Wood species classification based on local edge distributions

The wood species classification system is developed in order to give fair service to the wood industry. An image is characterized mostly by the low level features such as color, shape and texture. Among them texture features play a major role in characterizing the image and the statistical features are found to be efficient for classification. For a better classification result, wood image first has to be preprocessed in order to get meaningful feature extraction. Among various preprocessing techniques, Contrast Limited Adaptive Histogram Equalization is found to be an appropriate method for wood image enhancement. The edges in an image reveal spectral discontinuity, contrast, directional information, and structure pattern. In this paper, in addition to ten statistical features, saliency index is another significant parameter used for texture classification of different wood species. Edges are used as the main source of information to compute the saliency in the local window of every pixel. The computation of saliency index is simple and it is more powerful for recognizing the wood species. The back propagation neural network is used for classification of wood species. 100 number of wood images are used for training the neural network and 50 number of images are used for testing which are taken from prospect data base available for wood images. The percentage of classification accuracy in this work is more encouraging and the classification accuracy is 90%.