Color-based Human Hand Segmentation Based on Smart Classification of Dynamic Environments

Abstract: Color is an effective and widely used feature for hand detection. In order to dealwith the problematic situations such as hand color diversity and the variationin background and lighting conditions, a multi-classifier supervised learning approachis proposed using the color information of each pixel. Training imagesare first clustered into dozens of groups based on their global color histograms,and a linear SVM classifier is independently trained within each group. In thisway each group can represent a specific chrominance or luminance situation,and each classifier is optimized for a specific segmentation task.In this project, the optimal clustering solution for the classification task onthe given dataset is explored, and different combinations of four color spacesRGB, HSV, YCbCr and LAB are tested to construct clustering histograms ortraining features. Test results on the dataset show that clustering in LAB colorspace and training in multiple color spaces RGB HSV YCbCr LAB can producebetter results than other combinations, especially those using single color space.It also outperforms a recent hybrid color space solution for skin detection. Theproposed multi-classifier approach is easy to implement and computationallyefficient compared to most existing methods for hand segmentation.