Abstract: Skin color is important in image retrieval, face analysis and gesture analysis. Performance of skin detection is determined by the color space used. Histograms to visualize the specific distribution of skin samples in 2- and 3-dimensional color spaces are adopted. An evaluation function is proposed based on the Fisher's criterion to perform a quantitative analysis between skin and nonskin samples. The comparison result is validated by the verification experiment with Gaussian model. Through such intuitive observation and quantitative comparisons, better general color spaces are obtained for skin detection.

Key words: color space, Fisher's criterion, histogram, skin detection