基于分块LBP的树种识别研究

Wood recognition based on block local binary pattern(LBP)

本文基于获取便捷的木材体视图进行木材树种自动识别研究。首先对木材图像进行标准化预处理,然后采用分块LBP提取特征,分别使用欧式、卡方、衰减3种不同的距离进行分类,最后采用最近邻进行识别。讨论了木材图像特征分块方式对识别结果的影响,并比较了在不同距离下的识别效果。结果表明:采取不同的分块方式对最终的分类影响较大,其中沿着年轮线方向上的分块呈现下降趋势,而在垂直年轮线方向上进行适当的分块可以提高分类的识别率;采用卡方距离可以得到最好的识别率,最高可达93.3%,比欧式距离高出2.5%。

The automatic wood recognition is studied in this paper through wood stereogram for its convenient way to obtain. Firstly, a standardizing preprocess of wood images was carried out. Secondly, block local binary pattern （LBP） was selected to extract wood features and three different distances （European distance, Chi-square distance, and diffusion distance） were introduced to classification experiments. At last, we used nearest neighbor classifier to identify wood features, discussed effects of block LBP features on recognition results and compared recognition rates in different distances. Results show that different block ways have significant influence on the final classification, among which block along the tree ring direction shows downward trend and proper block will improve the recognition rates in the vertical direction of tree ring. Chi-square distance can obtain the best recognition rate, up to 93.3% , 2.5% higher than that of European distance.