【作者】 王业琴；

【导师】 王克奇；

【作者基本信息】 东北林业大学 ， 控制理论与控制工程， 2006， 硕士

【摘要】 木材表面颜色是反映木材表面视觉和心理感觉的重要特征,直接关系到木制品以及室内环境的质量评定,因此对木材表面颜色进行表征和分析具有一定的研究价值。本文以柞木、红松、水曲柳、落叶松和白桦五种东北常见树种为研究对象,利用计算机视觉技术分别对其径切、弦切木材表面颜色进行定量分析,并按照木材表面颜色特征对木材进行自动分类。 本文介绍了彩色图像处理中常用颜色模型及其特点,对获取的木材图像完成预处理过程,包括直方图均衡化,去除噪声,真彩色增强等。对预处理后的木材图像分别用颜色直方图法、颜色矩法提取了木材表面的颜色特征。另外我们对基于R、G、B分量融合的彩色特征矩阵提取了三阶矩参数作为木材表面颜色特征。通过对每组特征参数的分析,从类间均值、方差的分布,来预测参数的可分性情况。 介绍了模式识别技术以及神经网络分类器的特点,由于BP神经网络具有强大的非线性映射能力,因此本文设计了BP神经网络分类器。分类实验结果表明:颜色直方图特征分类识别率较低,尤其对于弦切样本颜色分类的特征参数是无效的,分析原因主要是颜色空间量化过程中丢失了颜色信息造成的;颜色矩特征对于径切、弦切样本都取得了较高的分类识别率;基于融合的彩色特征分类时,径切样本分类识别率与颜色矩特征的分类识别率相同,降低了特征矢量的维数,加快了运算速度,因此融合的特征参数对径切木材表面颜色特征具有较好的描述性,但是融合的彩色特征弦切样本颜色分类效果不是很理想;最后通过三组参数综合性能的比较,确定RGB颜色矩参数为木材表面颜色分类的最佳参数。 本文的研究成果能够实现木材表面颜色的自动分类,提高生产加工中木材表面颜色分类的自动化水平和木材利用率。为木材学提供先进的研究手段并丰富计算机视觉领域关于颜色分析分类的方法。更多还原

【Abstract】 Wood surface color is an important feather to reflect the visual and psychological feelings of wood surface, and it is directly related to the quality evaluation of wood products and environmental indoors, so the research of presentation and analysis on wood surface color is valuable. In this paper five tree species which are common in northeast are researched, such as Pinus koraiensis, Larix gmelinii, and Quercus Mongolic. The quantitative analysis of wood surface color is carried on by using computer vision technology and the wood is automatic classified according to the feathers of wood surface color.The common color models and their features in the process of color images are introduced in this article, and the preprocess of acquired wood images is finished, such as noise removal, color enhancement and so on. The wood surface color characters of wood images after preprocessing are abstracted with the methods of color histogram and color matrix. Moreover the 3-rank moment parameters, as the wood surface color feathers, are abstracted from color characteristic matrix based on the R, G and B color characteristic integration. Through the analyses of each kind of feature parameters, the classification is forecasted from the distribution of a distance category and mean square.The characteristics of pattern-recognition and neural network classification are introduced in this article. Because the BP neural network has powerful nonlinear shine capacity, the BP neural network classification device is designed. Classification experiments shows that the identification rate of color histograms classification is lower;especially feature parameters for color classification of tangential samples are invalid. The main reason is the loss of color information in the course of color space quantization. The identification rate of classification with color moment feathers is higher for radial and tangential samples. When classified basing on the integrated features, the correct rate of classification of radial samples is the same as that of color moment feathers, and the vector-dimension is lower, so the processing speed is accelerated. For this reason, the integrated features can describe the wood surface color characters of radial wood very well, but they are invalid for tangential samples. Finally comparing the all round performances of three sets parameters, the RGB color moment parameters are the best parameters for wood surface color classification.The results of this research can achieve automatic classification of wood surface color, and improve the automatic level of classification of wood surface color in production processing and wood utilizations. It provides advanced research tools for wood and enriches classification methods of color analysis on the field of computer vision.更多还原