Research On Adult Image Recognition Based On Detection Of ROIs

With the steady growth of the amount of images publicly available on the Internet, adult image recognition is of great significance for web security and content monitoring. In this work, the single-person adult pictures were studied. Current skin-based adult image recognition algorithms usually have a high false positive rate. In order to reduce the false-positives efficiently, an adult image recognition algorithm taking model's torso part as ROIs(Regions of Interest) is proposed in this thesis.This algorithm utilizes Poselet to detect the ROIs with plentiful discriminative information. Each poselet provides examples for training a linear SVM classifier which can then be run over the image in a multi-scale scanning mode, and the outputs of these poselet detectors vote for the localization of torso and body parts. Then based on the ROIs, we obtain discriminative Fisher Vectors for nude breast image classification. However, due to the varieties in human body appearance, there is a certain shifting between the ground-truth position and the output of the torso detector. In order to overcome the weakness, an adaptive algorithm is proposed in this paper. The algorithm selects several torso candidate areas according to the confidence value of the torso detector, and then integrates the discrimination results of several areas to obtain the final result. In addition, in order to train the SVM classifier based on torso, a set of 30,000 porn images of single persons was collected, and the pornographic regions in the images are manually labeled. The labeling information can be used to generate the training data automatically.In order to evaluate the method we have built a new large dataset, including adult images, benign images and bikini images. Experiments on this dataset reveal that the proposed method obtains accuracy of 91.7%, which is much higher than the traditional skin color-based method.The poselet-based torso detection obtains more porn-related information, thus the proposed method can detect adult images with a high detection rate as well as a low false positive rate, achieves requirement of practical application.