earticle

영어

This paper presents for isolated handwritten Roman numerals recognition a research interested for carrying out both comparisons between the performances in terms of precision and rapidity, the first comparison is realized between four hybrid methods used to extract the features from numerals that are the zoning combined with Radon transform in first time, then combined with Hough transform in second time, then combined with Gabor filter in third time, finally combined in the same time with all these three descriptors, while the other comparison is performed between three classifiers which the first one is neuronal that is the multi-layer perceptron whereas the second is probabilistic that is hidden Markov model, the third classifier is realized via a combination between these both classifiers. For this purpose we have employed for pre-processing each numeral image the median filter, the thresholding, the centering and the edge detection techniques.