Image/ video's automatic annotation considering semantics' tolerance relation

Abstract

In this paper, an approach of image/video's automatic annotation considering semantics' tolerance relation between pre-defined classes is proposed. The image/videos' semantics is represented by a vector of associative values with the pre-defined classes, and eliminating the semantic gap induced in the process of automatically generating associative values is dealt with by using the constructed semantic tolerance relation model (STRM). Moreover, the method how to generating associative values with pre-defined classes regarding different dimensions is described. Image/videos' retrieval results based on the predefined classes shows the effectiveness of the proposed methods.