Weighted fusion of LBP and CCH features for effective content based image retrieval
Swati Jain, Tanish Zaveri, Shailee Patel · 2016
Content based image retrieval for remote sensing images is challenging research area in recent years. In this paper, the weighted image fusion algorithm of morphological feature circular covariance histogram (CCH) and local binary and tetra patterns (LBP and LTrP) is presented. These features are effective for defining the image texture content and the content of the image but has limitations in defining the overall content of the image when individual features are used. Using proposed algorithm, retrieval results are improved when the features are combined compared to retrieval results of the individual features. For a given query image, there are instances where non-relevant images are assigned better ranks than relevant images. Such instances are minimized, thus weeding them out. This minimization process results into a weight matrix, which is used for obtaining the combined distance of the query image with database images. This combined distance is finally used for ranking the images. The proposed algorithm is experimented with UC Merced LULC image data set of 2100 images with 21 classes and 100 images in each class. All the 2100 images are taken as query image at a time and performance is evaluated for each image as a query image. The retrieval results obtained using proposed algorithm is compared with results obtained using individual and combined feature set with and without weight matrix. Simulation results are evaluated based on precision gain ratio and normalized rank which is significantly improved that proves that proposed algorithm is more effective.