摘要

Skin detection is an important preprocessing step of numerous image processing applications. Almost all the arithmetics address this issue pixel by pixel only allowing for the color information but humans do not. In this paper, a novel approach is proposed for skin detection, which takes use of both color information and region information. Meanwhile, taking into consideration the weakness of the statistical model, we do some post processing on the binary result to improve the skin detection effect. In the experiment, we firstly establish a histogram model based on training dataset. And then do the detection on the test dataset based on the renowned Compaq dataset which was widely used. Finally some special images are also selected to verify the effectiveness of our approach. And the experimental results also demonstrate that our proposed method is superior than the state of art arithmetics.

全文