Motion estimation and memory interface for video compression

Abstract

Block matching motion estimation has been widely adopted to reduce the temporal redundancy of video sequences in various video compression standards. In those applications, motion estimation has a large effect on overall computational complexity and memory bandwidth. In this thesis, an efficient algorithm is proposed to reduce the computational complexity of block matching motion estimation by using the characteristics of spatial correlation. Identifying the boundary macroblocks associated with identical motion vectors, the proposed algorithm skips the motion vector search for inside macroblocks. To compromise the computational complexity and the estimation accuracy, the number of inside macroblocks is restricted to the threshold and conditionally switched to the minimum for the alleviation of the incorrect motion vector propagation. Experimental results show that the proposed algorithm reduces the number of search points, executed instructions, and data memory references by 51.1%, 47.0%, and 45.2%, respectively, compared to the conventional motion estimation at the cost of negligible performance degradation, resulting in reducing the computational complexity and the memory bandwidth requirement of video compression. The thesis also proposes a memory interface to increase the bandwidth of synchronous memories that are widely adopted for video compression applications. Memory latency and energy consumption can be reduced by effectively managing the states of banks. The local access history of each bank is considered to predict whether the successive memory access refers to the same row in the same bank or not and manage the states according to the prediction. Two-bit state machines are employed to predict the next memory mode based on the history of memory references. Experimental results show that the proposed interface, at the cost of negligible area overhead, reduces the memory latency and the energy consumption by 44.6% and 46.9%, respectively, over the co...