Robust candidate frame detection in videos using semantic content modeling

Videos are of the most popular rich media formats carrying large amount of visual, audio and textual information. In recent years people all over the world show great interest in video mining to extract meaningful patterns and knowledge to enhance the smart level of video applications. In this work Speeded Up Robust Features (SURF) are used to detect the candidate frames among the set of key frames extracted from a video content. By eliminating the presence of duplicate key frames the computational and time complexity of processing a large number of frames is reduced. From the identified candidate frames semantic objects with meaningful content are extracted which improves the efficiency of video mining applications like Video recommendation systems, Video concept detection etc. Experimental results show that the proposed approach eliminates the duplicate frames without a prior knowledge of the video content.