A new one bit transform for low complexity block based motion estimation for HEVC

摘要

A new one-bit transformation kernel for block matching motion estimation with low complexity is presented in this paper. The Alter kernel proposed in this paper is used to obtain one-bit image frame by comparing the original eight-bit image frame with the image frame filtered with the proposed kernel. The proposed kernel does not require multiplication operation and the convolution operation is performed using additions and shift operations to construct one-bit frames. Compared to other one-bit transformation based methods, the proposed method provides better motion estimation results at macroblock level as well as for the smaller macroblock sizes. As in the proposed method, no multiplication operation is required, it can be implemented using integer arithmetic namely, addition and shift operations and thereby reducing the overall computational complexity.