SkinSeg-cascade: a two-stage fusion framework for robust face detection via skin-region prior and cascade verification

With the rapid advancement of computer vision technology, face detection, as a crucial component, has been widely applied across various fields such as security surveillance, identity recognition, and intelligent interaction. This paper proposes a face detection and filtering method based on skin tone modeling and cascade verification. Initially, the image to be detected is transformed from the RGB color space to the YCbCr color space, and the Gaussian skin tone model is utilized to calculate the probability of skin tone, followed by applying Otsu's thresholding method for image segmentation. Subsequently, the Viola-Jones detection algorithm is employed for preliminary face detection, yielding a series of detection boxes. Then, the skin tone areas within each detection box are analyzed, and the proportion of skin tone is calculated as a confidence score. A cascade verification process is applied to filter out detection boxes with high confidence scores, ultimately generating accurate detection results. Experimental results demonstrate that this method can effectively detect facial areas and significantly improve detection accuracy through skin tone verification, while maintaining high operational efficiency. The approach presented in this paper offers a new perspective in the field of face detection and holds promising prospects for application.