基于 PCA 特征融合与 LDA 分类的实木地板纹理判别方法

纹理一致性影响着实木地板档次，针对目前实木地板纹理分类速度慢、精度低的问题，提出一种适合区分直纹、抛物纹、乱纹3类纹理的在线检测方法。方法首先对纹理图像进行缩小，运用视觉心理学的Tamura方法提取粗糙度、对比度、方向度、线性度、规整度、粗略度等6个纹理特征；同时在原图像提取反映图像全局信息的灰度均值、方差、熵等3个统计量；然后，运用主成分分析法（PcA）对3类纹理9个特征进行降维融合操作；最后，采用线性判别分析方法（LDA）构建3类纹理的辨识模型。采用200幅实木地板纹理图像进行实验，当主成分个数为7时，分类正确率稳定达到85％，较传统Tamura方法的83％和全局基本统计量的70％有所提高；特征提取时间为0．5548S，比缩小前图像的Tamura特征提取时间55．7000S明显减低，而分类正确率没有明显变化。

Wood Board Texture Judgement Based on PCA Feature Fusion and LDA Classification

The main factor to improve the grade of solid wood products is the consistency of texture. To increase the rate and improve the ac- curacy in the online classification of solid wood boards, a new classification way was put forward which can identify three types of wood tex- ture, including straight grains,parabolic and disorder grains. First, the gray- scale image of wood texture was adjusted to a smaller one, six texture features were extracted with Tamura of visual psychology, which are roughness, contrast, orientation, linearity, regularity, rough de- gree. At the same time, three basic statistics （gray mean, variance, entropy） which can reflect the global information were extracted from the original gray - scale image. Second, PCA method was used to fuse the nine wood texture features of three kinds of wood texture. At last, LDA method was adopted to construct cognition model. 200 images of wood floor texture were used to conduct experiment. The experiment showed that ： when used seven principle components, the classification of accuracy is 85% which is improved, compared with 82.5% of the traditional method of Tamura and 70% of basic statistics. The classification time is reduced from 55.70 s to 0.55 s. The classification accuracy is not changed.