Research On Classification Of Wood Texture Based On Feature Fusion

Along with social progress and development, human artistic qualities of wood textures are increasingly high requirements. Wood surface texture timber directly affect the level of strength and value in use; and beautiful patterns of wood or of wood products processing also have a decisive role. Therefore, the wood surface texture classification has important practical significance. This paper solve this problem, this paper wood surface texture can be divided according to the shape ruled, parabola and random pattern texture three textures, and considering the time and accuracy of two factors, on the wood texture recognition methods were studied.The characteristics of the wood surface texture extraction problem, this paper, the method of wavelet transform, wavelet decomposition after layer to determine the ratio of the sum of the top layer of the three high-frequency detail energy use symlets4wavelet basis of three types of samples two wavelet decomposition of the image, and the decomposed wavelet coefficients of the seven sub-graph processing, as the mean and standard deviation of wavelet transform top14feature15feature value for the entire image entropy. The experiment shows that the15feature extraction by the wavelet transform method can be used on wood texture analysis, and the computation time is short, but the method is not suitable herein the parabola texture and random texture. To overcome this shortcoming, the use the curvelet transform method to decompose the three types of sample image, for the second layer of the detail layer, extraction of the mean and standard deviation of the odd direction coefficient in the first direction and the second direction, a total of16characteristics as a feature of curvelet transform parameters. Although the method is to identify the effect of each type of wood grain has a good, but long computation time. The advantages and disadvantages of the two methods, the characteristics of the fusion method,16characteristic parameters obtained by the wavelet transform and15characteristic parameters obtained by the curvelet transform to the normalization processing, using genetic algorithms on the feature fusion from the14wood samples classified most effective feature screened31feature parameters.Finally, the wood samples14feature fusion characteristic parameters as sorting features, using a sample of100oak300contains three types of texture, of which50per class into the BP neural network classifier training network and BP network of150sample image classification validation. Associated with GLCM extracted, contrast, angular second moment, variance and mean and identification method in the correct classification rate and time comparing this method can improve the accuracy and improve efficiency.