ANDImage: An adaptive architecture for nude detection in image

Abstract

With the popularization of Internet access, institutions and parents have encountered serious problems to prevent access by employers and children to inappropriate content such as pornographic pages. The detection mechanisms try to circumvent or at least mitigate this problem by using of filters or mechanisms that enable the nudity detection in digital images. This paper proposes ANDImage, an adaptative architecture for nude detection in images, which uses a module zoning with color information to describe a nudity image. The SVM classifier is used in classifying samples. Experimental results show the efficiency of the model can achieve accuracy rates of up to 94.3% in a base composed of 2.250 nudity images and 3.110 neutral images.