A New Skin Detection Approach for Adult Image Identification

With rapid proliferation of adult content on the internet, development of software to detect and filter this content is gaining more and more attention. In this study, a new method for adult image identification is proposed. Accurate human skin detection and extraction of relative features are the bottlenecks in this regard. In the proposed skin detection method, first, Hue color information is utilized to examine whether skin is present in the image, and if so, using dynamic thresholding an estimate of the skin region is obtained. In the second step, for images that contain skin, an exponential function is fitted to the histogram of the estimated skin area. Based on the parameters of the fitted function, the final skin map of the image is extracted. Area, texture, and a new shape feature are extracted from each region and used for image classification using neural networks. 95% accuracy for skin detection and 93.7% accuracy of adult image detection, compared to other methods showed the significance of the proposed method.