Pretreatment Of Banknote Testing And Improved Method Of Template Matching

Template matching is the commonest problem in machine vision,image and video processing technology.In banknote printing industry,machine vision gradually replaces artificial screening,becoming main means of par testing.Now there are many problems in detecting system of banknote printing industry.The frist,the matching template is too small to contain the more information.The matching template is 10px*10px in banknote printing industry.As everyone knows,the larger the template size,the longer the matching time.So In order to ensure high efficiency,the small template is used.In summary,efficiency of algorithm without template size is urgent required.The second,the template gets to image edge.Now template of industry is too small,in order to improve the amount of information contained in templates,edge locations are often selected.Because of overprint that is unavoidable in banknote industry,efficiency of algorithm without image edge is urgent required.The third,integral graph algorithm of Paul Viola overly tedious,there are more optimization space.In view of the above problems,this paper proposes a statistical method for template matching.The main achievements of this paper are as follows:1.In paper,pretreatment of banknote testing has been studied and improved method of template matching has been proposed,which is named statistical operator.Statistical operator is the feature vector that describe every pixel locals in section,the number of all section is the feature vector.Then,optimum matching position is found through comparison of template's feature vector and projection image's feature vector.After finding optimum matching position in discrete state,it using Taylor expansion formula to calculate optimum matching position in continuous state.Next,source image is moved in subpixel level by three linear interpolation.Finally,equalization processes the image,which eliminates the effect of brightness.Advantages of statistical operator is : high efficiency,300 times of the current method in template 100 pixel;template size independence;unaffected by brightness.The efficiency of template matching based on statistical method is independent of template size.Large scale templates containing more information can be used in matching,and templates at boundaries can also be avoided.2.Based on calculation method of integral graph proposed by Paul Viola,this paper proposes its improvement scheme.Calculation method of integral graph proposed by Paul Viola calculated by three integral value near target point.The time complexity is O(n2),which is a square order;involving two image memory,three variable memory and five times calculation.And there is initial condition,so it need determines whether initial conditions are satisfied in ever calculation.New project in this paper using the variable that is the sum of row value.The time complexity is square order O(n2);involved of tow image memory,three variable memory and three times calculation.Because the size of integral image has been Enlarged,no initial condition in this method.Although the two algorithms are consistent in the dimension of time complexity and spatial complexity,the number of calculations is reduced and the number of variable reads is reduced in improved integral graph,result in execution efficiency increased by 17.66%.