结合目标提取和压缩感知的红外与可见光图像融合

Fusion of infrared and visible images based on target segmentation and compressed sensing

针对红外与可见光图像融合易受噪声干扰从而使目标信息减弱的问题,提出了一种基于目标区域提取和压缩感知的融合算法。首先,在频率域上对红外图像进行显著区域检测得到其对应的显著度图,并在显著图指导下结合区域生长法提取红外图像的目标区域,有效抑制噪声与复杂背景的干扰。然后,用非下采样剪切波变换对待融合的图像进行分解,采用不同的融合策略分别对目标与背景区域的高、低频子带进行融合。针对背景区域提出一种新的基于多分辨率奇异值分解和压缩感知的融合规则,最后,进行非下采样剪切波逆变换得到融合图像。与其他算法的对比实验结果表明,本文算法能更好地突出目标区域,保留图像细节信息,抑制噪声干扰;图像质量评价指标中的信息熵、标准差、互信息、边缘保持度分别提高了3.94%,19.14%,9.96%和8.52%。

The image fusion of infrared and visible light is susceptible to noise and the target information is weakened easily. Therefore, a new fusion algorithm based on target area extraction and compressed sensing was proposed. Firstly, the infrared image was detected in a salient region at frequency-tuned domain to obtain a corresponding salient map. Under the guidance of the salient map, the infrared target area was extracted together with region growing method to effectively overcome the effects of noise and complex background interference on target segmentation. Then, non-subsampled shearlet transform was adopted to decompose the source images and the high and low frequency subbands in the target and backgound regions were fused respectively. Finally, a new fusion rule was proposed based on multi-resolution singular value decomposition and compressed sensing, and the fused image was reconstructed by the non-subsampled shearlet inverse transform. As compared with the other algorithms, experimental results show that the algorithm highlights the target area, preserves the details of the source images and suppresses the noise interference. The image fusion quality evaluation indexes including information entropy, standard deviation, mutual information and transferred edge information have increased by 3.94% ,19.14%,9.96%and 8.52%, respectively.