Abstract
In this thesis, the complexity scalability for H.264 motion estimation is to be explored. The model proposed uses Markov Chain Model to relate the prediction modes of the MB in the previous frame to that of the MB at the same position in the present frame. The Markov Chain Model includes a transition probability matrix which indicates the mode probability of MBs in the previous frames. The Markov Chain Model also includes a decision matrix which is composed of binary elements and 1 means the mode on while 0 means the mode off. In this way, we can build a set of constraint equations that the optimal solution represents the optimal mode decision.In order to make the problem practical, model simplification is executed. The decision matrix is modified and not all of elements in the decision matrix are binary. Hence, if we restrict the target complexity by a restricted complexity level, x, we can get the optimal mode decision at such complexity level. The experimental result shows that the proposed algorithm can not only utilize the target complexity well, but also control the motion estimation time well. Together with the proposed scalable inter-mode decision model we can make the motion estimation time more flexible with acceptable performance degradation.