Accurate Estimation of Rotation Angle through Zernike Moments and Image Matching

Abstract

In this paper, we propose a Zernike moments (ZMs) based global features to improve accuracy rate of the content based image retrieval system. In our approach, the real and imaginary components of ZMs are corrected by using the phase relationship among query and training images. The corrected real and imaginary components are used individually to measure the similarity among images. Therefore, the proposed descriptor attains twice the number of features at a given order as compared to using ZMs magnitude only feature as in traditional methods. Besides, the proposed system eliminates the step of estimation of rotation angle for correcting the phase coefficients of ZMs in order to make them rotation invariant. The accuracy of the system is improved further by incorporating local features, which are obtained by computing histograms of distances of linear edges to centroid of image. The linear edges are detected using Hough transform. Both the global and local features, used in the proposed descriptors are robust to geometric and photometric transformations. The results of an extensive set of experiments demonstrate that the proposed system supersedes existing approaches to image retrieval.