Shadow detecting using particle swarm optimization and the Kolmogorov test

An algorithm combining both gray level information and geometric features is introduced to detect cast shadows in gray level images. A simply connected candidate shadow region and a corresponding region are segmented by setting gray level thresholds, and neighbor-matching regions are constructed with a mathematical morphological algorithm. A shadow–non-shadow region pair is obtained from the result of Kolmogorov test for statistical features of both candidate neighbor-matching regions. Shadow regions are obtained by selecting the region with relatively lower average gray level from the matched region pair. The particle swarm optimization (PSO) algorithm is used to facilitate the feature extraction during the matching process. Experimental results showed the effectiveness of the proposed algorithm for cast shadow detecting in a single gray level image.