In this article, the authors introduce a new algorithm to identify adult images that can effectively filter out images of naked human bodies in the internet. The algorithm detects eyes, which are known as the most salient component of a human face, and makes a statistical skin color distribution model directly from each input image by choosing reliable skin samples in facial areas near the detected eyes. Skin areas over the entire image are segmented robustly with the online constructed skin color model. The authors then extract a set of representative features characterizing naked bodies from the segmented skin areas and verify if the skin regions contain naked bodies through multilayer perceptron neural networked-based learning and inference of the representative features. Experimental results are given to demonstrate that the proposed adult image detection method can identify various types of nude images effectively compared to other conventional methods.
Seok-Woo Jang, Young-Jae Park, Gye-Young Kim, Hyung-Il Choi, Min-Chel Hong, "An Adult Image Identification System Based on Robust Skin Segmentation" in Journal of Imaging Science and Technology, 2011, pp 20508-1 - 20508-10, https://doi.org/10.2352/J.ImagingSci.Technol.2011.55.2.020508