Study And Application Of Paper Counting Method Based On Machine Vision

The quantity statistics of paper and card is important for the automation of printing and packaging industry.At present,paper counting method based on machine vision is mostly adopting the way of image segmentation in accordance with the global characteristics of paper image or the way of edge detection to achieve paper number.Although the method is simple,there are still some more higher requirements for the thickness and the edge quality for the paper.At the same time,the shortcomings is existing in the aspect of difficulty of segmentation threshold selecting and slow speed of counting.To solve the problems above,this paper is studying a two-stage counting method based on feature of spatial level and the gray level for roughness and accuracy of paper counting based on spatial and gray feature.The improved curve peak location algorithm is adopted to complete the rough paper counting and the feature clustering method is used to complete accurate paper counting.The paper counting system needs to achieve after completing the study of methods.The main contents and conclusions of paper are as follows:Considering the basic periodic variation of the whole image and the distribution of spatial gray having peak-valley feature,the paper counting method based on peak-valley feature is designed.Firstly,we can obtain the projection curve by using gray projection method.Then,we can overcome the shortcomings of traditional detection algorithm according to the feature of curve.The detection method is improved by using the way of second-searching regional extreme value.According to the number of statistical regional minimum points to calculate the amount of paper and detect the position.When this method achieves good results in the aspect of paper image,as the same time the high accuracy will be still obtained while the paper color and paper edge strips is changing.To solve the problem of the counting accuracy is reduced as paper adhesion in paper arrangement,while can not detect the paper number in adhesion area using rough paper counting method.Taking this intoconsideration,we can design classification method according to the anomalous changing feature of paper width and paper gray in adhesion area.Firstly,region growing method is used to get binarization image.Statistical pixel method is used to get paper width and calculate the gray value difference of projection curve trough point and the projection curve mean value;Then calculate the average gray difference of projection curve between and average width of the paper.Constitutes two-dimensional feature vector with the paper width and gray difference;Finally,use K-Means clustering method to classify the feature vector and decide the paper number in adhesion area.The experiment results show that this method can solve the adhesion problems and improve the accuracy of paper counting under certain conditions.Realize the paper counting system based on machine vision and use for actual test at last.The results show that the counting accuracy achieve more than 99.7 % and count time less than 1s based on the conditions of paper thickness is 0.07mm-0.09 mm,scanner resolution is not above 1800 dpi,and the angle between paper and scanner horizontal is 45 degrees.