Efficient video retrieval system using virtual 3D space

Abstract

We propose a technique for efficient retrieval of video scenes in a large number of digital videos. With the technique, cut frame images are automatically detected from videos, and arranged based on a visual feature in a virtual 3D space so that users can easily catch out the target image. We conducted experiments to confirm the effectiveness of the technique. The results show that retrieval by our technique is about 20 times as fast as retrieval by the fast forward playing mode at three times the normal speed.