Hand Detection Using Robust Color Correction and Gaussian Mixture Model

Reliable and fast hand detection is crucial for gesture-based human-computer-interface. It is natural to utilize hand-skin color to detect hands in images. However, skin-color feature is sensitive to variations in illumination, and occurrence of shadow. Therefore, color correction plays an important role for skin-color based detection algorithms. But classical Retinex method is not suitable for correcting hand-skin-color because that hand is a very uniform region. To deal with this problem, we propose to utilize an advanced color correction method: RACE. The RACE algorithm is a combination of Random Spray Retinex (RSR) and Automatic Color Equalization (ACE). Based on the corrected colors, we employ the Gaussian Mixture Model (GMM) to describe the skin-colors. Compared to single Gaussian model, the GMM can capture more complex variations caused by the difference of human races, gender, and etc. Experimental results demonstrate the effectiveness of the proposed algorithm.