Special Video Content Detection Based On Semantic Concepts

There are massive videos spreading on the internet with the development of information technology and popularization of smart devices, and among these videos there are a lot of terrorist ones which are seriously threaten the public safety. It is meaningful to study the issue of harmful video content recognition, retrieval and filtering, and it is full of challenge as well.In this paper, we studied the problem of terrorist video detection and associated technologies, and the main works are as follow:1. We constructed a video dataset for terrorist video detection algorithms evaluation. We firstly studied the progress of special video content detection, and associated datasets. We constructed the terrorist video dataset refer to existence datasets, and did statistical analysis about the video collection channels, video specifications, video concepts for annotation and annotation methods and the final annotation results and so on.2. Features and classifiers for visual semantic concept detection were discussed. Two kinds of color histogram construction strategies were proposed, and the problem of parameter selection for color histogram was studied. Furthermore, we did simulation experiments for using color moment, local binary and histogram of gradient as feature in visual semantic concept detection tasks. In addition, the performance of seven classifier instances and their best matching feature for each visual semantic concept were studied.3. The key technologies for terrorist video detection were studied and terrorist video detection based on visual semantic concepts was implemented. A gray mass center based video key frame extraction method was proposed, and best feature and classifier combination for visual semantic concept was discussed. Simulation experiments were implemented to evaluate the performance of the proposed visual semantic concept based terrorist video detection framework.