Robust CBIR using sectorisation of hybrid wavelet transforms with Cosine-Walsh, Cosine-Kekre, Cosine-Hartley combinations

Content based image retrieval(CBIR) is a process of retrieving relatively similar images from image database based on content of an image. In proposed algorithm a novel idea of sectorisation is applied on hybrid wavelet transformed images to extract the unique feature of images. Here 4 individual orthogonal transforms Cosine, Walsh, Kekre and Hartley and their hybrid combinations like Cosine-Walsh, Cosine-Kekre and Cosine-Hartley are considered. To test the performance of the proposed method, total 1000 queries were fired on the image database containing 1000 images. Feature comparison is done using Manhattan distance. Performances of variations of proposed algorithm are evaluated using average precision. Result of proposed algorithm shows that Hybrid Cosine-Walsh, Cosine-Kekre and Cosine-Hartley gives better results than individual Cosine transform in image retrieval.