The Research on Real-Time Performance of Character Recognition Technique Based on Machine Vision
|School||Hebei University of Technology|
|Course||Communication and Information System|
|Keywords||OCR Real-time Feature extraction Template matching|
The topic of this paper is the real-time character recognition. Through the presentation of a key step of character recognition technology, how to improve the real-time performance of character recognition is analyzed. The corresponding improved algorithm in the stages of character location, feature extraction and matching can be used to enhance the recognition speed.First, some basic theories of image processing are introduced in this paper, and image processing techniques are used to filter and de-noise . The method of real-time iterations to seek global threshold is bright in to achieve image binaryzation. The median filter is applied for noise filtering. Second,the related character location technologies are researched, based on the existing location algorithm theory, two possible characters for the label positioning algorithm are proposed, also compared two algorithms,the advantages and disadvantages of them are analyzed. Considered the real-time performance of character recognition, the character location and correction methods based on anti-rotation template matching is selected to isolate characters region in the image taken from the industry occasion, and then characters image should be rotated to a horizontal position in order to achieve a single character segmentation by using method of projection. After analysis of the characteristics of the digital and capital alphabetic characters, special nodes and external contour with template match technique are used to recognize these characters,which improve the real-time performance of the system. At the stage of recognition, using multi-level classification of template matching, the number of endings is considered as integrate features of those 10 digital and 26 alphabetic character images, which can be used for character coarse classification. Features of some key points are viewed as the major specific features, and then the characters on industrial containers can be recognized, and the external contour of character is used to discriminate similar characters. Finally Character extraction and character recognition in work pieces are designed to verify the practicability of the proposed algorithms.Experimental results show that: the proposed approaches are practical and effective, the average recognition rate for the printing characters of the digital characters is 98.49%, while for the alphabetic characters is 98.32%. The same time, the entire system algorithm is quite simple, complexity is low, and the character recognition time of the tags of work pieces is about 54ms/piece to meet the requirements of fast identification. So the system has broad application prospects.