Research On Target Image Localization Algorithm Based On Machine Learning

In recent years,intelligent life is one step closer to our life,accurate and fast image target detection and recognition is playing an increasingly important role.Bar code is a widely used commodity label in intelligent life.It is also a structured image,which is easy to detect and recognize.With the rapid development of the retail industry,the work of the scan code becomes more and more heavy,and the majority of the bar code detection in the daily life is semi manual,and the human resource is a great waste.In complex background,fast and accurate positioning of barcode location is the premise and foundation of accurate identification.it is also a hot issue of scientific research in recent years.In this paper,we have carried out a thorough research on the bar code positioning algorithm.Aiming at the existing problems of the existing code positioning algorithm,in the complex background,especially in the presence of multiple bar codes,and in the case of the occurrence of rotation and deformation cannot be accurately positioning.This paper proposed a method,which is based on Hough Transform and LBP feature combined with SVM to improve the positioning accuracy of bar code for bar code recognition.Methods and contents of the study include the following aspects:(1)In order to solve the problem of existing detection algorithm,this cannot do the precise positioning of the rotated bar code.This paper proposed a method,which is based on the linear distribution of the bar code line in the image.According to the characteristics of the distribution of the bar code line in the image,the angle of the distribution density is obtained by the 2D Hough,and rotated the bar code image by the angle.Experimental results show that this method has a small angle error,and can be adjusted correctly for most of the complex background.(2)If the image have many bar code or defect barcode,the existing detect algorithm cannot precise detect.This paper proposed a method that using LBP feature describes the bar code.The situation of the bar code in the image is complex,and the LBP feature can be used to solve the problem,which is in the defect bar code and many bar code situations.(3)In order to increasing the accuracy of the bar code positioning,consider the characteristics of bar code,which consist of black and white lines,the color histogram is introduced to describe the bar code and combined with LBP feature to extract the feature of bar code.The extracted data are trained to get the SVM classifier model.Then use the classifier,to detect the image of the bar code detection,after post-processing gets the results.Our experiment is achieved in the international general database(Westf?lischeWilhelms-Universit?t Muenster Barcode Database,WWU),and compared with experimental results of the traditional bar code positioning algorithm verify the effectiveness of the proposed algorithm.