Affine invariant texture feature extraction based on invariant centroid

The affine invariance is a challenging problem in pattern recognition. This paper proposes a new method of achieving affine invariance based on global features. Originating from the mathematical model of affine transform, the algorithm converts the coordinates of images from Cartesian plane to polar plane based on the invariance of centroid of images in the affine transform. Using log transform and Radon transform, anisotropy scale transform and rotation transform are converted to shift transform respectively. At last the stationary wavelet transform is exploited to extract the shift-invariant features of images. Thus affine invariant features of images can be obtained. The algorithm is applied to the affine texture classification. Experimental results show a good performance.