Sensitive Identification Detection And Recognition In The Terrorist Image

With the continuous development of Internet technology, massive pictures and text services have been unable to meet the increasing demand for information acquisition. Online video services which are more intuitive and contain more amount of information become the center of attention. The rise of network television and video websites brings network video services into thousands of households. Nowadays the situation of world anti-terrorism becomes increasingly severe, and organizations like ISIS, Taliban, East Turkistan and Tibetan separatist are poisoning the world. Online video provides them a way to spread religious extremist thought. Therefore filtering online videos related terrorism is extremely urgent. Based on this, we propose a method for detecting and identifying sensitive identification (logo and masked men) in terrorist videos. It lays the foundation for the research of identify the attribute of online videosThe main work of this paper includes the following three aspects. Firstly, taking full advantage of the terrorism-related video data provided by the national network office to build a small sensitive identification image library of the terrorism-related video. It includes training data set and test data set, which includes 50000 pictures of logo and masked men. Secondly, we use rectangular features to describe logo image structural features and combine Adaboost cascade classifier to detect logo in terrorist videos. Finally, we take advantage of the characteristic features that masked men images are multimodal distributed to detect the masked men’s head profile with Gaussian mixture model (GMM) and histogram of oriented gradients (HOG). Based on the detection result of masked men contour, we judge the masked men. We use Local Pixel Difference (LPD) method based on pixel statistics and complete masked men detection after secondary identification.In conclusion, the detection and identification of sensitive identity are divided into two categories, logo detection and detection of masked men. In order to verify the validity of the above methods, we conduct a lot of contrast experiments based on the sensitive identification data sets we built. The results prove the validity of the method.