基于改进FT显著性检测的NSCT红外和可见光图像融合  

Fusion of NSCT infrared and visible images based on improved FT saliency detection

作　　者：王贤涛 赵金宇[1] WANG Xian-tao;ZHAO Jin-yu(Changchun Institute of Optics,Fine Mechanics and Physics,Chinese Academy of Sciences,Changchun 130033,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院长春光学精密机械与物理研究所,吉林长春130033 [2]中国科学院大学,北京100049

出　　处：《液晶与显示》2023年第7期933-944,共12页Chinese Journal of Liquid Crystals and Displays

基　　金：国家自然科学基金(No.U1831106)。

摘　　要：为了克服传统算法存在对比度低、细节和纹理缺失严重以及基于显著性检测算法存在对噪声敏感、适应性和抗干扰能力不强的问题,提出了一种基于改进的频率调谐(FT)显著性检测的非下采样轮廓波变换(NSCT)红外与可见光融合方法。首先,采用改进的显著图提取算法瞄准红外图像,用于从背景中区分目标。其次,使用NSCT对红外图像和可见光图像进行高、低频子带的分解。利用计算得到的红外显著权重图对低频子带系数进行指导融合,可以很好地保留目标和背景之间的对比度;对高频部分采用局部加权能量的规则进行抉择,再通过加权最小二乘(WLS)优化可以获得更多的细节信息和减小噪声影响。最后,对融合后的高频和低频子带系数进行NSCT逆变换得到最终的融合图像。通过4组图像的实验对比结果可知,在主观上,本文方法相较于其他方法具有目标突出、细节提取丰富、边缘伪影现象消除明显、视觉效果更好的优点。本文方法在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)上都处于最好的状态,与5种对比方法相比,AG、IE、SF、MI的平均值分别提高了8.19%、5.34%、8.54%、68.18%,说明了所提出方法的可靠性和有效性。In order to overcome the problems of low contrast,serious lack of details and textures in traditional algorithms,and noise sensitivity,weak adaptability and anti-interference ability of saliency-based detection algorithms,a non-subsampled contourlet transform(NSCT)infrared and visible light fusion method based on improved frequency tuning(FT)saliency detection is proposed.Firstly,an improved saliency map extraction algorithm is used to target the infrared image to distinguish the target from the background.Secondly,the infrared image and visible light image are decomposed into high and low frequency subbands using NSCT.The coefficients are used to guide the fusion,which can well preserve the contrast between the target and the background.For the high-frequency part,the rule of local weighted energy is used to make a decision,the weighted least squares(WLS)optimization can obtain more detailed information and reduce the influence of noise,and the fused high-frequency and low-frequency subbands coefficients are processed.Finally,the fused image is obtained by inverse NSCT transform.The experimental comparison results of four sets of images show that the method in this paper has better visual effect than other methods in terms of prominent target,rich details extraction,and obvious elimination of edge artifacts,subjectively.It is in the best state on the average gradient(AG),information entropy(IE),spatial frequency(SF),and mutual information(MI)of four objective evaluation indicators.Compared with the average of the five comparison methods,the AG,IE,SF,and MI are increased by 8.19%,5.34%,8.54%,and 68.18%,respectively.Thus,the proposed method is reliable and effective.

关 键 词：图像融合 非下采样轮廓波变换 频率调谐显著性检测 红外显著图 加权最小二乘优化 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]