基于滚动导向滤波的可见光与红外图像融合嵌入式系统研究

Research on Embedded System for Visible and Infrared Image Fusion Based on Rolling Guided Filter

在不利的环境条件下,红外图像往往会出现暗部纹理信息缺失不全、主体轮廓模糊等问题,现已提出了多种相关的算法来解决这些问题,获得了极大的成功。但这些算法大多停留于软件仿真,本文研究了一种基于滚动导向滤波的嵌入式的图像增强算法并对其融合方法进行一定的改进:首先,使用滚动导向滤波器对红外图像和可见光图像进行多尺度分解得到一系列Gauss金字塔,再对每层Gauss金字塔分解出一张Laplace金字塔,得到红外和可见光图像的高频部分和低频部分,之后对低频部分和高频部分使用Gauss分布权重矩阵进行加权平均处理,根据图像灰度级生成一个可视化权重图,生成基层融合图像,之后计算两张图像高频部分的差异矩阵,接着将差异矩阵与两张图像的高频部分进行加权平均得到一个中间图像,使用该中间图像和两张原始图像的高频部分构建一个线性的差异矩阵,根据矩阵得到一个线性特征的方程组,通过求解该方程组得到待融合图像的细节层,实现图像的细节边缘性特征保存,最后进行逆变换从而得到融合图像。之后通过一系列图像质量评价的方式:STD、AVG、SSIM、FD,证明了该方法的可行性。最后探索了基于该方法的FPGA实现方式,在嵌入式平台实现了仿真,满足了图像融合的要求。

Under adverse environmental conditions,infrared images often have problems such as incomplete dark texture information and blurred subject contours,and a variety of related algorithms have been proposed to solve these problems and have achieved great success.However,most of these algorithms stay in software simulation,and this paper studies an embedded image enhancement algorithm based on rolling guide filtering and makes certain improvements to its fusion method:first,a series of Gauss pyramids are obtained by multi-scale decomposition of infrared images and visible light images using rolling guide filters,and then a Laplace pyramid is decomposed for each layer of Gauss pyramids to obtain the high-frequency and low-frequency parts of infrared and visible images.Then,the low-frequency part and the high-frequency part are weighted and averaged using the Gauss distribution weight matrix,a visual weight map is generated according to the grayscale level of the image,and the underlying fusion image is generated,and then the difference matrix of the high-frequency part of the two images is calculated,and then the difference matrix is weighted and averaged with the high-frequency part of the two images to obtain an intermediate image,and a linear difference matrix is constructed using the intermediate image and the high-frequency part of the two original images,and a system of equations with linear features is obtained according to the matrix.By solving the equation system,the detail layer of the image to be fused is obtained,the detail edge feature of the image is preserved,and finally the inverse transformation is carried out to obtain the fused image.Then,through a series of image quality evaluation methods:STD,AVG,SSIM,FD,the feasibility of the method is proved.Finally,the FPGA implementation method based on this method is explored,and simulation is realized on the embedded platform,which meets the requirements of image fusion.