Research And Application Of Optical Character Recognition System Based On Improved Drop Fall Segmentation Algorithm

With the improvement of people's living standards,food quality problems are increasingly concerned by the public.Spray code as the most critical product quality tracking information has been widely used in all walks of life.Laser spray code equipment is the most common methods for marking on the PET bottles.Spray code quality detection,as an important part of product appearance inspection,at present mainly adopts the artificial light inspection and sampling inspection.However,the manual detection is easy to fatigue and the error rate is very high especially when the speed of production is very fast.For these reasons,we need to find stable and efficient quality detection method.Based on practical production,deep researches are carried out on the positioning and segmentation of laser spray code in this paper.The existing spray code detection systems are all designed for ink spray code,which deformation is small and has few adhesion phenomenon.Meanwhile,laser spray code is easily adhered like handwriting characters,which means it is difficult to locate and segment.Combining morphology clustering and drop-fall segmentation,this paper has developed an entire laser spray code detection system,even if there is many bottle scratch and adhesion interference.The paper's work is summarized as follows:?In the preprocessing stage,blackhat operation is used to extract the character contour and remove most bottle scratch.The rest bottle scratch is remove by using the proximity information analysis algorithm.Then,the code is located successfully.? K-means algorithm is used to separate the upper and lower two lines of codes.In order to improve the stability of K-means,the projection method is used to calculate the initial center of clustering.? Projection method and drop-fall algorithm,which is improved from several aspects,are used to segment the adjacency codes.This paper proposes an improved drop-fall segmentation algorithm based on Harris corner information,path score and recognition feedback.This algorithm uses the local projection extreme point to determine the starting position,and the Harris corner information is used to guide the dropping direction at the groove.Finally,the score evaluation system is introduced,in which the score of path and recognition are taken into account.Experiments and actual production have demonstrated that this improved algorithm can effectively separate the adherent laser spray codes.? In the character recognition part of the system,based on the characteristics of the code,this paper selects the regional structural characteristics of the codes to train and identify using SVM which has achieved good recognition rate.