关键词: 色情识别, 级联模型, 深度学习, 卷积神经网络, 色情封面检测, 精细至粗略

Abstract: In the Internet era, as people communicating and data exchanging frequently, the spread of malicious information is becoming more and more serious. Therefore, it is particularly important to identify and filter pornographic images on the Internet. In recent years, the appearance of deep learning has greatly pushed forward the frontier of computer vision research, and computer vision tasks like image classification and recognition have greatly benefited from it. Compared with traditional methods, the features automatically extracted by deep model have better representing power. In this paper, we propose a pornographic image detection method based on the deep learning and model cascade. For the sake of getting better features, the strategy of fine?to?coarse classification is adopted to deal with the diversification of data distribution. Specifically, faced with the problem of pornographic cover detection, we combine traditional image processing methods and deep learning model. As a result, it improves the performance of cover detection effectively.

Key words: pornographic recognition, cascade model, deep learning, convolutional neural network, pornographic cover detection, fine to coarse