邓巍, 丁为民. 基于EM算法的图像融合质量评价[J]. 农业工程学报, 2007, 23(5): 168-172.
    引用本文: 邓巍, 丁为民. 基于EM算法的图像融合质量评价[J]. 农业工程学报, 2007, 23(5): 168-172.
    Deng Wei, Ding Weimin. Evaluation of image fusion quality based on EM algorithm[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2007, 23(5): 168-172.
    Citation: Deng Wei, Ding Weimin. Evaluation of image fusion quality based on EM algorithm[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2007, 23(5): 168-172.

    基于EM算法的图像融合质量评价

    Evaluation of image fusion quality based on EM algorithm

    • 摘要: 为提出一种量化评价经图像融合所产生的结果图像的质量评价方法,采用一种混合瑞利(Rayleigh)概率密度函数(pdfs)对图像边缘强度直方图建模,混合模型中各项的参数和权重通过EM算法迭代估算得到。在建立的混合瑞利概率密度函数模型中,最小参数混合项对应图像的弱边缘,最大参数混合项对应图像的强边缘。因此,取最小参数项的标准方差作为噪声的标准方差估计,实现噪声盲估计;取最大参数项的标准方差作为图像模糊度的定量评价指标。通过估算混合瑞利pdfs模型中的参数变化可以评价图像质量。与其它图像质量评价方法相比,这种方法的最大优点是不需要知道图像构造等细节信息,不需要图像变换,只要有原始图像即可对其进行评价。而且对较小噪声也能较精确地估计。研究表明这个技术很强健,并对要评估的图像依赖很小。

       

      Abstract: The purpose of the research is to put forward a method for quantitatively evaluating the quality of a image obtained by fusing several images. The method of the research is to model the image edge intensity using a mixture Rayleigh probability density functions(pdfs). The parameters and weights of mixture terms in the mixture model can be obtained using the EM iterative algorithm. The term with the smallest parameter corresponds to the weak edges, or the low-frequency background fluctuation. The term with the largest parameter corresponds to the strong edges. Therefore, the smallest variance parameter is considered as the noise variance estimation. Thus the blind estimation of the noise can be realized. And the largest variance parameter can be used to monitor the blurring. The results and conclusions of the research are that the image quality can be evaluated by studying the change of parameters in the mixture model. Compared with other image quality evaluation methods, this technique only needs the images to be evaluated and does not use detailed information about the formation of the images, and need not transform the images. The approach can be employed to estimate the smaller noise. These are the advantages of the approach. The investigation shows that this technique is quite robust and has low dependency on the image under evaluation.

       

    /

    返回文章
    返回