Tropical Wood Species Recognition System Based on Gabor Filter as Image Multiplier

The main problem in wood species recognition system is the lack of discriminative features of the texture images. Some of the wood species have similar patterns with others and some have different patterns even though they are of the same species. Moreover, the growth rings for tropical wood changes slightly due seasonal changes in climate. One of the ways to improve the system is by providing more features representation of each species. In this work, Gabor filter is proposed to generate multiple processed images from a single image so that more features can be extracted and trained by the neural network. After the raw image has been sharpened and contrast enhancement has been applied at the preprocessing stage, the image will be convolved with Gabor filters. The output of the convolution generates Gabor images which are images extracted based on frequency and spatial information of the original images. These Gabor images will be used by grey level co-occurrence matrix (GLCM) for feature extraction. A multi-layer neural network based on popular back-propagation (MLBP) algorithm is used for classification. The result shows that increasing the number of features by means of Gabor filters as well as the right combination of Gabor filters increases the accuracy rate of the system.