10
TROPICAL WOOD CLASSIFICATION BASED ON LBP-LIKE DESCRIPTOR AND NEAREST NEIGHBOR CLASSIFIER
Authors: Chao Xiaofei, Cai Cheng, Zhang Zhiyong, Nie Liangbing, Li Shuqin
Number of views: 423
This paper compares the discriminative ability of six LBP-like texture descriptors for tropical wood classification, LBP, uniform LBP, rotation invariant LBP, rotation invariant uniform LBP, covariance of LBP and covariance of LBP difference were considered. Experiments on wood image dataset with 54 wood species was carried out, experimental results show that these six descriptors combined with nearest neighbour classifier achieve recognition rate of 97.50%, 96.64%, 92.84%, 88.55%, 54.40% and 56.53% respectively. LBP is the best and one of the efficient wood texture descriptors among these six LBP-like descriptors. LBP8,8 is the best and most stable wood texture feature, the recognition rate of and are 97.84% and 97.41% respectively, the time to classify one image by them is 0.28 second and 0.08 second. Compared with existing wood image classification methods, the combination of LBP descriptor with nearest neighbour classifier is very simple, it does not need the feature selection and training process, and it achieves much better time efficiency and a slightly lower recognition rate than the existing algorithms.