A Comparison between the Performances of Several Hybrid Methods of Features Extraction for Isolated Printed Tifinagh Characters Recognition

Abstract

In this paper, we present two comparisons in isolated printed Tifinagh characters recognition, in fact the first comparison is between four hybrid methods exploited in features extraction which are the retinal coding combined with the Hu then with Legendre then with Zernike invariant moments, finally with these tree moments at the same time; in contrast the second comparison is performed in order to deduce what is the most powerful between both kernel functions used in the support vectors machines classifier. For this purpose we have used for pre-processing each character image the median filter, the thresholding, the normalization, the thinning, the centering and the skeletonization techniques. Furthermore, the experiments results that we have obtained demonstrates really that the most powerful hybrid method is that combines between retinal coding and all tree invariant moments concerning features extraction while the Gaussian kernel is more performant than that polynomial concerning classification.