METHOD FOR FILTERING MALICIOUS VIDEO USING CONTENT-BASED MULTI-MODAL FEATURES AND APPARATUS THEREOF

摘要

A method for blocking a malice moving picture by using a contents-based multi-modal feature value and an apparatus therefor are provided to gradually block the moving picture including malice information by using a key frame-based judging engine and a frame-based judging engine and block the malice moving picture by using multi-modal feature values such as contents-based visual feature information, audio feature information, and temporal axis feature information in which malice feature information is extracted from the moving picture. An SVM(Support Vector Machine) model unit constructs a discrimination model including a malice moving picture classification model and a non-malice moving picture classification model(S110). A separating unit separates a video stream and an audio stream from moving picture contents(S120). The first extracting unit extracts a certain visual feature value from the separated video stream, detects a shot boundary, extracts at least one or more key frames from each shot, and generates the shot and key frame information(S130). The second extracting unit performs a certain pre-process for the video stream, and extracts a multi-modal feature value of the key frame on the basis of the shot and the key frame information(S140). A discriminating unit inputs the multi-modal feature value to the discrimination mode, and judges the malice of the shot(S150). The discriminating unit synthesizes the judged result of the shot, and judges the malice of the moving picture contents(S160).