An improved system for concept-based video retrieval

Abstract

In this paper, we present a common framework of concept-based video retrieval and propose several methods to improve the performance of the system. 12 kinds of features, including color, texture, shape and local features are examined, including a modified HOG which is defined on image edges to reduce its computational complexity. The concept cooccurrence matrix and several assistant methods (B&W detection, audio detection and motion detection) are suggested to enhance the performance of the video retrieval system. Extensive experiments on TRECVID 2010 show the effectiveness of our proposed methods.