Research on Key Technologies of Content-Based Erotic Image Filtering and Its Application

Abstract

Research on content-based erotic image filtering is a new technology emerged in recent years. This paper studies some key technologies of content-based erotic image filtering, analyses and compares three algorithms of detecting skin-color pixels in common use, estimates the effectiveness of the features extracted for classifying the images using the method of statistical threshold. In the end, a decision tree classifier is constructed for filtering erotic image. In order to reduce the error rate of classification for portrait image, the human face detector is used in the filter. The test of the filter is carried out on our image set with 1207 images (624 erotic images and 583 benign images), the correct rate of classification of our filter is 84.01%, while the precision for erotic image recognition is 89.74%, the precision for benign image is 75.13%