Research On The Technologies Of Special Content-based Erotic Image Filtering

With the rapid development of the Internet, network security and network health problems have been suffering from. Due to the openness of the network, making some sensitive pornographic images on the network the wanton spread, with the arrival of the era of big data some anti pornography software appeared in the details of the lag has not reached the expected effect of anti pornography. According to CNNIC statistics, the number of Internet users in China has been ranked first in the world, but the number of Young's development is very rapid. In order not to let the erotic content network environment pollution, don't let young people's physical and mental healthy growth affected, eliminates the waste of network,network to clean environment is particularly important. In this context, the research on content-based image filtering technology is very meaningful. The central idea of image filtering is the picture of the picture library classification process. At present,there are many kinds of technical means for the filtering or shielding of erotic images, but most of the use of URL filtering, keyword matching, based on the sensitive image filtering. In this paper, we study the problem of sensitive image filtering based on content. In this paper, the image processing technology, database, pattern recognition technology and computer vision technology,the sensitive image filtering technology research,research goal is to improve the positive detection rate and detection speed of image filtering. In this paper, the filtering system is composed of skin color detection model, texture detection model, face recognition,image feature extraction and classifier.Color detection in sensitive image filtering system is very important, can accurately find the skin is the key part of the whole filter system. The commonly used color detection algorithm includes color space model algorithm, the algorithm of the seed pixel neighborhood expansion model and the Bias classification algorithm based on the color histogram. Bias classification algorithm this paper adopts the statistical histogram of skin detection.Skin color detection from the point of view of color, will be a lot of color and color of the same color as the skin to deal with, so the rate of miscarriage of justice will be very high,texture detection can effectively reduce the miscarriage of justice. The texture detection method based on statistical model is more, skin texture detection method based on signal processing methods of research are relatively few. In this paper, we use Gabor function and edge detection combined, obviously some obvious texture in many areas can reject the edge detection method and carefully use Gabor function were detected, ultimate goal is the first step in skin color detection and along with accurate and complete extraction of the true color.This paper introduces the artificial fish swarm algorithm in edge detection to improve the detection efficiency and accuracy.The study found that the vast majority of sensitive images contain human face information, and the number of faces is generally less than 4, the face detection is introduced into the filter system can be very good to distinguish whether the image is sensitive image. In this paper, we use the improved AdaBoost face detection algorithm based on chaotic artificial fish to detect the human face, and the detection speed is improved obviously.Classification feature vector extraction and classifier selection. Seven classification feature was ultimately chosen as the input of the classifier, the specific features are: connected part of the skin color of the image accounted for percentage, number of part of the skin color components, maximum skin color area accounts for image proportion, skin color area and skin color rectangular area ratio, maximum skin color connected zone width and image aspect ratio, the average probability of skin color pixels and face a number. The selection of the classifier is the key of this paper to achieve the desired filtering effect. Because the decision tree model is simple, the operation speed is fast and so on, this paper finally selects the decision tree as the classifier of this paper. Finally,the experimental results are achieved in this paper.