Abstract— In this paper, we propose a novel method for improvingthe precision of Content based image retrieval [CBIR] by segmenting image intosub regions in a new way and use colour histogram with area coefficients. A newmethod is proposed to divide an image into 5 simple rectangular sub regions andassign weights according to their area. A bigger central rectangular area ischosen and assigned more weight as the centre region contains importantinformation about the image. Colour histogram feature is computed for eachregion separately and the resultant features are clubbed into a fusion vectoraccording to the region area coefficient. Additionally, we use univariatechi-squared statistical test for features selection to select best 35% of thefeatures for image similarity measurement. Chi-squared distances are used tomeasure the feature similarity for image retrieval. The proposed method hasbeen validated through experiments conducted on standard Corel1K image dataset.It is observed that the proposed CBIR method improves average precision of theimage retrieval by 6.33% as compared against baseline CBIR using colourhistogram feature over entire image.
Keywords: Content based image retrieval, Colourhistogram, Feature Vector.