基于适应性多列生长模型的鲁棒图像修复方法

Robust Image Restoration Based on Adaptive Multiple Columns Generative Model

电影的后期制作中通常需要将图像进行去噪以及对象进行扣除后的图像进行补全,针对这种噪音去除以及修复技术已经有基于包的修复方法.本文提出了一种基于生长模型的鲁棒性图像背景修复算法（AMC-RBM）,这是一种新奇的技术,这种技术不仅使用多RBM（受限玻尔兹曼机）通过解决一个非线性优化程序来计算最优的列权值,还训练一个分布式网络来预测最优的权值.在测试的时候我们不需要明确需要噪音或者补全的空洞的种类,也不用关心统计数据,而且对于训练集之外的空洞我们甚至也可以表明系统足够鲁棒.

During the process of film post-production, images usually need denoising and completion. And aiming at finishing these twoaspects, there emerged a package-based restoration method. This paper proposes an algorithm of image background restorationbased on generative model （AMC -RBM）. This algorithm is a novel technology, which uses RBM （restricted boltzmann machine）not only to calculate the optimal weights by solving a nonlinear optimization program, but also to train a distributed network topredict the optimal weights. At the time of testing we don＇t need the concise void categories of noise or completion and care aboutstatistics either. Moreover, we can even show that system is robust enough with regard to voids beyond the training set.