基于全变分的权值优化的多尺度变换图像融合

A Total Variation-based Lowpass Weight Function Optimization in Multiscale Image Fusion

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。

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.