Extended Performance Analysis of Various Combinations of Hybrid Wavelet Transforms for Sectorisation Based Image Retrieval

In Content Based Image retrieval (CBIR) process, images are retrieved by using contents like shape, texture, color and transformed image content. Hybrid Wavelet Transform (HWT) can be formed using any two orthogonal transforms combining qualities of constituting transforms. Previous work has proved that image retrieval using sectorisation of Hybrid Wavelet Transformed images with Cosine first HWT Combinations performs better than respective individual orthogonal transforms. In this paper, extended performance analysis of various combinations of Hybrid Wavelet Transforms for sectorisation (with 4, 8, 12 and 16 sectors) based image retrieval is done. Here seven orthogonal transforms like Sine, Cosine, Haar, Walsh, Kekre, Slant and Hartley are used for generating Hybrid Wavelet Transforms. Experimentation is done on image dataset containing 1000 images of different categories. Performance is evaluated by using Average Precision. Manhattan Distance is used as a similarity measurement criterion. Results of proposed work show that Hybrid Wavelet Transforms having Sine as first transform performs better than the other considered HWT. Also Image Retrieval based on Hybrid Sine-Cosine transform sectorisation performs best in all experimented combinations.
This paper was published in IJICAR - International Journal of Integrated Computer Applications & Research - http://ijicar.com