Research And Implementation Of The Cylindrical Embossing Character Recognition Algorithm

In the process of industrial aluminum production,it is necessary to read the embossing label on the surface of the mold,and enter the production equipment management system,so as to facilitate the production process of quality control,production management and product tracking.The original solution is the human eye observation,and manually enter the computer system;the method is lo w efficiency,high labor costs and cumbersome operation.So the automatic entry of labeling information has become a bottleneck restricting the development of enterprise ERP.Designing a set of visual real-time identification systems for embossed character labels on molds has become an urgent need in the industry,Based on this,the research topic of cylinder embossing character recognition algorithm is put forward.In this paper,for the characteristics of cylindrical embossed characters as curved and concave characters,the lighting scheme of low angle annular red light is adopted in the image acquisition stage to improve the uniformity of illumination.The collected images contain noise interference,contrast difference and tilt defects,in the preprocessing stage,the filter denoising,image enhancement algorithm and character tilt correction are used to correct the character.The character region of interest is difficult to extract the characteristics,so the paper presents a composite positioning algorithm based on a priori knowledge of static ROI region rough positioning and morphological orientation of geometric features.The algorithm can quickly realize the rough positioning of characters.Based on the coarse positioning algorithm,the algorithm can reduce the computational complexity of the region and realize the efficiency and stability of the algorithm.Based on the complex imaging background of embossed characters,the rate of single character segmentation is low,the single character segmentation algorithm combining projection method and connected domain is proposed.Through the texture feature analysis,the single character segmentation can be achieved by using the connected domains method directly when the the connected domains are same as the number of characters.What the number of connected domains are not the same as number of characters are adhesion or fracture character,the projection method is used for segmentation.Thus overcoming the drawbacks of using the projection method alone for the lack of adaptability to the tilted character,but also to overcome the shortcomings that use of morphological methods for the adhesion and fracture of the character is not successful.As to the low recognition rate of embossed concave convex characters,the BP neural network classifier is trained for training data sets based on the collected cylinder embossed character images,At the same time,the feature extraction method of multi-feature fusion is used to extract the feature of normalized characters,and the character recognition rate is improved to the maximum extent.In this paper,the MATLAB is used as the tool verify algorithm.The test results show that the algorithm for cylinder embossing character recognition can achieve rapid recognition and the recognition rate is 89.33%.The optimized algorithm could be applied to the industrial character recognition system.