H.264 video encoder implementation and optimization based on DM642 DSP

This paper focuses on H.264 video encoder implementation and optimization based on TMSDM642 DSP. We analyze the algorithm of JM8.0, which is the reference model of H.264 and found that the computation load of motion estimation is very big. In order to overcome the limitation of the motion estimation we put forward an improved strategy which adopts judging (0, 0), SS (square search) and early termination threshold algorithm for the motion estimation. Then we combine with the hardware configuration and instruction characteristics of DM642. We transplants and optimizes the reference model using fast albilinear interpolation algorithm. The experiment verifies that the proposed video encoder can distinctly reduce the computational complexity and increase encoding speed with little quality loss of the rebuild image, hence it enable to meet the real-time encoding constraints.