基于引导滤波的分布式多波段图像融合算法

Distributed Multi Band Image Fusion Algorithm Based on Guided Filter

为进一步提高空基光电探测系统对复杂环境与目标的智能感知性能,采用多无人机平台组成分布式空基多波段融合探测系统进行多基线数字多波段测量,系统具备可见光、红外数据协同探测与多波段图像融合能力。为了结合红外与可见光多模态图像的互补信息,文中提出一种基于引导滤波的可见光与红外像素级异构图像融合算法。首先,采用Shearlet变换分别对两幅经过处理的相同视场的可见光和红外图像进行高低频特征分解;接着,通过引导滤波提取图像显著性特征和边缘特征;最后,计算融合的权重再对图像进行像素级融合。本算法在在Ubuntu18.04系统上采用Qt开发框架进行算法实现并在某分布式多机系统下得到验证,算法对于分辨率为640×512的红外与可见光多波段图像源融合时间为32 ms,满足实时图像融合需求,融合后的图像能突出目标的红外辐射特性并较好的保留背景细节信息,为后续目标智能识别提供了保障。

In order to improve the intelligent sensing performance and target recognition ability of airbased photoelectric detection system in complex environment,a distributed multi machine platform is used for visible and infrared data cooperative detection and multi band image fusion. In order to combine infrared and visible multimodal images,a pixel level heterogeneous image fusion algorithm based on guided filtering is proposed. Firstly,shearlet transform is used to decompose the high and low frequency features of two processed visible and infrared images with the same field of view. Secondly,salient features and edge features are extracted by guided filtering. Finally,the fusion weight is calculated,then the image is fused at pixel level. The algorithm is implemented on the Ubuntu 18. 04 system using Qt development framework,has been verified in a distributed multi machine system,The fusion time of 640 ×512 resolution infrared and visible image source is 35 ms,which meets the real-time image fusion requirements. The fused image highlights the infrared radiation characteristics of the target,and retains the background details,It provides a guarantee for target recognition.