Wavelet Based Multi-Trend Structure Descriptor for Effective Image Retrieval
Mithun Natarajan, Sarmitha Sathiamoorthy · 2019
We propose a novel descriptor for content based image retrieval called wavelet based multi-trend structure descriptor (WMTSD). The proposed WMTSD is an enhanced approach of multi-trend structure descriptor (MTSD). To diminish the computational cost of MTSD with the preservation of retrieval rate, WMTSD is derived. In order to diminish the time cost, an image is decomposed into optimum level using discrete Haar wavelet transform and it produces multiresolution pyramid image. The multi-resolution pyramid image at optimum level is utilized to estimate the feature descriptor. The optimum level for the decomposition of an image is determined empirically. The proposed WMTSD is exhaustively tested on Corel 1k, Corel 5k, Corel 10k and Caltech benchmark datasets. The experimental study very clearly illustrated that the proposed WMTSD and MTSD attains almost similar retrieval accuracy. But, the time cost of WMTSD is too small then the MTSD which is required more in the era of big data. Euclidean distance measure is adopted to assess the similarity between the images. Precision and Recall are the measures used to assess the proposed approach.