A comparison between a neural network and a SVM and Zernike moments based blob recognition modules

This paper proposes a new algorithm to recognize the printed characters as part of the blob recognition modules of Optical Character Recognition systems. The algorithm uses a SVM classifier and Zernike moments for feature extraction. A comparison with another algorithm based on a five layer convolutional neural network is done. An analysis of the accuracy and the time needed to process one character leads to useful conclusions on the advantages and disadvantages of each algorithm.