Handwritten digit string segmentation and recognition using deep learning

Elitez, Orçun
The main purpose of this thesis is to build a reliable method for the recognition of handwritten digit strings. In order to accomplish the recognition task, first, the digit string is segmented into individual digits. Then, a digit recognition module is employed to classify each segmented digit completing the handwritten digit string recognition task. In this study, a novel method, which uses deep belief networks architecture, is proposed in order to achieve high performance on the digit string segmentation problem. In the proposed method, images of digit strings are trained into a DBN structure by sliding a fixed size window through the images labelling each sub-image as a part of a digit or not. After the completion of the segmentation, in order to achieve the complete recognition of handwritten digit strings, the segmented digits are classified using both DBN algorithm and support vector machines and the results of these algorithms are compared over CVL Digit Strings Dataset. The result of the segmentation which uses the proposed method is compared with the result of the segmentation algorithm using water reservoir concept. Moreover, the results of some benchmark algorithms which use the same database of handwritten digit strings are included in the comparison. The proposed method outperformed the state of the art methods and also the baseline algorithm using water reservoir concept for digit segmentation on the CVL Digit Strings Dataset.
Citation Formats
O. Elitez, “Handwritten digit string segmentation and recognition using deep learning,” M.S. - Master of Science, Middle East Technical University, 2015.