Fusion Technique for Robust Human Skin Detection

Human skin color provides cue to infer variation in culture, race, health, age, wealth, beauty etc and promotes human recognition. Detection and classification of skin color and non skin color pixels is a challenging task and highly expensive. Human visual system incorporates color opponency. Variation in skin color of an image is caused by camera characteristics, ethnicity, subject appearances, background colors, shadows and motion. In fusion technique we used the product of two features to perform automatic skin detection. The frequency variation in RGB color space is removed before applying LO color and includes Logarithm coding and indexing. The skin threshold value(s) are calculated by using online dynamic approach and elliptical gaussian mixture model with 2D histogram and smoothened densities is used for classifying skin pixels. In our technique we used color image(s) obtained under different illumination conditions, different locations, to obtain single and/or multiple images. The proposed technique (i) reduces computational costs as no training is required, (ii) reduces the false positive ratio and provides comparatively more robust and accurate results in skin detection.