A Hybrid Skin Detection Model from Multiple Color Spaces Based on a Dual-Threshold Bayesian Algorithm

Abstract

As a preliminary step of many applications, skin detection serves as an irreplaceable role in image processing applications, such as face recognition, gesture recognition, web image filtering, and image retrieval systems. Combining information from multiple color spaces improves the recognition rate and reduces the error rate because the same color is represented differently in other color spaces. Consequently, a hybrid skin detection model from multiple color spaces based on a dual-threshold Bayesian algorithm (DTBA) has been proposed. In each color space, the pixels of images are divided into three categories, namely, skin, nonskin, and undetermined, when using the DTBA. Then, nearly all skin pixels are obtained by using a specific rule that combines the recognition results from multiple color spaces. Furthermore, skin texture filtering and morphological filtering are applied to the results by effectively reducing false identified pixels. In addition, the proposed skin model can overcome interference from a complex background. The method has been validated in a series of experiments using the Compaq and the high-resolution image datasets (HRIDs). The findings have demonstrated the proposed approach produced an improvement, the true positive rate (TPR) improves more than 6% and the false positive rate (FPR) reduces more than 11%, compared with the Bayesian classifier. We confirm that the method is competitive. Meanwhile, this model is robust against skin distribution, scaling, partial occlusions, and illumination variations.