光学学报, 2017, 37 (8): 0810001, 网络出版: 2018-09-07
基于目标提取与引导滤波增强的红外与可见光图像融合 下载: 1049次
Infrared and Visible Image Fusion Based on Target Extraction and Guided Filtering Enhancement
图像处理 图像融合 红外图像 可见光图像 非下采样Shearlet变换 目标提取 image processing image fusion infrared image visible image non-subsampled shearlet transform target extraction guided filtering enhancement
摘要
为了使融合结果突出目标并发掘更多细节,提出了一种基于目标提取与引导滤波增强的红外与可见光图像融合方法。首先对红外图像依据二维Tsallis熵和基于图的视觉显著性模型提取目标区域。然后对可见光与红外图像分别进行非下采样Shearlet变换(NSST),并对所得低频分量进行引导滤波增强。由增强后的红外图像和可见光图像低频分量基于目标提取的融合规则得到融合图像的低频分量,高频分量则根据方向子带信息和取大来确定。最后经NSST逆变换得到融合图像。大量实验结果表明,本文方法在增强融合图像空间细节的同时,有效突出了目标,并且在信息熵、平均梯度等指标上优于基于拉普拉斯金字塔变换、基于小波变换、基于平稳小波变换、基于非下采样Contourlet变换(NSCT)、基于目标提取与NSCT变换等。
Abstract
In order to highlight the fusion result and dig out more details, a fusion method of infrared image and visible image based on the target extraction and guidance filtering enhancement is proposed. Firstly, the two-dimensional Tsallis entropy and graph-based visual saliency model are used to extract the target region of infrared image. Then the visible image and the infrared image are decomposed by non-subsampled shearlet transform (NSST), respectively. The low-frequency components of the visible image and the infrared image are enhanced with guided filtering, respectively. The low-frequency component of the fused image is obtained from the enhanced low-frequency component of the infrared image and the visible image based on the fusion rule of target extraction, and the high-frequency components of the fused image are gained according to the maximization criterion of the directional sub-band information sum. Finally, the fused image is obtained by inverse NSST transform. A large number of experimental results demonstrate that the proposed method can improve the spatial resolution of the fused image, effectively highlight the target, and is superior to the method based on the Laplacian pyramid transform, the method based on wavelet transform, the method based on stationary wavelet transform, the method based on non-subsampled contourlet transform (NSCT), the method based on target extraction and NSCT in the quantitative evaluation indexes such as information entropy and average gradient.
吴一全, 王志来. 基于目标提取与引导滤波增强的红外与可见光图像融合[J]. 光学学报, 2017, 37(8): 0810001. Yiquan Wu, Zhilai Wang. Infrared and Visible Image Fusion Based on Target Extraction and Guided Filtering Enhancement[J]. Acta Optica Sinica, 2017, 37(8): 0810001.