基于RPCA模型的红外与可见光图像融合技术研究

发布时间：2018-05-12 07:36

本文选题：图像融合 + 非下采样Contourlet变换　；参考：《南昌航空大学》2017年硕士论文

【摘要】：随着传感器成像质量的不断提高,如何利用红外与可见光图像融合技术增强图像质量与清晰度逐渐成为图像处理与机器视觉研究领域的热点问题。可见光传感器成像符合人眼观察,含有丰富的细节信息,但是容易受到天气影响,不能全天候工作。红外传感器成像稳定,能够很好地显示隐藏的目标,受照明条件和恶劣天气的影响较小,但是所得红外图像对比度较低,目标细节的反映能力比较差。因此将红外与可见光图像融合,可弥补两者的不足,发挥各自的优势,使得融合图像同时具有红外与可见光图像的优点。近年来,红外与可见光图像融合技术已经取得较大进展,但融合图像失真、纹理细节信息缺失、目标显著性等问题仍是图像融合研究领域尚未完全解决的重点与难点。针对上述问题,本文提出基于RPCA分解模型的红外与可见光图像融合方法,主要工作如下:1.针对自然场景下的红外与可见光图像精确配准问题,本文首先对红外与可将光图像融合的过程、层次、常用方法以及融合规则进行概述;然后对图像预处理过程进行介绍;最后对图像预处理过程中的图像配准技术进行阐述,并通过实验与分析验证本文所选用红外与可见光图像配准方法的精度与稳定性。为后续红外与可见光图像融合的预处理提供可靠基础。2.针对红外与可见光图像的特征描述问题,本文在鲁棒主成分分析的基础上,通过对红外与可见光图像进行RPCA分解,并对分解后所得稀疏矩阵图与低秩矩阵图所包含的源图像特征信息进行分析,提出红外与可见光图像的RPCA分解模型。3.针对传统基于非下采样Contourlet变换的图像融合方法易出现融合图像失真、纹理细节信息缺失的问题,本文在RPCA分解模型的基础上,提出基于RPCA分解模型的NSCT域红外与可见光图像融合方法。首先对红外与可见光图像进行RPCA分解,得到相应的稀疏矩阵;然后利用NSCT变换将红外与可见光图像进行分解,得到源图像的低频子带和高频方向子带;对于低频子带,采用基于稀疏矩阵的融合规则进行融合;对于高频方向子带,最高层方向子带采用基于稀疏矩阵的绝对值取大法进行融合,其它层则采用基于PCNN的方法进行融合;最后对融合后的低频子带与高频方向子带进行NSCT逆变换,从而获得最终的融合图像。4.分别对标准图库与真实场景图像集使用Contourlet变换融合方法、双重NSCT融合方法、NSCT-小波-PCNN融合方法以及本文方法进行测试实验。实验结果表明,相对于其它几种融合方法,本文方法不仅能够突显红外图像中的目标人物信息,而且将可见光图像中的纹理以及细节信息表现的更为细腻,同时能够减少融合图像失真。并对融合后图像使用图像互信息、平均梯度、标准差、峰值性噪比、结构相似度指数这五项客观评价指标进行评价,评价结果表明本文方法取得较为优异的表现。
[Abstract]:With the continuous improvement of sensor imaging quality, how to improve image quality and clarity using infrared and visible image fusion technology has gradually become a hot issue in the field of image processing and machine vision. Visible light sensor imaging is consistent with human eye observation and contains rich details, but it is vulnerable to weather and can not work all the time. The infrared sensor imaging is stable and can display the hidden target very well. It is less affected by the illumination condition and the bad weather, but the contrast of the infrared image is low, and the target detail ability is poor. Therefore, the fusion of infrared and visible images can make up for the shortcomings of the two images, give play to their respective advantages, so that the fusion images have the advantages of infrared and visible images at the same time. In recent years, infrared and visible image fusion technology has made great progress, but the fusion image distortion, texture details missing, target significance and other problems are still not fully resolved in the field of image fusion. Aiming at the above problems, this paper proposes an infrared and visible image fusion method based on RPCA decomposition model. The main work is as follows: 1. Aiming at the problem of accurate registration of infrared and visible images in natural scene, this paper firstly summarizes the process, level, common methods and fusion rules of infrared and visible images, and then introduces the process of image preprocessing. Finally, the image registration technology in the process of image preprocessing is described, and the accuracy and stability of the infrared and visible image registration methods selected in this paper are verified by experiments and analysis. To provide a reliable basis for the subsequent infrared and visible image fusion preprocessing. 2. Based on the robust principal component analysis (PCA), the infrared and visible images are decomposed by RPCA, aiming at the feature description of infrared and visible images. The characteristic information of the source image contained in the sparse matrix graph and the low-rank matrix graph is analyzed, and the RPCA decomposition model of infrared and visible image is proposed. In view of the problem that the traditional image fusion method based on non-downsampling Contourlet transform is prone to the distortion of fusion image and the lack of texture detail information, this paper is based on the RPCA decomposition model. An infrared and visible image fusion method in NSCT domain based on RPCA decomposition model is proposed. First, the infrared and visible images are decomposed by RPCA, and the corresponding sparse matrix is obtained. Then the infrared and visible images are decomposed by NSCT transform to obtain the low frequency subbands and the high frequency directional subbands of the source image. The fusion rules based on sparse matrix are adopted. For the high frequency directional subband, the maximum direction subband is fused by the absolute value based on sparse matrix, and the other layers are fused by PCNN method. Finally, the NSCT inverse transformation of the low frequency subband and the high frequency direction subband is carried out to obtain the final fusion image. 4. The Contourlet transform fusion method, the dual NSCT fusion method and the NSCT- wavelet-PCNN fusion method are used to test the standard image library and the real scene image set. The experimental results show that compared with other fusion methods, this method can not only highlight the target information in infrared images, but also make the texture and detail information in visible images more delicate. At the same time, it can reduce the distortion of fusion image. The five objective evaluation indexes such as mutual information, average gradient, standard deviation, peak noise ratio and structural similarity index are evaluated. The evaluation results show that the proposed method achieves excellent performance.
【学位授予单位】：南昌航空大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】