基于多尺度方向引导滤波和卷积稀疏表示的红外与可见光图像融合

Fusion of Infrared and Visible Images Based on Multi-Scale Directional Guided Filter and Convolutional Sparse Representation

基于引导滤波和非下采样方向滤波器,提出了一种多尺度方向引导滤波图像融合方法,该方法兼具边缘保持特性和方向信息提取能力,能够有效提取源图像的有用信息。所提方法对源图像进行多尺度方向引导滤波,得到了包含低频近似部分和强边缘部分的低频分量,而后通过高斯低通滤波将其进行有效分离,分别应用基于卷积稀疏表示和区域能量自适应加权平均的融合规则;对高频细节方向分量应用显著性与引导滤波相结合的融合规则,以保持空间一致性,得到了相应的高频细节融合分量。结果表明,所提方法能更好地提取源图像的目标特征信息,保留丰富的背景信息,客观评价指标优于现有方法,融合结果具有更好的主观视觉效果。

A new multi-scale directional guided filter image fusion method based on guided filter and nonsubsampled directional filter bank is proposed. The proposed method possesses the feature of edge preserving and extracting ability of directional information, and can capture the useful information from the source images more effectively. The low-frequency subbands, which are obtained by the multi-scale directional guided filter, include the low- frequency approximation components and strong edge components. These components are separated by Gaussian filter. The low-frequency approximation components and strong edge components are fused based on convolutional sparse representation and adaptive regional energy, respectively. The detail directional subbands are fused via a strategy combined saliency and guided filter to preserve the spatial consistency. Experimental results demonstrate that the proposed method could effectively extract the target feature information and preserve the background information of the source images. The fused results have better subjective visual effect and objective evaluation criteria.