SNIF: a simple nude image finder

The lack of control of the content published is broadly regarded as a positive aspect of the Web, assuring freedom of speech to its users. On the other hand, there is also a lack of control of the content accessed by users when browsing Web pages. In some situations this lack of control may be undesired. For instance, parents may not desire their children to have access to offensive content available on the Web. In particular, accessing Web pages with nude images is among the most common problem of this sort. One way to tackle this problem is by using automated offensive image detection algorithms which can filter undesired images. Recent approaches on nude image detection use a combination of features based on color, texture, shape and other low level features in order to describe the image content. These features are then used by a classifier which is able to detect offensive images accordingly. In this paper we propose SNIF - simple nude image finder - which uses a color based feature only, extracted by an effective and efficient algorithm for image description, the border/interior pixel classification (BIC), combined with a machine learning technique, namely support vector machines (SVM). SNIF uses a simpler feature model when compared to previously proposed methods, which makes it a fast image classifier. The experiments carried out depict that the proposed method, despite its simplicity, is capable to identify up to 98% of nude images from the test set. This indicates that SNIF is as effective as previously proposed methods for detecting nude images.