The Research On Pornographic Image Recognition

With its fast development, the Internet is getting much closer to every field of the society. It changes our life style and speeds the information communication. It benefits us a. lot, at the same time, the pornographic information spreads widely in the Internet. The pornographic information, such as the pornographic image, does much bad effect to the society, especially to young people. Researchers pay more and more attention to the problem. By now, some Internet-eroticism filter systems usually filter eroticisms by blocking Web addresses and filtering erotic texts, but it is not enough, so the algorithm for content-based image recognition and filtering attracts more and more attentions. In this paper, the technique for content-based image recognition is studied. To detect pornographic image, a character choosing method and a pornographic image recognition method are proposed based on Adaboost (Adaptive boost) algorithm.Through the study of the current illumination compensation methods, an improved Gamma correction method is proposed to weaken illumination effect on image quality. Through the improvement of non-linear function, the performance of improved Gamma correction method is enhanced without more computation complexity. Different illumination compensation methods for color image are compared. Experimental results demonstrate that the proposed method performs better in weakening illumination effect and improving image quality.To remove salt-and-pepper noises from an image, an algorithm is proposed on the basis of linear prediction. Each pixel is estimated to be a signal pixel or possible noise pixel according to characteristics of the salt-and-pepper noise. Linear prediction is employed to the possible noise pixels, which are then replaced by the predicted values, while the signal pixels are kept untouched to preserve the detail of the image. Experimental results prove that the presented method has less calculation and shows strong denoising ability, especially for highly corrupted images.To detect pornographic image, the method to detect skin color mask image is introduced. The skin pixel is detected by the pixel’s color in certain color space and the texture of the corresponding skin region is analyzed in skin color mask image. Then according to the characteristics of skin adjacency region, some noises are wiped off.Some features of skin color mask image are extracted and put to Adaboost classifier. Adaboost classifier has two steps: learning step and test step. In learning step, the features obtained from the pre-classified images are put into Adaboost classifier and train it. Simulated annealing method is adopted to train the Adaboost classifier separately. In test step, features of image are put into learned Adaboost classifier. Whether the image is porngraphic one or not is up to the output of Adaboost classifier.The experimental results show that, the character choosing method based on Adaboost algorithm chooses the Characters of pornographic image and utilizing Adaboost classifier can recognize pornographic image effectively.