成像光谱重建及纹理色差客观评估研究

【作者】 蔡普庆；

【导师】 沈会良；

【作者基本信息】 浙江大学 ， 电磁场与微波技术， 2008， 硕士

【摘要】 多光谱成像是获得和显示精确颜色信息的重要技术,原因之一是多光谱图像包含了更多的光谱信息,原因之二是多光谱成像技术很好地克服了同色异谱现象。本文从光谱反射率重建这一成像系统的基本点着手,研究了反射率重建的技术,并针对成像系统聚焦模糊的现象提出了适当的模糊图像恢复技术,最后采用纹理分析方法初步研究了纹理各特性对视觉色差的影响。维纳估计重建光谱反射率是多光谱成像中最常用的技术,本文在未知样本光谱反射率的前提下,通过自由选择适当的训练样本来求得维纳估计方法中的协方差矩阵,从而达到重建高精度反射率的目的。最后重建结果还同传统的维纳估计在不同信噪比和不同成像通道数情况下进行比较,得出在系统低信噪比或成像通道少于7的情况下,本方法求得的光谱反射率重建图像时,光谱误差和色度误差都较小。鉴于多光谱成像中不同波段滤光片焦距的差异导致图像模糊的情况,需要采用合适的图像处理方法来恢复图像质量。本文提出对传统Unsharp Mask滤波的高频成分做加权因子修正,使图像在增强细节的同时减少噪声影响,并根据基于人眼视觉特性的质量描述参数Q值来确定最佳参数,从而实现聚焦模糊图像的恢复。且与其它方法相比,经本文所提出方法恢复后的图像在整体上具有更高的清晰度,也与理想清晰图像更为接近。纹理这一对颜色有着重要影响的因素已成为颜色质量评价和色差估计技术中关注的焦点,本文通过实验也给出了初步的定性和定量的分析结果。定量分析中通过提取多光谱纹理图像的特征统计变量,分析其对明度方向上的视觉色差的影响,并且在分析过程中注意某些统计量之间的相关性,通过SAS软件分析减少相关统计量,以便精简最后的多变量线性回归模型。更多还原

【Abstract】 Multispectral imaging is of great importance in color acquisition and reproduction. One reason is that multispectral images contain more spectral information; the other reason is that multispectral images can effectively eliminate the phenomenon of metamerism. This thesis presents a new method for spectral reflectance reproduction in multispectral imaging system, and introduces an image restoration method to solve the out-of-focus image blurring. We also conduct preliminary investigation on the texture effect in visual color difference evaluation.In multispectral imaging, Wiener estimation is widely adopted for the reconstruction of spectral reflectance. We propose an improved reflectance reconstruction method by adaptively selecting training samples for the autocorrelation matrix calculation in Wiener estimation, without a prior knowledge of the spectral information of the samples being imaged. The performance of the proposed adaptive Wiener estimation and the traditional method are compared in the cases of different channel numbers and noise levels. Experimental results show that the proposed method outperforms the traditional method in terms of both spectral and colorimetric prediction errors when the imaging channel number is 7 or less.Because of the different focus of the filters in different channels, out-of-focus blurring always occurs in the channels other than the one whose focus is appropriately adjusted, and therefore image restoration techniques are required to eliminate or reduce the blurring effect. We propose such an image restoration method by modify the strength parameter function of the traditional Unsharp Mask filter for the sake of both image sharpening and noise removal. The optimal parameters are decided according to a visual evaluation term called Q-value. Compared with other techniques, the proposed method can produce perceptually more clear images, which is also closer to the ideally focused one.Surface texture is becoming the most interesting factor in color quality assessment and color difference evaluation areas. This thesis gives some preliminary research results through certain analysis. During the quantitative assessment process, we acquire some statistical variances of texture features through the light direction, then pays attention to the correlation between statistical variances in the hope of eliminating some variances, simplifying the final linear regression model.更多还原