Research On Handwriting Area Detection And Character Segmentation Based On Machine Vision

As an applied research field, optical character recognition has an extremelyimportant significance in reducing the workload of the workers and improving workefficiency and accuracy of the target certification recognition. In handwriting recognition,character segmentation is the prerequisite and basis of the text recognition, and thequality of character segmentation directly impact on the pros and cons of text recognitioneffect. However, in order to carry out the character segmentation, we must first detectand locate the text area in image or video. Therefore, the question of how to detect andlocate the text area quickly and correctly and how to segment characters correctly hasgreat significance for improving the overall performance of the character recognitionsystem.Based on the actual demand of the handwriting recognition applications, the paperuses the edge detection method to extract the text area in image or video, and combinesthe projection features of the image and characteristics of the character itself forcharacter segmentation. First, as to the detection and extraction of text area in the image，the first step is to use wavelet transform method to smooth the original image, then use8-direction Sobel operator for edge detection. Second, because in the actual collectedimage or video, some handwritten text may appear different degrees of tilt, this paper usea tilt correction method to correct the detected text area. Third, in this system, we use thehorizontal and vertical projection method to coarsely segment the text characters, andthen combine the characteristics of the character itself to reprocess the coarse splitresults.Experimental results show that, based on edge detection and connected componentanalysis method, the detection and location on handwritten text area has a good effect. Atthe same time, with the condition of simple adhesion between characters or a tilt of thetext area, using the integrated approach of tilt correction and projection Combination can segment the characters effectively with high accuracy. But for some complex charactersadhesions, the further research methods to resolve are needed.