一种改进的压缩感知图像融合方法

西安科技大学 通信与信息工程学院,陕西 西安 710054

图像融合; 非下采样contourlet变化; 压缩感知; 标准差; 自适应

An improved compressive sensing image fusion method
ZHANG Ye,LIU Xiao-pei

(College of Communication and Information Engineering,Xi'an University of Science and Technology,Xi'an 710054,China)

image fusion; NSCT; compressive sensing; standard deviation; adaptive

DOI: 10.13800/j.cnki.xakjdxxb.2018.0425

备注

传统基于压缩感知的图像融合算法在通过压缩感知观测图像高频分量时会丢失分量的空间信息,仅能采用简单规则进行融合,导致融合图像纹理细节等信息效果较差。针对此,文中提出了一种结合图像高频特征的自适应融合规则。首先,对融合图像进行非下采样contourlet变换(NSCT),分解后得到的低频子带系数采用区域能量融合规则,较传统低频系数处理更好的保留源图像的背景信息; 然后,由于高频子带系数具有较高稀疏性,因此通过压缩感知进行压缩后根据标准差特征自适应选择融合规则; 最后,对重构系数进行非下采样contourlet逆变换。实验结果表明:与传统经典算法相比,新方法不仅精准提取到红外目标,同时充分保留可见光图像的细节信息,兼顾了待融合图像的背景信息和红外目标信息,有效提高了融合效果和主观感受。

The traditional image fusion algorithm based on compressive sensing(CS)can only adopt the simple fusion rule when the high frequency component spatial information is lost by CS observation,which leads to poor information such as detail of texture.In this paper,an adaptive fusion rule combining high frequency feature of image is proposed.First,making a NSCT decomposition to the infrared image and visible light image,and after that to fuse them using regional energy criterion for the low-frequency sub-bands to obtain a better fusion result than the traditional low frequency coefficients.Next,Since the high-frequency sub-band coefficients have high sparsity,they are compressed by CS and adopted adaptive fusion rule according to the standard deviation feature.Finally,inverse NSCT transform is carried out for the reconstructed coefficients.Compared with the traditional compressive sensing method,the new method not only accurately extract the infrared target,while retaining the full details of visible light images and containing the background information and infrared target information,which effectively improves the image fusion effect and subjective feelings.