节点文献
基于NSCT和改进型PCNN的多源图像融合算法研究
【作者】 王波；
【导师】 周冬明；
【作者基本信息】 云南大学 ， 通信与信息系统（专业学位）， 2015， 硕士
【摘要】 在当今飞速发展的图像融合技术领域中,由于图像融合技术在各行各业都得到了不同程度的应用,因此该技术受到了人们越来越多的关注。而作为这项技术领域的重要组成部分——多源图像融合技术,因为其能在不要求单个传感器更多性能的基础上较大的提升系统的整体性能,并且在图像识别、图像特征提取等方面展现出的良好性能,所以作为图像融合的分支之一多源图像融合技术也同样引起了很多的研究者关注。多源图像融合技术是指将从多传感器接收的图像信息整合成人们需要的目的图像,这种处理方法可以获取更加可靠、更加清晰、更加适合人类视觉感知的融合图像。本文主要先是对多源图像融合技术的基础理论、基本概念进行了研究,在研究现有的一些融合算法上更进一步的研究如何提升多聚焦图像融合质量的问题。本文研究的主要内容如下：1、系统的介绍了本文内容所涉及到的多源图像融合理论、图像融合效果的评价体系、PCNN以及NSCT等的基本原理和结构。2、传统的PCNN算法中是把图像灰度值直接作为神经元的激励,这不能很好分辨物体的细节部分,而空间频率是基于单位像素的变换率,它能够很好的符合人眼对图像的边缘细节敏感这一特点,因此本文选取局部领域内的空间频率当作神经元的输入激励,通过实验验证该方法取得不错的融合效果。3、由于在传统的多尺度分解与重构图像融合方法中,会存在重影等现象,这导致实际得到的融合图像效果并不十分的清晰,针对这一现象本文在深入研究了具有平移不变性特点的NSCT变换后,提出了基于NSCT的多聚焦融合方法,并经过一组实验证明了本文选取的方法在主观和客观上均优于传统的算法。4、对图像进行多尺度分解后的低频分量含有图像的大部分内容信息,而高频分量又代表着图像的细节信息,因此针对这一特性,本文在低频部分的处理采用区域统计的方法,而高频部分采用能够很好描述图像边缘细节信息的SF,并将其值作为PCNN的激励输入。实验结果表明,该方法不仅在人眼的视觉效果上有不错的清晰度,同时其客观评价的标准也有明显的优势。
【Abstract】 In the field of image fusion technology,image fusion due to its wide application, so it got more and more applications. And in numerous branch of the image fusion technology, Multi-source image fusion technology is the source image fusion of multiple sensors to receive objective image, because it can improve the overall performance of the system, but does not require a single sensor’s more performance, And it showed good performance in the aspects such as image recognition, image feature extraction, So the multi-source image fusion technology has attracted more and more attention of researchers. In the multi-source image fusion technology, in order to obtain the fusion image more reliable,more clear and more suitable for human visual perception. In this paper, the basic theory of multi-source image fusion technology, the basic concept is studied,on the basis of some existing multi-source image fusion algorithm,the Further study is how to improve the image’s quality on the more focus image fusion problem.This paper studies the main content as follows:1、Detailed introduces the content of this article involves the multi-source image fusion、Image fusion effect evaluation system, the basic principle and structure about the PCNN and NSCT.2、In the traditional PCNN algorithm, image grey value is directly as a neuron stimulation, but this method does not well distinguish the details of the object, spatial frequency is a transformation rate on the basis of unit pixels, It can very good satisfy the human eye is sensitive to the edge of the image details,therefore this article selects the spatial frequency in the field of local as neurons input.3、because of in traditional multi-scale decomposition and reconstruction of image fusion method, There is the phenomenon such as ghosts, This leads to the actual effect of the fusion image is not very clear, In reaction to the phenomenon in this paper, Based on the further study of the translation invariance characteristic of NSCT after transformation, this paper proposed the More focus on image fusion method based on the NSCT. The superiority of the method is verified by the experiments.4、the low frequency component after image multi-scale decomposition contains most of the image information,and high frequency component represents the image details, according to this feature, In this paper, the processing of low frequency part adopts the method of regional statistics, the processing of high frequency part adopts the method of the value of SF as excitation input of PCNN. The experimental results show that, This method not only has good clarity on the visual effect of the human eye, at the same time, the objective evaluation standard also has obvious advantages.