A Semantic Segmentation Based Approach for Segmentation and Recognition of Touching and Overlapping Digits

Text line segmentation is one of the most important stages in the Optical Character Recognition systems. In order to acquire a good recognition rate, text lines must be segmented in high accuracy. Accurate separation of characters that overlap and touch each other in handwriting documents increases the accuracy of segmentation and recognition. For this purpose, we propose a new approach to segment the touching or overlapping digits. Two systems are designed for segmentation of overlapping digits and digit recognition. Proposed method is implemented to synthetic images containing two digits touching or overlapping each other. As a result of this implementation, it is seen that overlapping and touching digits are segmented effectively with 99.31% segmentation accuracy. Considering these results, it has been shown that the proposed method can be used as an effective tool in document analysis since it can provide solutions to problems such as text line segmentation and character recognition.