Skin detection based on contextual information

Abstract

In this paper a skin detection method based on the two-dimensional histogram of color images is presented. First, from the statistical perspective, we select the discriminant features in the hope of getting a better average TPR and FPR. Despite the other methods, new statistics of these features are extracted based on the 2-D histograms leading finally to the considering the contextual information. At the last step, the decision is reached based on a feature fusion strategy. The experimental results on the known databases (containing 103 images of humans under uncontrolled condition) demonstrate the performance of the proposed method.