形态学平整运算尺度空间在图像融合中的应用

摘要

In the multi-scale image transforms, scale-spaces which are built on linear operators such as gaussian filter, have the same important drawback; edges in the image are blurred and the contours of objects are distorted. Mathematical morphological levelings ( MML) , as powerful operators that possess a number of desired properties for the construction of nonlinear scale-space, can overcome this drawback. Considering the advantages of MML, this paper applies the MML scale-space in image fusion, in order to improve the visual effect of fused image. Firstly, we analyze different methods that compose the MML operators, and different methods of generating marker image, choose proper method from them. Secondly, suitable fusion rules are carefully chosen for different kinds of image fusion. Finally, observing that the MML can generate rather fine detail images, enhancement factors are added to improve the effect of fused image. Experimental results show that our method can better preserve the nonlinear features in the fused image, such as sharp edges in multi-focus image fusion, and high intensity objects in visible/infrared image fusion. Thus, it performs better than popular linear scale-space based image fusion methods, such as atrous wavelets and Nonsubsampled Contourlet Transform (NSCT).%在图像的多尺度变换方法中,由高斯滤波器等线性算子构成的尺度空间,不能很好地刻画图像中的非线性特征,且具有共同的缺点:在多尺度分解过程中图像中的边缘会被模糊而较难定位,且图中物体的轮廓会被扭曲.而形态学平整运算是一种能在多尺度分解的同时保持物体边缘位置不变的形态学滤波器,由其构成的非线性尺度空间能有效克服这一缺点.本文提出了一种基于形态学平整运算的尺度空间的图像融合方法,从融合的三个步骤分别进行分析:分解时,对分解算子和标志图像的生成方法对融合的影响进行分析并从中选择最佳的方法；融合时,对不同类型图像融合,选择最适合的融合规则；重构时,引入增强因子,进一步提高图像融合效果.实验结果证明本文方法能比基于线性尺度空间的方法更好地保护图像中的非线性特征,如边缘和亮度,从而取得更好的融合质量.