Research Of Paper Defects Image Processing Algorithm Based On Fuzzy RBF Fusion

Paper appearance disease is an important factor which affects the quality ofpaper. Quick and accurate identifying paper diseases during paper productionline and taking measures in time are greatly significant to paper manufacturers.With the development of paper technology, paper transfer rate is faster, and papersheet is wider on paper production line. Traditional method that using humaneyes to detect paper defects has been gradually replaced by online paper defectsdetection system based on machine vision technology. Nowadays machine visiontechnology is widely used and constantly develops. But current paper defectsdetection methods have two radical problems. Firstly, every current method canonly identify one or few defects. Secondly, current methods can hardly detectcomplex paper defects accurately, such as fold and crack. In view of theseproblems, for five representative paper defects, i.e.,black, holes, bright spots,fold, crack, on the basis of existing paper disease detection technology, a newalgorithm that combining information fusion to detect paper diseases is proposedin this paper. This paper mainly studies image preprocessing method and paperdefects identification algorithm based on fuzzy RBF fusion, main work is asfollows.1）In terms of preprocessing algorithms, research begins from imagefiltering. By comparing mean filtering and median filtering in different sizetemplates, improved median filtering is chosen as paper image denoisingalgorithm. After comparing paper defects edge processing results of differentedge operators, a “3*3”Sobel operator edge detection is selected to obtain edgeinformation of paper diseases from background. For futher extracting paperdefects eigenvalue, next step is to achieve and sign minimum enclosing rectangleof every paper defect area, and then to segment target paper disease area byprogressive-scanning the image after edge detection and filling original paperimage. 2）Here Cyclone IV of Altera is chosen as hardware platform, FPGA is usedto implement image preprocessing. As paper preprocessing results show,processing results of both are almost same, and the processing speed of FPGAhardware implemention is faster, can realize real-time paper preprocessing whenpaper image data rate is60frames per second.3）In terms of paper defects identification algorithm, summarize differentkinds of paper defects identification methods, and analyze the advantages,disadvantages and complementarity of neural network and fuzzy logic system,fuzzy RBF fusion is used to conduct feature layer fusion with some paper defectcharacteristic values, aiming to identify more paper defects more accurately. Thisfuzzy RBF fusion has the advantage of simple structure and rapidity.4）On basis of analysis of gray, geometric characteristics and texturalfeatures of five paper defects, mean gray value, gray variance, fractal boxdimension, fold template matching degree are chosen as four input values offuzzy RBF fusion. Sample paper defects image are used to train fuzzy RBFfusion, and then verify its identification effect. Simulation experiments haveshown that the fusion presented is practicable to identify five paper defectsaccurately.After all, this paper defects detection system presented based on FPGA andfuzzy RBF fusion can not only speed up paper preprocessing speed, achievereal-time paper defect detection, reduce the data volume of host computer, butalso identify five paper defects accurately. It has a good application prospect.