Orthogonal moments have been widely used for image recognition and classification duo to their useful properties such as being less sensitive to noise and being very accurate in image reconstruction. However

their do not natively possess scaling invariance

essential image normalization and binarization process will lead to error of resampling and requantifying. A new scaling and rotation invariant analysis method for image recognition is proposed. In the proposed method

the Radon transform is utilized to project the image onto projection space

and then the analytic Fourier Mellin transform is applied to the projection space to convert the rotation of the original image to a phase shift and the scaling of the original image to a scaling of amplitude. In order to achieve a set of completely invariant descriptors

a rotation and scaling invariant function is constructed. Based on four features of the invariant function

a k-nearest neighbor classifier is employed to implement classification. Theoretical and experimental results show the high classification accuracy of this approach in comparison to the orthogonal moments based methods as a result of using the rotation and scaling invariant function instead of images binarization and normalization

it also shows that this method is more robust to white noise than the orthogonal moments based methods.