An adaptive grid-point detector by exploiting local entropy map

This paper describes an adaptive grid-point detector for the feature detection task in a pseudo-random structured light pattern. In the algorithm, a local entropy map is firstly constructed to evaluate the distribution of the projected pattern elements in the captured image. A mask in the shape of a cross is then used for the preliminary detection of grid-point candidates. With reference to the entropy map, the size of the cross mask can be determined adaptively. With considering the local symmetry property around the grid-points, a correlation procedure is then introduced for the final grid-point localization with sub-pixel accuracy. Experiments on real human face and comparison with previous methods are used to demonstrate its high performance.