梯度域多曝光图像融合

为解决多曝光图像融合的问题,提出了一种新颖的直接将多曝光图像融合成适合显示的低动态范围图像的方法,并且具备消除运动伪影的能力。使用联合柱状图探究两幅图像之间像素值的对应关系,对每幅图像检测分别得到一幅伪影权值图,用来构造融合时的度量。在梯度域中进行融合,根据多维黎曼流形几何(Riemannian Geometry),从多个输入图像的结构张量中推导得出融合的梯度域,然后使用均值滤波器在多尺度上进行梯度域的非线性修改,通过解泊松方程得到融合图像。最后,还提出了一种新颖的色彩复原算法,用以避免造成色彩偏差。实验表明,该方法能够有效融合含运动物体的多曝光图像,无论哪幅图像被选作参考图像,都能获得高质量的结果图像。该算法同时具有处理的有效性和时间的高效性。

Multi-exposure images fusion in gradient field

For the multi-exposure image fusion problem,a novel method is presented for directly fusing multi-exposure images into a low dynamic range image suitable for displaying with the ability of removing ghost. The joint histogram is proposed to generate ghost maps for each image,which are used to construct metrics. The fusion is conducted in the gradient field. The fused gradient is derived from the structure tensor of inputs based on multidimensional Riemannian geometry with the constructed metric. And then an effective gradient modification algorithm is applied in multi-scale for the dynamic range compression. The result is obtained through solving a Poisson equation. Finally,a new color restoration algorithm is proposed for avoiding color artifacts. Experimental results on real images demonstrate that the proposed method is especially effective at fusing images no matter which image is selected as reference,and it is of both efficiency and effectiveness.