Abstract—The goal of paper is to describe a flexible hierarchical classification algorithm and a new image similarity computing model based on mixing several image features for promoting the performance and efficiency of speed for content-based image retrieval. With an experimental comparison of a large number of different representative point selection approach, we are trying to seek for a method of uniform division of image space, eventually design a novel approach enlightening by high-dimensional indexing and social networking, that introduces the directivity to image classification that is used to explain the convergence of images to edge points of the high-dimension feature space in this paper. Meanwhile we find the laws of parameter setting of this algorithm through experiments and these laws acquires satisfied effects in different dataset. In addition to that algorithm, we also find some features assembling with reasonable formula to represent images better in color, texture and shape. Experimental results based on a database of about 50,000 person images demonstrate improved performance, as compare with other combinations in our descriptor set consisting of several general features mentioned below.
—CBIR, hierarchical classification, feature combination.
The authors are with the Department of Computer Science and Engineering, Beijing University of Aeronautics and Astronautics, Beijing, CHN (e-mail: lq@ nlsde.buaa.edu.cn).
Cite: Qiao Liu, Jiangfeng Chen, and Hui Zhang, "A Flexible Hierarchical Classification Algorithm for Content Based Image Retrieval," International Journal of Future Computer and Communication
vol. 2, no. 3 pp. 237-241, 2013.