Row, column and fused row-col R, G, B plane´s feature vector generation using DCT, DST and Kekre wavelet for CBIR

The feature extraction approach for CBIR proposed in this paper is based on frequency domain information of the image. Three different wavelets namely DCT, DST and Kekre wavelet transform coefficients are obtained from the image´s row and column mean vectors and are used as components of the feature vectors to be compared. New variation used as feature vector for this CBIR ; which has produce very good retrieval is fusion of row column mean vector coefficients obtained by applying the DCT, DST and Kekre wavelet transforms. Results are obtained for row vectors, column vectors and fusion of row-column vectors separately using two similarity measures namely absolute distance and Euclidean distance. Performance of the system is evaluated using two parameters `Precision Recall Cross Over Point (PRCP)´, and `Longest String´. After analyzing the results obtained separately for three planes, we have combined and refined them using the `Criterion OR´.