Recognition Of Paper Defect Image With Complex Background Noise Based On SVD And SVM

Paper appearance defect is an important factor which affects the quality of paper. With the increasing demand for paper making speed, traditional method using human eyes to detect paper defects has been unrealistic. Therefore, developing new intelligent online detection system combined with computer technology to accurately identify paper defect is the first problem.In paper making production line, paper defect images with complex background noise result in the existing recognition methods in practical application is not able to correctly identify paper defect.With the image feature of black, holes, bright spots, folds and crack five common and typical paper defects as the research object and on the basis of the existing paper defect detection system, a new paper defect identification method is proposed. The main work is as follows:1) Image preprocessing based on FPGA. To reduce the amount of processed data in the paper defect extraction and recognition phase, and improve recognition accuracy, image information obtained by CCD camera is needed to be preprocessed by FPGA. Improved median filter is selected to remove noise by comparing noise filtering effect of mean and median filter under different templates and analyzing intensity difference histogram.2) Built of noise model. Obvious unevenness and serious image background noise exist in paper defect background image. In order to quantitatively analyze the impact caused by paper defect image unevenness, temporal noise and spatial noise, using nature modal separation method(EMD), transverse gray-scale distribution trend function of background image is established to remove its uneven ingredients. Then mathematical model of time and space noise is established by AR model.3) Paper defect recognition algorithm research based on SVD and SVM. Through the analysis of the model parameters, we know that the power spectrum of paper defect image is uneven and time changeable and traditional filtering technology can’t only filter out noise very well, but also easy to destroy the paper defect information. So a new paper defect identification method is proposed: Firstly two-dimensional wavelet is used to remove background noise and paper defect information in the low frequency band is retained. Then SVD is employed to extract paper defect characteristics and SVM is used to recognize paper defect.With noise elimated by three layers two-dimensional wavelet, feature extracted by SVD and paper defect type classified by SVM, the experimental results show the effectiveness of the method. It can effectively identify all kinds of paper, and the identification is above 98%. At the same time, paper defect identification accuracy and speed are developed by using FPGA to lock the small area of paper defect.