A Two-Layer Data Model for Image Retrieval Systems

Abstract

Content-based retrieval is one of the most important research issues in multimedia databases. Since the low-level features are used in content-based image retrieval usually, the gap between low-level features of images and high-level human concept becomes a problem. In this paper, we propose a two-layer image data model including the object layer and the image layer to resolve the problem. The object layer uses low-level image features such as color, shape and texture to describe each object of the image. The image layer represents an image by composing the objects in the object layer. Based on the two-layer image data model, we develop a prototype system for similar image retrieval. Two main algorithms in the system, the object-pair matching algorithm and the similarity matching algorithm, are proposed to measure the similarity between images. A two-layer relevance feedback mechanism is also proposed to update the weights in the two layers according to the user's responses for satisfying human subjective concept. The experiments show that the proposed approach can capture human perception and retrieve relevant images in two or three feedback processes effectively.