Multiscale principal components analysis for image local orientation estimation

This paper presents an image local orientation estimation method, which is based on a combination of two already well-known techniques: the principal component analysis (PCA) and the multiscale pyramid decomposition. The PCA analysis is applied to find the maximum likelihood (ML) estimate of the local orientation. The proposed technique is shown to enjoy excellent robustness against noise. We present both simulated and real image examples to demonstrate the proposed technique.