ABSTRACT

this paper proposes the new Walsh Wavelet generation and its use in the Content based image retrieval. The Walsh Wavelet has been used to transform the images and sectored it into various sector sizes to generate the feature vectors of those images. The transformation of all images in the database has been tried three fold i.e. column wise, row wise and full transformation (with two planes plane1 and plane2). The retrieval performance results of these approaches has been compared with the results of Walsh transformed (row wise, column wise, full- plane1 and plane 2) image sectorization. The comparison of these results has been done based on individual class wise average performance of five randomly selected images per class for each sector sizes. The overall average performance comparison of all methods proposed has been done and it has been found that the column wise approach performs well. The performance measurement has been done by means of average precision-recall cross over point, LIRS (Length of initial relevant string of images), LSRR (Length of string to recover all relevant images).Since similarity measures play a very important role in CBIR. We have employed two similarity measures i.e. Euclidian distance and sum of absolute difference.