The Research On Classification Of Wood Surface Texture Based On Multi-resolution Fractal Dimension

The wood surface texture has the complex structure. It is extremely difficult to carry onthe quantitative analysis and the description with the ordinary. Only depending on the artificialdirect-viewing to evaluate the wood surface texture's characteristic. It will result the losing ofobjective and the usability. Therefore, the solution of wood surface texture research has thevital significance. This subject is a technological unit of the natural science fund inHeilongjiang projects "The research of wood surface visual characteristics"(C2004-03).This paper has gathered massive wood surface texture picture, and pretreated. To beginwith the fractal and the wavelet theory, it has selected the wavelet base and the waveletdecomposition layer according to various wavelets quality and the error of the reconstruction.After that, it separately studied the wood surface texture fractal dimension on the condition ofsingle rate of distinguish and the multi-rate of distinguish. The research indicated that, it willhave a better effect to differentiate each kind of different wood surface texture characteristicbased on the multi-rate of distinguish than single rate of distinguish. It carried on 2 levels ofwavelets of the decomposition and reconstruction to the wood face texture picture, and soughtto the fractal dimension on the multi-rate of distinguish of 10 kinds of altogether 1,000 woodsamples.Taking the rate of recognition of the Nearest Neighbor Classifier as the price function, Itchoosed the parameter of fractal dimension on the condition of the multi-rate of distinguish forSimulated Annealing Algorithm.And defined characteristic vectors to input the Classifier.Usingthe Nearest Neighbor Classifier and the K-close neighbor Classifier, it separately makeclassification to the wood surface texture, and studied influence of the noise to the parameter offractal dimension and the result of Classifier.The experimental result indicated that, inmostsuperior combination of the parameter of the normal condition, the rate of recognition ofthe Nearest Neighbor Classifier and K-close neighbor Classifier achieved higher precision. Butthe rate of recognition of two Classifiers obviously had droped in noise environment.And therate of recognition had enhanced arter Midian Filter filtered.