A Fast PDE Algorithm Using Adaptive Matching Criterion for Motion Estimation

In this paper, we propose an algorithm that reduces unnecessary computations, while keeping almost same prediction quality as that of the full search algorithm. In the proposed algorithm, we can reduce unnecessary computations efficiently by calculating initial matching error point from first partial errors. To do that, we use tighter elimination condition as error criterion than the conventional PDE algorithm. Additionally, we use different search strategy compared with conventional spiral search pattern. By doing that, we can increase the probability that hits minimum error point as soon as possible. Our algorithm decreases the computational amount by about 50% of the conventional PDE algorithm without any degradation of prediction quality.