Gradual Integration of Local Color information for Image Retrieval by Content: Application to Cell-CCV Method

Abstract

The goal of Content-Based Image Retrieval (CBIR) systems is indexing and retrieving images from the analysis of its visual content (color, texture, shape, etc.…). They are frequently faced with the problem of effectiveness of retrieving images. The performance of CBIR system depends much on the choice of descriptors used to characterize the image content. We have recently proposed a local color based method called Cell Color Coherence Vector (Cell-CCV); which was considered an effective method as it reflects both color and spatial properties of an image and requires less space and computing time overheads than Local Color Histogram descriptor. In this paper, we refined the above proposed method, to further improve the retrieval precision. Thus, we propose Gradual Cell Color Coherence Vector (G-Cell-CCV). In the proposed method, many Cell-CCV(s) operators are combined using multi-size grid of cells. The experiments were carried out on Wang dataset (Corel 1K) of 1000 color images. Results show that G-Cell-CCV performs significantly better than Cell Color Coherence Vector (Cell-CCV) and other CBIR systems in terms of average precision metric.