Research On Identification Methods Of Wood Surface Defects Based On Texture Features

The detection technique of wood surface defects based on computer vision technology and pattern recognition theory has many advantages, such as non-breakage, rapidity, accuracy, economical efficiency and so on. It plays an important role to rank the lumber classes automatically, to improve the commodity value of sawtimber and to accelerate the automation of wood processing.In this paper, taking wormhole, dead knot and live knot three kinds of typical wood defects as research objects, an intensive study of the pattern recognition methods to wood surface defects are made. The main contents of the research are including the image segmentation of wood surface defects, the evaluation of segmentation performance, feature extraction and the identification of defects.Image segmentation is the first step. And it is also the essential step from image processing to image analysis. In view of the deficiency of traditional gray level thresholding and edge detection, two-dimensional thresholding technique based on-gray level-gradient co- occurrence matrix and the maximum entropy principle is adopted to segment the wood defect images. Because wood defect is natural texture stuff, combined with Fuzzy C-Means cluster algorithm, texture segmentation method based on gray level co-occurrence matrix is presented. And simultaneously using morphology as a tool which has strong operation function processes the segmented images. After post-processing, it strengthened the invisibility and integrity of the segmented images and enhanced the precision of defect extraction.According to the gray system theory, a model to evaluate the performance of image segmentation based on the analysis of grey incidence is proposed. The integrative appraisal of the segmentation performance of the improved two-dimensional thresholding algorithm and the texture segmentation algorithm based on gray level co-occurrence matrix was conducted. The judgment results of this model can basically keep in step with the algorithm performance.To identify the wood defects, the defects are described from two aspects, the texture features (14 gray level co-occurrence matrix parameters)and geometrical characteristics (elongation and degree of rectangle). Acording to the distrabution of these parameters, the parameters which have small standard deviation are selected as the input eigenvector of the classifiers. And the features are extracted by the principal components analysis which can reduce the texture feature dimensions and eliminate the relevance between feature modes and highlight their difference to satisfy the input request of the recognition level. Using BP NN classifier and the improved K-Near Neighbor classifier to identify the defects, the correct rates of pattern recognition achieved 92% and 88% respectively. The experiment result proves that according to the digital image processing technology, it is an effective way to solve the segmentation and identification of wood surface defects by texture features of wood surface defect images. And the software identification experiment system of wood surface defects has been developed to facilitate the users.