A new Gabor based approach for wood recognition

Correct wood recognition has an important meaning in the rational use of wood resources. To complete this task automatically, based on wood stereogram images, we propose a new Gabor based wood recognition approach in this paper, which has been successfully applied in many pattern recognition fields for its robustness against local distortions. However, only a few approaches can make full use of the information in Gabor patterns. To obtain more information of Gabor feature for wood recognition, we first use a set of 40 Gabor patterns to represent a wood image, which consist of important information at different orientation and scales. Then, we apply the block-based feature extraction with more statistical features besides mean and standard deviation on these Gabor patterns to enhance the discriminative ability of our approach. Finally, we reduce the dimensionality of the proposed feature descriptor by using feature selection. Only a few features are selected to achieve both high recognition performance and computational efficiency. We evaluate our approach on the wood database in Zhejiang A & F University (ZAFU), which contains 24 wood species. Experimental results show that by adopting proper sub-block numbers and blocking schemes, our approach outperforms the most of state-of-the-art approaches.