A Differentiable Approximation Approach to Contrast-Aware Image Fusion
A Differentiable Approximation Approach to Contrast-Aware Image Fusion
复制标题
DOI：
10.1109/lsp.2014.2314647
复制
发表时间：
2014
影响因子：
3.9
通讯作者：
K. Hara;K. Inoue;K. Urahama
中科院分区：
文献类型：
--
作者：
K. Hara;K. Inoue;K. Urahama
我们提出了一种新的图像融合权重优化方法，以获得增强的图像。给定在不同摄影条件（例如曝光时间和焦深）下捕获的一组静态场景图像作为输入，该算法根据视觉显着性修改输入图像，然后搜索使梯度幅值总量最大化的图像线性组合。通过使用 log-sum-exp 函数逼近不可微拉格朗日函数，然后迭代更新封闭式解析解直至收敛来执行搜索。这种简单的算法收敛速度快，并且与几种传统技术相比，图像质量有了显着的提高。
We propose a new weight optimization method for image fusion to obtain enhanced images. Given as input a set of images of a static scene captured under different photographic conditions such as exposure time and depth of focus, the algorithm modifies the input images based on visual saliency and then searches for a linear combination of the images that maximizes the total amount of gradient magnitudes. The search is performed by approximating a non-differentiable Lagrangian with the log-sum-exp function and then iteratively updating the closed-form analytical solution until convergence. The simple algorithm has converged fast and has demonstrated significant improvement in image quality over several conventional techniques.