A noise resistant image matching method using angular radial transform

• A new improved distance measure for angular radial transform (ART) is developed.
• Recognition performance of the proposed method is better than the Euclidean distance measure.
• The proposed similarity measure provides excellent recognition performance on noisy images.
• The performance of ART is comparable or better than the Zernike moments (ZMs) features.
• ART is very fast and simple to implement than ZMs.

In this paper, we extend the concept of the optimal similarity measure, originally developed for Zernike moments (ZMs) which belong to a class of orthogonal rotation invariant moments (ORIMs), to angular radial transform (ART) which is non-orthogonal. The proposed distance measure not only uses the magnitude of the ART coefficients but also incorporates phase component unlike the existing L1L1-distance   and L2L2-distance   measures which use only the magnitude of ART in image matching problems. Experimental results show that the new distance measure outperforms L2L2-distance measure. The performance of the proposed method is highly robust to Gaussian noise and salt-and-pepper noise even at very high level of noise. The results are compared with the ZMs-based optimal similarity measure. It is shown that the recognition rate of the proposed distance measure is comparable to that of the ZMs, however, at very low computational complexity.