Research Of Horror Video Scene Recognition Based On Multi-Instance Learning

Along with the ever-growing Web comes the proliferation of objectionable content,such as pornography, violence, horror information, etc. An effective web filtering isrequired to prevent end-users, especially children, from these harmful materials.However, compared with great progress in pornography content filters, horror videosfilter is still on the stage of earlier exploration, although its threat to children’s health isno less than the former one. In this paper, three algorithms are proposed based onMIL(Multi-Instance Learning) to promote horror video scene recognition. Thealgorithms include:(1) context-aware horror video scene recognition via cost-sensitiveSparse coding;(2) horror video scene recognition based on Multi-view Multi-instanceLearning;(3) horror video scene recognition based on multi-instance learning viadiscriminative instance selection. Experiments on horror scene data and benchmark MILdatabase show that the proposed methods are effective for horror scene recognition. Inaddition,(2) and (3) are also effective for other MIL problems. The research in thispaper will help to build healthy web environment and advance the development ofCBVR in affection analysis.