A neural network approach for content-based image retrieval using moments of image transforms

Abstract

Due to ever-increasing explosion of multimedia and storage devices available, accessing large image databases becomes inevitable. Furthermore, the availability of high-speed Internet has raised drastically the level of multimedia exchange by users across cyberspace every second. Hence, content-based image retrieval is gaining importance day by day. Therefore, this work proposes content-based image retrieval based on moments in edge map of different transforms such as Walsh–Hadamard transform (WHT), discrete cosine (DCT), and discrete wavelet transform (DWT). The first- and second-order moments of edges of these transform coefficients are combined with moments of color and color objects to improve the average retrieval efficiency. It has been shown that DWT exhibits better average retrieval efficiency compared to FWHT and DCT on Corel data base.