Character Recognition Based On Topological Structure

Inkjet Industry plays a important role in the domestic automation production,and it greatly improves the level of automation production,especially in the food industry,the accurancy of coding and marking the production date is closely related to the health of consumers.In this paper,we have a in-depth study about the coding chracter which is based on the machine vision,and improved the inkjet character key technology and system implementation.After analyzing the form and characteristics of the inkjet on the current market,we present an overall technical program of the inkjet character recognition,and describe the logic composition of character recognition.By reading a lot of relevant literature we analyzed and summarized the current development level and extensive application prospect of character recognition.In the whole process of character recognition,according to the distribution characteristics of the inkjet characters,it is divided into a bar inkjet and ring inkjet.Ring inkjet can be further corrected for bar inkjet,this paper presents a method which can adjust the bar area to the ring region based on moment model.In the pretreatment process,according to the scattering distribution of inkjet ink dots,the inkjet ink dots are valid "closed" by using the grayscale morphological,we proposed an image segmentation method based on a column,a good solution to the poor quality of image segmentation that caused by uneven illumination;In the character region corrective,respectively compared the maximum aspect ratio correction method,the horizontal projection of the largest variance correction method,the minimum axis of inertia and least squares fitting correction method,we propose a method based on the least relevant axis correction.In order to reduce the redundant portion of the character,highlighting the structural features,using refined methods to extract the skeleton of the character;In the character segmentation discussed the vertical split method of segmentation and connected component labeling method,and proposed a method of binary image based on the line seed tag.In character feature extraction,compared to the structural features and the contours feature,and according to the shortcoming of the two,presents a character-based topology of feature recognition method.In recognition algorithm, introduced the feature template matching algorithm and naive Bayesian classifier,we chose the Bayesian classification algorithm finally according to the special property of characteristics.Finally,according to the above theory we developed a system for a certain types of barcode recognition,and tested the recognition rate and the processing time,and got a goo d result.