the quality of decoded images to a great extent. However, motion estimation contributes to the most part of computational load in the whole video coding system. Thus research on efficient motion estimation algorithm becomes a focus in video coding field.At first, this paper gives an overview of the specialties of MPEG-4,then introduce the basic theory and key points of its core technology, which is block-matching motion estimation. On the basis of the analysis on popular fast block-matching algorithms, two novel algorithms are respectively proposed, aiming at the improvement of quality and speed.In this paper, a new algorithm called inner predictive diamond fast motion estimation is proposed, applying inner predictive technique to speed up original diamond search. It also takes advantage of the correlation of adjacent macro-blocks to predict the initial search point, and uses appropriate search strategy for different motion style macro-blocks. Experimental results show that the proposed algorithm greatly improves searching speed.Traditional estimation algorithms based on the local unimodal error surface assumption have an inherent shortcoming of getting into the local optima, which leads to degeneration on the quality of video images. Therefore, an adaptive motion estimation algorithm using simulated annealing is proposed to overcome this shortcoming. It is highly adaptive since its parameters are setup according to the characteristic of macro-block in search process. Moreover, there is refined search after SA to enhance matching precision. The simulations show that this proposed algorithm is robust against local minima and improves searching accuracy remarkably. In conclusion, it depends on the different requirement between quality and speed to choose the more suitable one of above proposed algorithms.