Image recognition method based on high-order statistics

A new image recognition method based on high-order statistics is proposed. Firstly, Radon transform is used to project 2-D image to 1-D space, high-order statistics of the projection data are calculated, the scale and shift invariant is constructed in the bi-frequency domain. Secondly, Rapid transform is used to get the rotational invariant, which is better than other methods for image recognition. Thirdly, threshold analysis is used for feature data optimization. Finally, the feature data set is reduced to a small set based on the principle component analysis (PCA). The pattern recognition procedure is presented. Simulation results are given to obtain an insight into the efficiency of the proposed method.