Image retrieval using colour features

Abstract

With the significant outgrowth of hardware technology, the amount of pictorial information has increased remarkably. Due to this increase in image data, the need for an efficient indexing technology arises. In earlier retrieval systems, images are described using plain text and retrieval is based on the text description using conventional information retrieval techniques. These systems are good for retrieval abstract concepts captured in images. However, it suffers the problem of subjectiveness and manual entry of text description. To overcome these limitations content-based retrieval techniques using image features have been proposed. The main features used for image retrieval are shape, texture and colour. The purpose of our work is to develop an efficient content-based retrieval technique based on colour. Researchs on colour feature have shown very successful results. However, these techniques have been applied to the overall image content. The processing time for extracting colour features is another problem. In this paper, we propose three extensions to the basic colour histogram matching technique: (a) the use of hue value of HSI colour space; (b) the use of segmentation for object level; and (c) the use of clustering.