视觉检测系统中颜色测量关键技术研究

发布时间：2018-03-27 11:09

本文选题：视觉检测　切入点：颜色测量　出处：《北京交通大学》2017年硕士论文

【摘要】：基于机器视觉系统实现印刷品质量在线检测已成为一种趋势,其中颜色是印刷品质量的重要评价指标,但机器视觉检测系统通过数字相机仅能获得待测目标在RGB颜色空间下的图像,同一待测目标在不同成像环境下的RGB值并不相同。在印刷品颜色质量的检测过程中,需要利用颜色测量技术才能得到待测目标真实的颜色值。本文利用物体表面的光谱特性实现了颜色测量。同时基于对颜色测量色差可控的要求,本文提出了设备无关的颜色空间样本设计方法限定颜色估计误差,解决了机器视觉检测系统中的色差控制难题。本文提出了一种基于R矩阵的光谱反射率重建算法,并根据不同颜色空间之间的映射关系,实现了颜色测量。该方法在十二通道线扫面CCD成像系统下,对主成分分析法与R矩阵法进行理论分析与验证,提出了利用主成分分析法对R矩阵法进行改进,实现光谱反射率重建的方法。然后将光谱信息转换到色度颜色空间中,从光谱反射率的重建精度以及颜色色差两方面对颜色测量的精度进行评价。实验结果表明,改进后的R矩阵法能够实现颜色在色度与光谱两者上的高精度匹配,可用于实际的工业印刷品颜色质量检测。本文提出了基于CIE 1976 L**b*颜色空间均匀分割的样本设计方法。该方法将CIE 1976 L*a*b*颜色空间变换为以颜色的明度、饱和度以及色调表示的极坐标LCH形式,采用等明度、等饱和度、等色调的分割方式,对CIE 1976 L*a*b*颜色空间进行均匀取颜色点,将颜色空间的颜色估计误差限定在均匀的样点之间。为了检验训练样本对颜色测量模型建立的精度影响,利用BP神经网络法以及支持向量机法建立颜色空间转换模型,并对随机生成的测试样本进行测试,实验结果表明,基于CIE 1976 L**b*颜色空间均匀分割的训练样本,能够控制颜色测量结果的精度,提高了颜色测量结果的有效性。
[Abstract]:It has become a trend to realize print quality on-line detection based on machine vision system, in which color is an important evaluation index of print quality. However, the machine vision detection system can only obtain the image of the target under the RGB color space through the digital camera, and the RGB value of the same target under different imaging environment is different. In order to obtain the true color value of the target to be measured, the color measurement is realized by using the spectral characteristics of the object surface. At the same time, based on the requirement of controllable color difference measurement, the real color value of the target can be obtained by using the technique of color measurement. In this paper, a design method of device independent color space samples is proposed to limit the error of color estimation, which solves the problem of color difference control in machine vision detection system. In this paper, a spectral reflectivity reconstruction algorithm based on R matrix is proposed. According to the mapping relationship between different color spaces, the color measurement is realized. The principal component analysis method and the R matrix method are theoretically analyzed and verified in the 12-channel linear scanning CCD imaging system. A method of spectral reflectivity reconstruction is proposed by using principal component analysis (PCA) to improve the R matrix method, and then the spectral information is transformed into chrominance color space. The accuracy of color measurement is evaluated in terms of spectral reflectivity reconstruction accuracy and color difference. The experimental results show that the improved R-matrix method can achieve high accuracy matching between chrominance and spectrum. This paper presents a sample design method based on uniform segmentation of CIE 1976 Lnb * color space. This method transforms CIE 1976 Lnab * color space into the brightness of color. In the polar coordinate LCH form of saturation and hue representation, using the method of equal brightness, equal saturation and equal hue, the color space of CIE 1976 is evenly selected. In order to test the effect of training samples on the accuracy of color measurement model, the color space conversion model is established by BP neural network method and support vector machine method. The experimental results show that the training samples based on uniform segmentation of CIE 1976 Lnb * color space can control the accuracy of color measurement results and improve the effectiveness of color measurement results.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】