Research On Pattern Recognition Of Wood Surface Color

Color is one of the important natural characters of wood,which affects the visual perception and evaluation of wood and wood products.In view of that there was no national standard or industry standard in wood industry now to define and describe the wood surface color,image processing technology and pattern recognition was used to study on the method to implement automatic classification of wood surface color.A set of parameters to describe the wood surface color and the method of pattern recognition to classify the wood based on these parameters were established.It would provide an important reference and theoretical basis for classifying wood products and decoration wood by wood surface color,and provided a solid basis of theory for wood color classification automatically.Combining with our research subject,the wood sample databases used for wood surface color analyzing and classifying were founded at first.With image processing technology preprocessing of the wood images in sample databases was made to eliminate the influence of noise.Three method of wood surface color expression was studied and three kinds of color feature were extracted correspondingly.The methods of feature selection and classification in the patter recognition system were also studied,and classification results of three different classfieres with the features before and after feature selection was discussed.With comprehensive consideration to feature dimension and classification results,a set of wood color parameters and the method of pattern recognition were established finally.In view of the complexity of wood color and its narrow distribution in color space,three kinds of color feature were proposed after the analysis of wood color expression:â‘ As HSV color space could reflect the perception and identification of human being to color,histogram statistical features of three independence variables Hue,Saturation and Value were proposed to express the information of wood color;â‘¡Color moments of Hue,Saturation and Value were used to express the information of wood color;â‘¢With the characteristics of high resolving power of L~*a~*b~* uniform color space,five color features,seven color difference features and seven standard deviation features of color difference were also proposed.Problems and method of feature selection in pattern recognition system were studied,and classification accuracy of Nearest Neighbor(NN) classifer was adopted as evaluation criterion. The theories of Simulated Annealing Algorithm(SA) and Genetic Algorithm(GA) were stated. Feature selections based on SA-classification-accuracy and GA-classification-accuracy were realized,and then the searching performance of the two optimization methods was compared.The theory of support vector machine(SVM) which was established based on statistical learning theory was stated.The classification performance of k-Nearest Neighbor(KNN),BP neural networks and SVM,as well as SVM with different kernel function,were discussed by experiments.For parameter selection of SVM and kernel functions,GA was used to parameter optimization,and got the parameter combination which could make a better classification performance for SVM.A set of wood color parameters and the method of pattern recognition were established based on theory analysis of pattern recognition and the classification result of three kinds of color feature.Automatic wood classification based on wood surface color was realized,which would provide an important reference for wood production and processing.It also provided an advanced method for wood science and riches the methods of color analysis and classification in image processing.