The Research On Dot Matrix Character Recognition Method In Food Packing Box

With the improvement of social economy and people’s living standards,the consumers’ demands for food,medicine and other product quality,safety and traceability are increasing.Local governments at all levels have established uniform cargo coding standards so that each item has a unique code.Based on this,all products are required to print dot matrix characters,which contain production date and traceable two-dimensional code during the production process.The key equipment to complete this task is the printer.The inkjet printer is widely used in modern food industry,cosmetics industry,pharmaceutical industry,automobile and other industries.Therefore,the research of dot matrix character recognition is an important part of industrial automation technology.The dot matrix characters in food packing boxes generally contain important information such as the production date,production batch,and origin of product.Therefore,to ensure that the dot matrix character information is complete at the factory and consistent with market access principles,it is necessary to detect information on the products.Traditional method of dot matrix character detection and recognition is performed manually,which has the problems of low efficiency and waste of labor costs.It is not suitable for large-scale production.Benefiting from the development of computers,industrial production is gradually shifting to automation.Using computer to automatically recognize dot matrix characters can not only improve production efficiency,but also reduce errors during manual inspection.The recognition methods of dot matrix characters in food packing boxes are studied in this paper.The main research results of this paper are as follows:1.The dot matrix character localization method based on FAST(Features from Accelerated Segment Test,FAST)corner detection algorithm is proposed,in order to solve the problem that the existing dot matrix character localization method is not applicable when the food packing box is interfered by various kinds of interference.This method firstly detects all FAST corners in the image based on FAST corner detection algorithm.Then,using the intensity value and spatial distribution of the corners,the wrong corners in the image are deleted.Finally,the remaining corners which belong to dot matrix characters are saved,and the location of the dot matrix characters is completed.2.The related algorithms of feature extraction in character recognition are studied.In the process of character feature extraction,this paper studies the following three types of features,including histogram of oriented gradient(HOG)features,pixel-by-pixel features based on principal component analysis(PCA),and grid features.The combination of these three features is applied to the character recognition algorithm.Experiments indicate that the features extracted in this paper are effective for dot matrix character recognition.3.A dot matrix character recognition method based on template matching and support vector machine(SVM)is studied in this paper.It overcomes the defect of low recognition rate which caused by using a single template matching algorithm or a single SVM algorithm.This combination method consists of the following three steps: First,the characters are recognized using a gray-based template matching method.Then,characters are recognized using a feature-based template matching method.In the end,compare whether the recognition results are the same.If they are different,the SVM is used for re-recognition.If they are same,the recognition result is output.Through experimental verification,the recognition method studied in this paper has higher recognition accuracy than a single template matching method and a single SVM method.4.A dot matrix character recognition method based on probabilistic neural network(PNN)is studied in this paper,for the problem of complex neural network structure and many training parameters are difficult to determine.In the feature extraction part,this method uses a combination of HOG features and grid features.Through experimental verification,this method shortens the time required for network training and has high recognition accuracy.