Paper Properties Research Based On Computer Vision

Paper can be seen everywhere in our daily life, with the improvement of people’s living standard, the requirement for the quality of the paper keeps on increasing. The quality of the paper includes appearance properties and mechanical properties. The traditional paper appearance quality detection is mainly via eye observation, and mechanical properties of paper mainly rely on artificial testing with the aid of all kinds of instruments.In order to realize the purpose of the paper test automation. In this thesis, we carried on related research on paper appearance properties by using computer vision and choosing appropriate light source. First, image segmentation, edge detection, and the hough transformation were used to detect paper disease such as dark spots, bright spot and folding, which laid the foundation for subsequent paper mechanical properties testing. Then, texture feature was extracted from paper image to describe appearance properties of paper by using LBP and gray gradient co-occurrence matrix. The mechanical properties of paper were classified as category labels, BP neural network and SVM were used to classify paper mechanical properties, average accuracy rate of the model was more than 95%, which proved that the paper appearance properties and mechanical properties were related. Final, multiple regression analysis was carried out on the paper mechanical properties such as ensile index, burst index, folding resistance and tear index and paper appearance properties. The regression model between mechanical properties and appearance properties of paper was established. After testing, the average relative error of model was within 10%, which proved that the model was correct.This thesis provides a new way for paper mechanical properties of automated non-destructive test by using computer vision methold.