Researches On Motion Estimation Algorithm Based On Dynamic Threshold And Oriented Cross Search Pattern

Motion estimation algorithm is the core of video coding. An excellent fast motion estimation algorithm can effectively obtain the similarities between adjacent frames, reduce the computational redundancy, and ultimately achieve good compression results. Therefore, the motion estimation algorithms have been widely applied to various fields that have relations with video, and it has been a research hotspot of digital imaging technology.In this paper, a new adaptive motion estimation algorithm based on dynamic threshold and oriented cross search pattern (DT-OCSP) is proposed, which adopts different search pattern with classification of motion contents. Dynamic threshold technique gets threshold TH1of slow movement and threshold TH2of fast movement by averaging the sum absolute difference (SAD) of macro blocks of same movement in the reference frames. The classification of motion type is judged by the threshold of the SAD in the center macro block, which is an approach that can provide accurate motion type prediction and avoid mass storage as well as complex sorting operations. For the slow movement case, a direction-extended small-CSP (DE-SCSP) searching pattern is proposed in this paper, which is based on the small diamond search pattern (SDSP) by adding two extended searching points along the searching direction to achieve higher searching accuracy. In the fast movement situation, a direction-extension large-CSP (DE-LCSP) searching model is proposed, which searches from rough to fine. It can accelerate the searching speed and enhance the searching accuracy at the same time. The experiment results indicate that the proposed DT-OCSP is a robust and efficient motion estimation algorithm. Compared with existing motion estimation algorithms, DT-OCSP possesses faster speed while maintain similar searching accuracy at the same time.