Block Histogram-Based Support Vector Machine For Skin Color Segmentation

Skin color segmentation by a block histogram-based Support Vector Machine (SVM) is proposed in this paper. The color model used is the Hue-Saturation (HS) model. Color information is represented by histogram of a block image on the HS space. To represent histogram information as accurately as possible, an irregular quantization partition approach on HS space is proposed. Histogram information of blocks from images under different environments is used to train SVM to make the method as robust as possible. SVM with Gaussian kernel is used. The proposed irregular histogram feature extraction method is compared with HS feature. Comparisons between the SVM classifier and other methods are also performed. The comparisons show that a better performance is achieved by the proposed approach.