基于压缩感知的红外图像与可见光图像融合

发布时间：2018-04-25 13:03

本文选题：压缩感知 + 图像融合　；参考：《长春工业大学》2016年硕士论文

【摘要】：图像融合技术是近几年的图像研究领域的热门,而红外与可见光的图像融合则更是在军事、遥感、安全和视频监控等领域得到了广泛的应用。红外传感器对热目标能很好地识别而可见光传感器能提供清晰的场景信息,通过对红外与可见光进行融合,可以同时利用目标的热量信息和场景信息,对目标进行更为精确的判断,有利于目标识别及其他后续的处理。传统的图像融合由于需要源图像的全部信息,不易存储和传输,融合效率低。压缩感知(compressive sensing,CS)能以较低采样率高效重构原信号,成为图像融合领域的研究热点。压缩感知在采样的同时对数据进行压缩,而且融合只对少量的测量值进行而不是整幅图像,可以显著降低系统的存储压力与计算复杂度。围绕压缩感知融合,本文主要做了以下工作:(1)给出了压缩感知的基础理论框架和结构,对基于压缩感知的现有融合规则进行了深入研究,并总结了常用的融合图像质量客观评价指标。(2)目前基于压缩感知融合的研究集中在DCT,小波变换,NSCT(非下采样轮廓波)稀疏变换上。而本文将非下采样剪切波变换(NSST)应用在压缩感知域,并提出新的压缩感知融合方案,仅对计算量较大的高频子带系数采用基于压缩感知的图像融合方法进行融合,即对高频子带进行星型测量,使测量值与结构信息相关联;对测量后的高频子带制定了基于空间频率加权的新融合规则。利用区域标准差与区域能量联合指导低频子带系数的融合。最后经NSST逆变换得到融合图像。实验结果表明,仅用单层NSST变换即可重建出高质量图像,融合效果好于其他几种传统的压缩感知融合算法。(3)在基于非下采样剪切波变换(NSST)压缩感知融合算法的基础上,对基于图像分块的压缩感知融合算法进行了深入研究。考虑到对图像分块采样,图像块之间缺乏整体特性的缺点,致使图像在经过重构之后会产生分块效应。在分块压缩感知融合算法中引入平滑投影(SPL)Landweber算法进行重构,有效去除了分块效应,并提高了融合速度。
[Abstract]:Image fusion technology is a hot topic in the field of image research in recent years, and the image fusion of infrared and visible light has been widely used in military, remote sensing, security and video surveillance. Infrared sensor can recognize thermal target well, while visible light sensor can provide clear scene information. By fusion of infrared and visible light, heat information and scene information of target can be used simultaneously. A more accurate judgment of the target is beneficial to target recognition and other subsequent processing. Because the traditional image fusion needs all the information of the source image, it is difficult to store and transmit, so the fusion efficiency is low. Compressed sensing CSS can efficiently reconstruct the original signal at a low sampling rate, which has become a research hotspot in the field of image fusion. Compression sensing can compress the data while sampling and fuse only a small number of measurements instead of the whole image which can significantly reduce the storage pressure and computational complexity of the system. In this paper, the basic theoretical framework and structure of compressed sensing are given, and the existing fusion rules based on compressed perception are deeply studied. This paper also summarizes the objective evaluation index of image quality in common use. Currently, the research based on compressed perceptual fusion is focused on DCT, wavelet transform and NSCT (non-downsampling contour wave) sparse transformation. In this paper, the non-downsampling shear wave transform (NSST) is applied to the compressed sensing domain, and a new compression sensing fusion scheme is proposed. Only the high-frequency subband coefficients with a large amount of computation are fused by the image fusion method based on compressed sensing. The high frequency subbands are measured by star pattern to correlate the measured values with the structure information, and a new fusion rule based on spatial frequency weighting is established for the high frequency subbands. The combination of regional standard deviation and regional energy is used to guide the fusion of low frequency subband coefficients. Finally, the fusion image is obtained by inverse NSST transform. The experimental results show that high quality images can be reconstructed by using single layer NSST transform only, and the fusion effect is better than that of other traditional compression sensing fusion algorithms. The compressed perceptual fusion algorithm based on image segmentation is studied in detail. Considering the shortcoming of image block sampling, there is a lack of overall characteristics between the image blocks, which leads to the blocking effect of the image after reconstruction. The smooth projection splitter Landweber algorithm is introduced into the block compression perceptual fusion algorithm for reconstruction, which effectively eliminates the blocking effect and improves the fusion speed.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】