An improved sub-optimal video summarization algorithm

During the last few years the amount of digital video content has been increasing exponentially as a result of the proliferation of media sources like digital TV, streaming video internet sites like YouTube and wider availability of digital video cameras. The video data volume is so large that the only way a user can browse these libraries is through the use of time-condensation techniques. Video summarization achieves time-condensation by choosing a sub-set of frames of the original video creating a summary hopefully representative of the source video. The frame selection process can be directed according to different principles, based on either subjective or objective frame-relevance measures. Previous works have used dynamic programming (DP) and greedy approaches to choose the frames that make up the video summary. We present an algorithm that performs better than the greedy solution achieving a performance closer to DP´s while keeping the greedy solution´s simplicity.