基于奇异值分解的木材细胞切片识别算法

摘要

For recognizing wood cell section images, dimension reduction should be made to extract the major char-acteristic data.Since the similar the cell images the closer eigen values of the norms, small images could hardly be identified.For this reason, an improved cell section identification method based on RealAdaBoost +SVD-n was suggested.Firstly, singular values of cell section image was calculated and it was arranged from big to small. Then the top n singular values was removed, and the remains was used as image characteristics for testing by Real-AdaBoost .The results showed that this method could not only recognize the wood cell section images with great differences but also could well distinguish similar cell section images.This recognize algorithm showed good robust-ness .%木材细胞切片识别需要对图片进行降维处理, 提取最大特征数据来进行识别. 相似度高的切片由于它们的最大特征的范数相近, 所以很难区分. 针对这种情况, 提出了一种基于RealAdaBoost+SVD-n的细胞切片识别算法, 该算法首先计算切片图像的奇异值, 将奇异值按从大到小的顺序排列, 然后去掉前n项较大的值, 使用剩余项作为图片特征, 利用RealAdaBoost分类器进行训练和测试. 实验结果表明, 该方法不仅能够识别差异较大的木材切片, 而且还能够很好地区分较为相似的细胞切片, 该识别算法具有很好的鲁棒性.