【作者】 戴天虹；

【导师】 王克奇；

【作者基本信息】 东北林业大学 ， 机械设计及理论， 2008， 博士

【摘要】 颜色分类存在于木材的生产、加工和应用的各个环节当中。对于工业化生产和应用,能够实现按颜色自动分类是非常重要并具有深远意义的。本研究采用计算机视觉与模式识别理论方法对木质板材进行了颜色分类,针对木材自身的特点,实现了木质板材颜色分类,对带有缺陷的板材进行分割。本文对以下工作进行了研究。基于图像处理方法,采用均匀颜色空间L~*a~*b~*的色差理论,提取5个颜色参数均值和7个色差参数均值以及7个色差参数的均方差,共计19个参数作为颜色特征,对五种树种十类板材进行了颜色分类的研究。研究了基于RGB颜色空间颜色矩颜色分类方法,提取RGB颜色空间R、G、B三个分量各三个颜色矩,再加两个整幅图像的均值和方差,共11个参数作为颜色的特征量。将不同树种(114种)运用自组织竞争神经网络对其进行粗分类,分成六类,再用BP神经网络进行确认。正确率在88%以上。研究了基于HSV颜色空间颜色矩颜色分类方法,提取HSV颜色空间的H、S、V三个矩阵分别提取一阶原点矩、二阶中心矩、三阶中心矩共9个特征参数作为颜色特征。将五种树种十类板材进行了颜色分类。研究了颜色直方图两种方法,一种方法是选取HSV颜色空间,选取颜色特征为HSV颜色空间的H,S,V分量,计算均值、方差、偏度、峰度、熵和能量,形成十八维颜色特征向量。分别对五种树种十类板材进行了颜色分类的实验。另一种是选取HSV颜色空间,先进行量化,形成72维直方图,建立标准模板,再与模板进行相似性测度,得到了72维特征向量,根据木材本身颜色的分布特点,去掉那些频数为0或频数很小的特征分量,就形成了30维颜色特征向量。分别对五种树种十类板材进行了颜色分类的实验。提出了主颜色思想,并给出主颜色提取方法。选取HSV颜色空间,先进行量化,形成72维直方图,建立不同的树种标准模板,再与模板进行相似性测度,得到了72维特征向量,从最大到最小提取频数,从而得到主颜色特征的参数。对五种树种十类板材进行了颜色分类的研究。尝试将颜色和纹理结合对木质板材进行颜色分类,主颜色特征为3个、5个、8个(bin),纹理特征为共生矩阵的6个参数,对五种树种十类板材进行了颜色分类的实验研究。构造了四大种分类器,遗传算法聚类分析、神经网络分类器、支持向量机分类器和k-近邻分类器。其中神经网络含有三种分类器,分别是BP、RBF和PNN分类器。对各个分类器的思想和原理加以阐述。并对上述方法进行了颜色分类实验。提出了基于颜色和数学形态学相结合的木质板材分割方法,建立不同的树种无缺陷标准模板,将带有缺陷木质板材与标准模板进行对应象素点色差,把所有的色差值求和取平均,得到所要求的阈值。将大于这个阈值的象素点提取出来,再进行数学形态学的运算,就可以得到木质板材缺陷分割结果了。提出了木质板材颜色分类的系统设计方案。并对系统组成、工作原理加以论证,介绍了其系统的硬件系统和软件系统组成部分。更多还原

【Abstract】 Color Classification exists in each link of the wood producing,processing and application. It is very important to the industrial production and application because it can achieve automatic classification by color,and have a further signification.Computer vision and pattern recognition method is used in this research to classify the color of wooden board.According to the characteristics of wood itself,classification of wooden board based on its color is achieved,and segmentation to the defective wood is also realized.The research contents in the study are as follows:Based on image processing theory,using the color difference of the uniform color space L~*a~*b~*,5 color parameter mean value,7 color difference parameter mean value and 7 mean square deviation of color difference parameter,altogether 19 parameters are extracted as the color features to research on color classification of 10 categories of 5 tree species.The method of wood color classification based on color moments in RGB color space is studied.Every color moments of R,G,B components in the RGB color space,and also the mean value and variance of the two whole images,altogether 11 parameters are extracted as the characteristic value of color.Using self-organizing competitive neural network to rough classification to different tree species(114 species),classified to 6 species,then confirmed by using BP neural network.The correct rate is above 88%.The method of wood color classification based on color moments in RGB color space is studied,three matrix H,S,V of HSV color space-respectively first-order origin moments, second-order center moments,third-order center moments altogether 9 characteristic parameters are extracted as the color featureThe method of wood color classification by using color comment in HSV color space is studied.mean,variance,skew ness the three matrixes difference extracting HSV color space H,S,V is extracted counts regulation 9 characteristics in total parameter is the color features. Ten sorts of wooden board of five wood species are classified by using their color comments features.Five species ten categories are classified by color.Two methods of color histogram are studied:one method is choosing HSV color space, choosing color feature of H,S,V component in HSV color space to calculate the mean value, variance,skewness,kurtosis,entropy and energy,forming 18 dimension color feature vectors. Then experiment of color classification is done to each 5 tree species 10 categories.The other method is choosing HSV color space,quantization first to form 72 dimension histogram,build standard template,then making similar measure with the template,and get the 72 dimension characteristic vector,according to the distribution feature of wood itself,remove the characteristic vector which the frequency is 0 or too small,then 30 dimension color characteristic vector is formed.Do the experiment to the five species ten categories.Main color idea is proposed,and extraction method of main color is given,choosing HSV color space,quantization first to form 72 dimension histogram,build standard template of different tree species,then making similar measure with the template,and get the 72 dimension characteristic vector,extract the frequency from big to small,in order to get the parameters of the main color feature.Research on color classification of five species ten categories is done.Try to combine the color and texture to classify the color of wooden boards,the main color feature is 3,5,8 bin,the texture parameter is co-occurrence matrix with 6 parameters, then do the experiment to the five species ten categories.Four classifiers are built:Genetic Algorithm cluster analysis,Neural Network classifier, support vector machines classifier and k-neighbor classifier,in which the Neural Network contains 3 classifiers,:BP,RBF,PNN.The ideas and principles of each classifier are described. Then use the method above to do the color classification experiment.The segmentation method of wooden board based on color and mathematical morphology is proposed.No defect standard template of different tree species is built,and then compared the color difference with defect and standard form board,and all the color difference takes an average,get the threshold that wanted.Extract the pixel point that greater than the threshold value,and do the calculation of mathematics morphology,so the defect segmentation effect of the wooden board can be get.In this paper,design scheme of classification system of wood surface color is established. Then the system components and principles are demonstrated,the hardware and software components of the system are introduced更多还原