关键词: 图像比例特征, 肤色一致性, 人体结构, 改进的人脸检测模型

Abstract:

Currently the exposed skin ratio is still frequently used by most network pornographic image filtering methods. However, it has some defects such as the interference of analogous skin color, the interference of nonkeyarea skin and sometimes the irrelevance of the whole exposed skin ratio and the exposed extent. We propose a network pornographic image filtering algorithm based on the consistence of skin color and human body structure, which calculates the face size ratio, the whole exposed skin ratio after the skin color is improved by face skin color data in the skin color model and the keyarea exposed skin ratio. The problems mentioned above have been effectively solved by using these three ratio features. Besides, we add a rotation process to the face detection module, so it can detect severely inclined face effectively. Experimental results show that the recognition rate of the proposed algorithm can achieve 91.5%, which meets the practical demand.

Key words: image ratio feature;consistence of skin color;human body structure;improved face detection module