Approach for Retrieving and Mining Video Clips

Abstract

Multimedia (include video, images, audio and text media) is characterized by its highdimensionality, which makes information retrieval and data mining even morechallenging. This research proposesa method to build an indexes database for hugecollection of video clips, to make the video retrieval and mining much more efficientand perfectthat by considering similarity in both text of sound and features of frames.The proposed method has the following steps: First, isolates video motion from soundin the video clips. Second, converts the sound to text and index the result withdatabase. Third converts video motion to shots, then select the master frame for eachone and extracts the feature vector for them such as color, texture, shape and others andfinally index the result with database. Fourth,combines the two resulted indexeddatabase (Second and Third steps)into one database and make it the final and standardfor both retrieval and mining.