Image Retrieval Based on Block Color Averaging

Although color averaging techniques are widely used in content based image retrieval system because of its computational simplicity and effective retrieval, color averaging technique still suffer in identification of different objects which share the same average color information. To overcome this limitation of color averaging technique, a new technique based on block color averaging is proposed. The proposed technique involves dividing the image to equal sized blocks. The local mean of each block is extracted for feature vector generation. The distance between the features of query image and the features of database images is computed using Euclidean distance. The proposed block color averaging technique is tested on an indexed image database and compared with existing color averaging technique. Experimental results show that the proposed block color averaging technique performs better compared to existing color averaging technique with high precision and recall rate.