Research On Product Code Recognition Based On Digital Image

Optical character recognition, as an important method for automatic informationextraction and record, has been widely used in all walks of life, such as recognition forairline tickets or other notes, electronic document record, vehicle license platerecognition, etc. With the popularity of digital cameras, digital image-based characterrecognition technology will be more widely used.The code on the packaging of goods contains the information for logistics,automatically recognizing and recording the code can greatly improve the speed andcost savings. In this paper, the problem has been studied, and a solution has beenproposed.To recognize the code, firstly it is necessary to find the location of code in theimage. Texture-based location method is generally not suitable for real-time processingfor the large amount of calculation. Connected components-based location methodrequires foreground and background easy to distinguish in color or brightness so as tosegment the image. And then it is needed find the connected components of foregroundto finish the location. But in many cases, the distinction between foreground andbackground in the image to be recognized is not so clear to meet the conditions, thus theconnected components-based method is difficult to obtain a good robustness. Cannyedge detection algorithm performs very well on detecting weak edges, which provide abasis for edge based location method. This paper proposes a method to extract edge ofcharacter using Canny edge, adaptive local image contrast and Ostu threshold. And thenthe edge is used to locate the code and estimate inclination of code lines. The proposedmethod shows good performance.In this paper, several typical binarization methods have been simulated andanalyzed. Then a character edge–based binarization method has been proposed based onthe study on existing algorithms. Binarization results of code images show the goodperformance of the proposed binarization method. Then the chars are segmented by theproposed connected components-based segmentation algorithm, and recognized by BPneural network. For samples of non-white background code image, our system obtains arecognition rate of97.8%, and for low-contrast images of code with white backgroundour system can also reaches a high recognition rate of86.3%.