基于非抽样形态小波与视觉显著计算的图像融合的研究

This project focuses on the research of visual and infrared image fusion based on undecimated morphological wavelets and visual saliency computation. It presents a method for constructing real-time undecimated morphological wavelets with the purpose to solve the problem of shift-variant property in wavelets; proposes a nonseparable undecimated morphological wavelet based on S-transform for a smooth image without the loss of edge detail; proposes a method for extracting saliency objects adaptable to various image categories, while maximizing the saliency information. A framework of image fusion is proposed based on undecimated morphological wavelet and visual saliency computation. Visual saliency computation is combined with pulse coupled neural networks(PCNN) by improving the linking strength in the PCNN model through the object's saliency. Meanwhile, visual saliency computation is combined with variant PDE to construct the energy function of variant PDE and the optimal solution is found. Using fuzzy neural network, three metrics, edge preservation, regional structural similarity and noticeable difference are combined to construct an evaluation system, which is able to reflect the visual perception characteristics, with the purpose to bring the objective evaluation into correspondence with subjective perception. Finally, an experimental platform is set up for subjective and objective evaluations. Achievements of this project can improve the visual effects and quantitative metrics of the fused image, and thus can be widely applied in military and civilian fields.

本项目以可见光与红外图像融合为研究对象，基于非抽样形态小波和视觉显著计算进行图像融合的研究。提出一种具有实时性的非抽样形态小波的构造方法，克服了形态小波的移变性；构造一种不可分离S-变换的非抽样形态小波，兼顾图像平滑和边缘保持特性；提出一种适应不同图像类别的视觉显著目标提取方法，最大限度地保留图像中的显著信息；构造基于非抽样形态小波和视觉显著计算的图像融合框架，结合视觉显著计算和PCNN，用目标的显著度构造PCNN模型中的链接强度；结合视觉显著计算和变分PDE，构造变分PDE的能量函数并求最优解；引入模糊神经网络，选择边缘保持度，区域结构相似度和可感知差异等三个指标构建反映视觉感知特性的评价指标体系，使客观评价与主观感知尽量保持一致；搭建实验平台，从主客观两个方面进行图像融合质量评价的实验研究。本项目的研究成果能够改善图像融合的视觉效果和量化指标，在军事和民用领域均有重要的应用价值。

本项目以可见光与红外图像融合为研究对象，基于非抽样形态小波和视觉显著计算进行图像融合的研究。用非线性的形态学滤波器替换非抽样小波变换中的线性滤波器，提出系统构造非抽样形态小波的方法并应用于图像融合。在这种统一框架下，构造一种形态非抽样小波(Dilate-Erode Morphological Undecimated Wavelet，DE-MUDW)，采用形态学膨胀和腐蚀的平均作为分析算子，特点是简单高效，高频信息具有较好的物理特性；构造了一种高帽变换的非抽样形态小波，应用该方法得到的融合图像较好地保持了源图像的轮廓；构造了不可分离S-变换非抽样形态小波方法，兼顾平滑和边缘保持特性；将DE-MUDW扩展为多方向三通道分解的非抽样形态小波方法，利用多方向结构算子提升融合质量。.构造基于非抽样形态小波和视觉显著计算的图像融合框架，结合视觉显著计算和PCNN，用目标的显著度构造PCNN模型中的链接强度；结合视觉显著计算和变分PDE，构造变分PDE的能量函数并求最优解；把视觉注意机制引入图像融合质量评价体系中，使客观评价与主观感知尽量保持一致；搭建实验平台，从主客观两个方面进行图像融合质量评价的实验研究。本项目的研究成果能够改善图像融合的视觉效果和量化指标，在军事和民用领域均有重要的应用价值。