Content-sensitive Approach for Video Browsing and Retrieval in the Context of Video Delivery: VBaR Framework

Information is doubling at a rate of once every few months and the rate of increase is growing. Video and media transport plays an important part of that growth. It has spurred the development of broadband access and slowly gained increasing prominence especially for multimedia-rich Internet contents. Operators are now under pressure to efficiently use available resources for delivering targeted contents in a reliable and consistent manner. The present work proposes a new strategy to select intra-coded frames that best represent the entire video. The proposed approach, tri-step approach, uses low-level image features to select representative key frames which enable random access to any part of the video sequence while avoiding the overhead incurred to transmit periodic intra-coded frames. The first step discriminates non-transition frames by checking adjacent frame characteristics using low level features. The second step refines our selection of key frame by dropping non-commonly browsed frames, being the non-informative frames. The final step verifies the remaining frames if they are far apart in time to maintain the efficiency of video delivery over low bit rate channel. Thirteen video sequences, obtained from MPEG-4 industry forum, are used in the experiments. A framework, named the Video Browsing and Retrieval (VBaR), is developed and it allows user to analyze input videos at their full control. Experimental results, using the proposed framework, show the ability to effectively select intra-coded frames in video delivery. Povzetek: Predstavljeno je kontekstno odvisno indeksiranje in iskanje video vsebin.