期刊名称：International Journal of Computer Science & Technology
印刷版ISSN：2229-4333
电子版ISSN：0976-8491
出版年度：2012
卷号：3
期号：1
页码：557-563
语种：English
出版社：Ayushmaan Technologies
摘要：The paper presents novel Content Based Image Retrieval (CBIR)methods using hybrid wavelet transforms generated from thecombination of Cosine transform matrix with seven differenttransforms namely Walsh, Haar, Kekre, Slant, Hartley, Sine (DST)and Cosine (DCT). Here the feature vector size per image is greatlyreduced by taking fractional coeffcients of the transformed image.The feature vectors are extracted in ffteen different ways fromthe transformed image. Apart from considering 100% coeffcientsof the transformed image, fourteen reduced coeffcients sets (as50%, 25%, 12.5%, 6.25%, 3.125%, 1.5625% ,0.7813%, 0.39%,0.195%, 0.097%, 0.048%, 0.024%, 0.012% and 0.006% ofcomplete transformed image) are considered as feature vectors. Theproposed CBIR techniques are implemented on a database having1000 images spread across 11 categories. For each proposed CBIRtechnique 55 queries (randomly selected 5 images per category)are fred on the database and average precision and recall valuesare plotted to get precision-recall crossover point. The resultshave shown improvement in performance (higher precision-recallcrossover point) with fractional coeffcients compared to completetransform of image at reduced computations resulting in fasterretrieval. The hybrid wavelet transform generated using DCTand Kekre transform matrices (DCT-Kekre) for 0.048% reducedcoeffcient set gives the best performance among the proposedCBIR techniques.
