基于稀疏语义表示的不良图像检测算法

Pornography images are spread out in the multi-media communication, which is harmful to the mental health of teenagers. Thus, the pornography image detecting techniques are most urgent needed. From the perspective of image semantic analysis, we propose a Gaussian mixture model based skin region extraction method, which is boosted by the Bootstrap algorithm. To reduce the feature redundancy, feature points in the skin area are obtained by saliency-based visual attention model. Then "bag of words" model is incorporated to represent the images by the dictionary. Finally, the pornography information is filtered by a sparse representation algorithm based on the dictionary in the "bag of words" model. The experimental results prove the validity of the proposed method. In addition, the efficiency of the proposed method under different distance measures is compared.