Abstract
Polar complex exponential transform (PCET) is superior to pseudo Zernike moment-based method in terms of kernel generation, numerical stability and easier implementation. Their performance degrades under additive noise such as white Gaussian noise. Moreover, these methods show poor performance against directional information of texture. In this paper, a new rotation- and scale-invariant method for texture analysis using Radon transform and PCET for textured image is proposed. Scale and translation invariance is achieved by normalization process in Radon space, and rotation invariance is obtained by combining Radon transform with PCET. A k-nearest neighbor classifier is employed to classify the texture. To test and evaluate the proposed method, several sets of textures were experimented with different scaling, translation and rotation in different noisy conditions. The correct classification percentage is calculated under the varying standard deviation. Experimental results show pre-eminence of the proposed method as compared to the existing invariant texture analysis methods.