CAJ | 학술논문

　　在基于多尺度变换的图像融合中，通常对分解后的低频图像采用加权求和的融合规则，而对于异源图像融合的情形，两幅源图的低频分量之间亮度差距较大，这时权值的选择直接影响到融合图像的视觉效果，简单的平均规则会使融合图像对比度降低。针对这一问题，该文提出将加权系数看作一个函数，用全变分方法对其进行优化；为使该方法可适用于所有的多尺度变换像融合，分别建立了图像域和变换域上的全变分模型。实验表明该方法能得到细致准确的权值函数，很好地保持源图像中的显著目标，提高了融合效果，从而证明了该方法的有效性和优越性。
　　재기우다척도변환적도상융합중，통상대분해후적저빈도상채용가권구화적융합규칙，이대우이원도상융합적정형，량폭원도적저빈분량지간량도차거교대，저시권치적선택직접영향도융합도상적시각효과，간단적평균규칙회사융합도상대비도강저。침대저일문제，해문제출장가권계수간작일개함수，용전변분방법대기진행우화；위사해방법가괄용우소유적다척도변환상융합，분별건립료도상역화변환역상적전변분모형。실험표명해방법능득도세치준학적권치함수，흔호지보지원도상중적현저목표，제고료융합효과，종이증명료해방법적유효성화우월성。
In the image fusion methods which are based on multiscale transform, the lowpass subbands are usually fused through a weighed sum rule, this may lead the intensity of the fused image to a balenced level, some bright or dark objects will not preserved especially when the two source images are taken from different types of sensor. This paper proposes a total variation method for the optimization of lowpass weight function, which can be built on either image domain or transform domain. Several multiscale transforms are chosen to verify the proposed method. Experimental results show that the method can obtain delicate and precise optimized weight function, and preserve the intensity of salient objects well, thus improve the visual quality of the fused image. Moreover, this method can be applied to any kind of multiscale transform based image fusion.