Recognizing objectionable pictures using sparse coding

In recent years different methods for detecting objectionable images have proposed. Generally these methods are based on skin color detection and extracting features from human body. In this paper a variant of SPM method is proposed in order to discriminate normal images from objectionable ones. In this method first SIFT features are extracted. Next features are learned by sparse coding the features of previous step. Finally classes are separated by a linear SVM. This approach remarkably improves the scalability of the training phase. The proposed system is tested on 80,000 images and experiments indicate that it outperforms other methods including methods based on histogram features and nonlinear classifiers.