Mobile OCR Using Centroid to Boundary and Backpropagation Neural Network

Mobile OCR Using Centroid to Boundary and Backpropagation Neural Network

Author		: BAKTI ANUGRAH YUDHA PRATAMA; TJOKORDA AGUNG BUDI WIRAYUDA; KURNIAWAN NUR RAMADHANI; Febryanti Sthevanie
Published on	: International Conference on Information and Communication Technology 2015

Abstract

In this research, the writer proposed feature extraction method using Centroid to Boundary. Centroid to Boundary has processing time better than some other methods, this method is also invariant to size and rotation of image. The method will get a feature of character based on distance from center point of character to its contour . For classification, Backpropagation Neural Network can improve the system performance better include computational processing time and accuracy. From the experiment, can be concluded that the OCR system built is well enough to recognize the character with the font type inside the training set. But, it must be trained furtherly to recognize other font type, especially with the significantly different characteristic font type. The image capturing distances affect the system performance significantly especially on the accuracy aspect. The smallest average of WER value generated by the distance about 1 ??? 10 cm with the size of image character 30 x 30 pixel.