基于视觉底层特征和多尺度分析的图像融合研究

发布时间：2018-03-25 06:00

本文选题：图像融合　切入点：视觉底层特征　出处：《昆明理工大学》2017年硕士论文

【摘要】：单图像传感器获取的图像信息有限,很难对场景进行精确的描述。图像融合技术就是为解决这个问题而发展起来的。图像融合技术通过对图像信息进行处理和计算的方式将多个图像传感器采集的同一场景的图像信息综合起来得到一幅包含更多有用信息的高质量图像,新的图像更利于人们进行后期处理。近年来,研究者们提出了很多图像融合算法,并且成功运用在计算机视觉,机器人,数码成像,自动目标识别技术,军事等领域。本文为提升图像融合性能展开了一系列研究,研究成果如下:(1)在红外与可见光图像融合研究中,针对在变换域融合规则设计中只考虑了图像的单一特征的问题,本文在非下采样轮廓波域内设计融合规则时充分考虑了图像的局部结构、对比度信息和亮度信息,并针对高低频子带系数分别提出了新的有效的清晰度度量算子,调整不同特征的重要性权重。(2)在多聚焦图像融合研究中,针对空间域聚焦特性检测算法中聚焦信息提取不完整的问题,提出了一种多尺度图像分析技术与非局部均值滤波相结合的融合算法。算法中为综合不同尺度图像的优势,提出一种新的多尺度图像分析技术,同时引入双边滤波器的思想设计新的图像聚焦特性度量指标来识别多尺度分析结果中的聚焦像素。为减轻空间域方法产生的边界效应,最后利用块的一致性验证和非局部均值滤波器对决策图进行优化,得到源图像的融合权重图像。(3)在医学图像融合研究中,针对传统的稀疏表示训练字典算法耗时,训练的字典表达能力有限的问题,提出基于视觉底层特征筛选的字典训练算法。根据医学图像的特点,提出利用反映图像块亮度信息的能量特征和反映图像块边缘信息变化的空间频率特征等视觉底层特征作为度量标准,筛选信息丰富的图像块来构建训练集,并通过KSVD算法分别训练细节信息字典和亮度信息字典,并基于训练的字典设计医学图像融合框架。
[Abstract]:The image information obtained by single image sensor is limited, It is very difficult to accurately describe the scene. Image fusion technology is developed to solve this problem. Image fusion technology through the processing and calculation of image information to capture the same image sensors. When the image information of a scene is synthesized, a high quality image containing more useful information can be obtained. In recent years, researchers have proposed many image fusion algorithms, which have been successfully used in computer vision, robot, digital imaging, automatic target recognition technology. Military and other fields. In this paper, a series of studies are carried out to improve the performance of image fusion. The research results are as follows: 1) in the research of infrared and visible image fusion, only a single feature of the image is considered in the design of fusion rules in transform domain. In this paper, the local structure, contrast information and luminance information of the image are fully considered when designing fusion rules in the non-downsampling profilometry domain, and a new effective sharpness metric operator is proposed for the high and low frequency subband coefficients, respectively. Adjusting the importance weight of different features.) in the research of multi-focus image fusion, focusing information extraction in spatial domain focusing feature detection algorithm is incomplete. A fusion algorithm combining multi-scale image analysis and non-local mean filtering is proposed. In order to synthesize the advantages of different scale images, a new multi-scale image analysis technique is proposed. At the same time, the idea of two-sided filter is introduced to design a new measure of image focusing characteristics to identify the focus pixels in the results of multi-scale analysis. Finally, using block consistency verification and non-local mean filter to optimize the decision graph, we get the fusion weight image of source image. In the medical image fusion research, the traditional sparse representation training dictionary algorithm is time-consuming. In this paper, a dictionary training algorithm based on visual bottom feature screening is proposed. Based on the energy feature of image block luminance information and spatial frequency feature reflecting the change of image block edge information, this paper proposes to select the image block with abundant information to construct the training set. The detail information dictionary and brightness information dictionary are trained by KSVD algorithm, and the medical image fusion framework is designed based on the training dictionary.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】