Affine invariant feature extraction using a combination of radon and wavelet transforms

The paper presents a new framework for the extraction of region based affine invariant features with the view of object recognition in cluttered environments using the radon transform. The presented technique first normalizes an input image by performing data pre-whitening which reduces the problem by removing shearing deformations. Then four invariants are constructed by exploiting the properties of radon transform in combination with wavelets which enable the analysis of objects at multiple resolutions. The proposed technique makes use of an adaptive thresholding technique for the construction of invariants. Experimental results conducted using three different standard datasets confirm the validity of the proposed approach. Beside this the error rates obtained in terms of invariant stability in noisy conditions are significantly lower when compared to the method of moment invariants and the proposed invariants exhibit good feature disparity.