Hand Shape Detection and Classification Based on Co-occurrence of Partial Contour Feature

Abstract- We propose a method for detecting a hand and estimating the shape. We define a hand shape as combinations of local features. We extract local features and we search for co-occurrence of hand shape features. Co-occurrence of many local features is high ability to describe a specific shape. It is difficult to search all combinations of co-occurrence because there are too many varieties of combinations of co-occurrence. Instead of searching all co-occurrence combinations, we take hierarchical search strategy. We start from searching for co-occurrence of pairs of local features. Next we search for co-occurrence of pair features. We repeat searching for larger features. Hand shapes are detected and classified when we find large shape-unique features. I