Abstract: The segmentation results of traditional skin detection algorithms are always unsatisfactory for the insufficient determination information and difficult choice of the threshold. So a novel skin color detection method is proposed, whose main idea is spreading from the initial skin seeds according to some rules referring to the connectivity and consistency of local information based on a sequence of images to obtain the final recognition result. Experimental results show that after combining with the proposed method, the results of traditional skin color segmentation algorithms can be significantly improved.