基于多判断和加权最小二乘优化的NSCT红外和可见图像融合

Fusion of NSCT infrared and visible images based on multi-judgment and WLS optimization

为了克服传统方法的一些缺陷和单一特征提取信息的不足,在进一步提高红外和可见图像融合的同时,寻找针对不同类型特点的适应能力强的方法,提出了一种基于多判断与加权最小二乘优化(WLS)的非下采样轮廓波变换(NSCT)红外可见图像融合方法。首先,采用NSCT对图像进行多尺度分解,得到图像的低频和高频子带。其次,低频子带选择局部平方熵和修正拉普拉斯和(SML)来相互补充,在保证好的对比度下提取少量细节信息;高频子带充分考虑底层特征的重要性,选择相位一致性(PC)、局部加权修正拉普拉斯算子和(WSML)以及局部加权能量(WLE)相互补充的方式融合细节层,对其进行WLS优化,融合后的图像细节更自然,更适合人眼视觉感知。最后,对融合后的低频和高频子带进行逆变换,得到融合图像。对不同类型特点的图像进行了实验验证,实验结果表明,与其他融合方法相比,本文方法在主观上目标显著、背景清晰、视觉效果好。在4个客观评价指标平均梯度(AG)、信息熵(IE)、空间频率(SF)、互信息(MI)中,在保证MI指标比较好的前提下,其他3个指标都处于最好的状态,尤其是对于光照均匀的camp图像,AG和SF与最好的数值相比提高了6.9%和4.8%,从而验证了本文方法的有效性。

In order to overcome some defects of traditional methods and the insufficiency of single feature extraction information,while further improving the fusion of infrared and visible images,a method with strong adaptability to different types of features is sought.A non-subsampled contourlet transform(NSCT)infrared-visible image fusion method based on multi-judgment and weighted least squares optimization(WLS)is proposed.Firstly,NSCT is used to decompose the image at multiple scales to obtain the low-frequency and high-frequency subbands of the image.Secondly,the low-frequency sub-band selects local squared entropy and sum-modified laplacian(SML)to complement each other,which extracts a small amount of detailed information under guaranteed good contrast.The high-frequency subbands fully considers the importance of the underlying features,and selects phase consistency(PC),the local weighted sum-modified Laplacian(WSML)and the local weighted energy(WLE)to complement each other to fuse the detail layer.They are optimized by WLS,and the fused image details are more natural and more suitable for human visual perception.Finally,the inverse transform is performed on the fused low-frequency and highfrequency subbands to obtain a fused image.Through the experimental verification of images with different types of characteristics,experimental results show that the proposed method has subjectively significant targets,clear backgrounds and better visual effects in comparison with other fusion methods.Under the average gradient(AG),information entropy(IE),spatial frequency(SF),and mutual information(MI)of the four objective evaluation indicators,on the premise that MI is relatively good,the other three indicators are in the best position,especially for uniformly illuminated camp images,AG and SF are improved by 6.9%and 4.8%in comparison with the best values,thus validating the proposed method effectiveness.