Querying image database by video content

Abstract

A content-based image retrieval system using an image sequence as a query is proposed in this study. The proposed system is applied to a fish database in Taiwan, which is collected by the Institute of Zoology, Academia Sinica. Major contribution of this study is threefold: (1) The proposed queryby-an-image-sequence scheme can eliminate the problems caused by non-rigid fish motions, serious deformations, and partial occlusions. (2) In traditional query-by-an-example schemes, queries with different imaging conditions may generate completely different retrieval results, but the proposed query scheme can overcome varying imaging conditions. (3) The proposed fish representation scheme is invariant to translation, scaling, and rotation. Experimental results have proven the effectiveness of the proposed system.