On the Use of Edge Orientation and Distance for Content-Based Image Retrieval

Recently, various features for content-based image retrieval (CBIR) have been proposed, such as texture, color, shape, and spatial features. In this paper we propose a new feature, called orientation-distance histogram for CBIR. Firstly, we transform the RGB color model of a given image to the HSI color model and detect edge points by using the H-vector information. Secondly, we evaluate the orientation-distance histogram from the edge points to form a feature vector. After normalization of feature, our proposed method can cope with most problems of variations in image. Finally, we show some results of query for real life images with the precision and recall rates to measure the performance. The experimental results show that the proposed retrieval method is efficient and effective.