2D经验模态分解与非下采样方向滤波器组的红外与可见光图像融合算法

Infrared and visible image fusion algorithm based on 2D empirical mode decomposition and non-subsampled directional filter banks

针对当前红外(IR)与可见光(VI)图像融合中细节保留能力不足及目标配准精度不高的问题,设计了一种多尺度2D经验模态分解耦合非下采样方向滤波器组(NSDFB)的红外与可见光图像融合算法。分别计算红外与可见光图像的熵值,并比较二者阈值的大小,计算阈值较大图像的残差。通过2D经验模态分解(2D-EMD)和NSDFB机制,构建了多尺度方向分解模型,将熵值较大图像的残差和熵值较小的图像变换为高频方向系数与低频系数,以获得源图像的细节和特征信息。对于低频系数,引入加权平均作为低频系数的融合准则;根据区域能量对比度与清晰度来定义融合规则,完成高频系数的融合。利用2D-EMD多尺度分解逆变换将获取的低频与高频系数生成新图像。实验表明:与当前常用红外与可见光图像融合对比,所提算法具有更高的融合质量,所输出的图像具有更好的对比度与丰富的细节信息。

Aiming at the problem of insufficient detail preservation and low registration accuracy in the fusion of infrared(IR) and visible(VI) images at present, a multi-scale 2 D empirical mode decomposition and(Non-subsampled directional filter banks, NSDFB) was designed for the fusion of IR and VI images. The entropy values of infrared and visible images were calculated to compare the thresholds, and the residual values of images with larger thresholds were calculated. A multi-scale directional decomposition model of 2 D-EMD was constructed by means of 2 D empirical mode decomposition(2 D-EMD) and NSDFB, Which is used to transform the residual error of image with larger entropy value and the smaller entropy value into the high frequency direction coefficient and the low frequency coefficient, which can effectively obtain the details and feature information of the source image. For low frequency coefficients, the weighted average is introduced as the fusion criterion for low frequency coefficients, and the high frequency coefficients were fused by comparing the regional energy contrast with the definition scheme. The 2 D-EMD multi-scale decomposition is used to transform the low frequency and high frequency coefficients to generate new images. The experiment shows that the proposed algorithm has a higher fusion quality compared with the current infrared and visible image fusion, and the output image has a better contrast and rich details.