Research On Image Segmentation Based Paper Defect Detection Methods

In the papermaking process, paper defections affect the evaluation of paperquality and the benefits of paper mills seriously. So, identifying and locatingpaper defection on line in papermaking process is of great significance. Atpresent, web defection system based on CCD（Charge Coupled Device）technology has been used in the market, however, there are still somedeficiencies on the algorithm. In order to satisfy the paper machines which arerunning rapidly nowadays, a simple and efficient processing algorithm of paperdisease images must be found through exploring and researching. This thesisbuilds an experiment device of paper defection system to solve the problems ofpaper images by simulatimg real-time paper machine. The image processingalgorithms are researched emphatically.The main work is as follows:1）The advantages and disadvantages of image processing algorithms areanalyzed and researched. The results of images are analyzed by introducingpreprocessing algorithm and segmenting algorithm, and the suitable algorithmsare found out for paper images by contrasting the results of image processing.2）The segmenting algorithm of paper image is studied.The basic steps ofpaper image detection are summarized by introducing detection system. In orderto process the paper image with two or subsize flaws, an improved iterativemethod and a dynamical double threshold segmentation algorithm are proposed.For the low contrast paper defection like folds, a Canny edge detectingalgorithm based on dynamical threshold is put forward. The results show thatthe improved algorithms could separate paper diseases from backgroundcorrectly.3）The design of whole processing algorithm for paper image is conducted.Firstly, images are processed by a3×3template median filter. Secondly, theflaw areas are segemented by the improved segmentation algorithm. Finally, the features of target images are extracted and classified.4）The fuzzy pattern recognition algorithm is applied and researched in theclassification of paper defections.Therefore, an idea of two-stage classifier isproposed. The representative features are selected and extracted by analysingtypical paper defections such as specks, holes, folds, slits, and bright spots.Aiming at the characteristics of the image after segmentation, The defections areclassified coarsely and then segmented precisely. In the part of second category,the corresponding membership functions and fuzzy rules are established. At last,paper defection could be classified rapidly by the Mamdani fuzzy inferencesystem.