摘要
为了对木材进行表面纹理分类,首先确定纹理的灰度共生矩阵描述参数、灰度共生矩阵的生成像素间距和灰度级数;求取分析了200个木材样本的纹理参数并输入给竞争神经网络进行分类验证.实验表明:1)以“角二阶矩”、“对比度”、“相关”、“熵”、“方差”、“逆差矩”作为描述木材纹理的特征参数是合适的.2)在比例为1∶1的512×512木材图像情况下,生成灰度共生矩阵的最佳像素间距为4,最佳图像灰度级数为128.3)木材纹理图像灰度共生矩阵的"角二阶矩"、"相关"和"熵"值最大的方向为纹理方向.4)竞争神经网络的分类正确率为88%.研究结论:按上述规则生成的6个灰度共生矩阵参数对描述木材表面纹理特征是有效的,据此对木材表面纹理分类是可行的.
To classify wood by surface texture, wood texture parameters of Gray Level Co -occurrence Matrix (GLCM) were first selected by relevance analysis. The building rules of GLCM ( the best gray levels and pixels interval) were confirmed by experiment. GLCM texture parameters of 300 wood specimens were calculated and assayed, and then placed into a Competitive Neural Network for classification output. The results are as follows: 1) GLCM parameters “Angular Second Moment contrast , correlation , entropy , sum of squares”and “inverse difference moment”are suitable to describe the wood texture, 2) the best pixels interval when building GLCM is 4 when the wood image size is 512×512 and proportion is 1: 1, and 128 levels image gray can best reflect the wood texture information, 3) the texture direction is exactly the direction of the maximum value of‘ Angular Second Moment', ‘Correlation'and ‘ntropy'of the GLCM of the wood texture image, and 4) the correct rate of classification of the Competitive Neural Network classifier is 88 percent. The six parameters of GLCM building rules mentioned above are valid to describe wood texture feature and it is workable to classify wood by surface texture according the six GLCM parameters.
出处
《哈尔滨工业大学学报》
EI
CAS
CSCD
北大核心
2005年第12期1667-1670,共4页
Journal of Harbin Institute of Technology
基金
哈尔滨市自然科学基金资助项目(2004AFXXJ020)
黑龙江省自然科学基金资助项目(C2004-03)
关键词
木材纹理
灰度共生矩阵
特征参数
竞争神经网络
wood texture
gray level co- occurrence matrix
feature parameter
Competitive Neural Networks