Study Of Container Number Smart Recognition Algorithm

Digital image processing and pattern recognition in recent years has been widely applied to traffic monitoring and management (License Plate Recognition) checks to identify cancer cells in medical image recognition, remote sensing image recognition in many fields, pattern recognition has become the era of focus on technologies. Container as an important part of the international transport industry, its scope of application and quantity in a substantial increase Furthermore, China’s container transportation showing the development of a leap, a substantial increase in the amount of goods to container tracking and statistics a higher demand, research and development of the intelligent container identification system demand has become a hot social projects.Departure from the theoretical knowledge of image processing and pattern recognition technology, elaborated the structure of the intelligent container identification system, a careful study of the system, from image acquisition (on video captured by the camera to grab images) began, box image preprocessing, box character regional location, correction and character segmentation into feature extraction and character recognition. Different approach throughout the design process every step of the experiment, analyze the pros and cons of the algorithm to select for this article, the main work is as follows:(1) For the container image preprocessing, the night image brightness, contrast characteristics, contrast stretching and histogram equalization method of combining the night images were processed separately, the result is daytime and nighttime images can be better binarization.(2) In the distortion and tilt for the character of the box, the box number correction and character correction method of combining a good correction of the creases of the container surface and the geometry of the image tilt.(3) In tank positioning, the Container character characteristics using coarse positioning and precise positioning, the combination of methods, during which used boxes arranged in a priori knowledge, rough location, box number priori knowledge in a wide combination of crude positioning box for precise positioning. (4) For the segmentation of the characters, the three ways to do the experiment, the color image segmentation method, the projection histogram segmentation method segmentation method based on regional growth experiments, the choice split better based on regional growth segmentation method.(5) For the container number recognition, were established digital network letter networks, digital-subtitles network, respectively, made the training and testing. Recognition rate of test results, integrated the advantages and disadvantages of each method, combined with the character-based structural characteristics to identify the advantages and disadvantages, and eventually adopted the characteristics of the BP neural network is a recognition and confusing characters into the secondary processing library, with character-based identification method for secondary identification, to ensure the reliability of the identification results.Application of Matlab2010b for image acquisition to the identification experiment, the experimental results show that a high recognition rate and the prospect of practical application.