Multi-operator combination for character segmentation in complex background
Yuchun Fang, Jialu Yao · 2014
Automatic character segmentation is the first most fundamental and crucial step for the Optical Character Recognition (OCR) system. Though there have been a lot mature commercial OCR systems for controlled environment, the techniques of OCR are not as popular as expected for the complex uncontrolled environment. The bottleneck is character segmentation from noisy background. In this paper, we propose a multi-operator combined character segmentation algorithm to partition the characters from complex background. To handle the uncontrolled lighting condition, we propose a localized Canny operator for pre-edge detection and refine it with the Compass operator to promote the accuracy of edge detection under complex background. The proposed algorithm involves the advantage of the localized Canny operator in speed and the advantage of Compass Operator in accuracy. By comparing with other algorithms and analyzing the performance of the proposed algorithm, it can be concluded that our algorithm can achieve a better result of character segmentation in complex scenarios.