Fusing magnitude and phase of wavelet moments for content-based image retrieval
Sanjay R. Patil, Sanjay Nilkanth Talbar · International Journal of Computational Vision and Robotics · 2015
The wavelet moments (WMs) are capable to represent the variations in the images as they provide both time and frequency localisation. In this paper the image retrieval system is implemented using complex wavelet moments (CWMs). The magnitude of the WMs of image is rotation invariant, but the phase component changes with image rotation. The magnitude and phase of WMs are incorporated and used both real and imaginary components of WMs so as to make WMs complex to describe as image. The rotation invariance of real and imaginary component is achieved by correction and comparison of phases of WMs of two comparable images with suitable distance metric. The proposed algorithm is applied to USC Rotated Texture database and Colombia Object Image Library (COIL-100) database and detailed analysis is carried out and performance is compared with other orthogonal moments. In many cases the complex WMs produce excellent performance. The average retrieval accuracy for 13 textures, Bark , Brick , Bubble , Grass , Leather , Pigskin , Raffia , Sand , Straw , Water , Weave , Wood and Wool are 100 , 85 , 92 , 99 , 91 , 92 , 95 , 97 , 76 , 94 , 99 , 74 , and 95 % respectively. Also, the experimental results on COIL-100 database reveals that the complex wavelet moment outperforms than complex Zernike moments and it is one of the useful descriptor for content-based image retrieval systems.