Development of Signal Processing Algorithms and Systems for Old Film Restration

本研究旨在修复由于老化、化学效应和机械变形等原因造成的几种损坏的旧电影序列。旧胶片序列中的斑点可以通过时间数字信号处理来去除，因为它们在时间上是不连续的。本研究提出了一种在老电影序列中生成和检测的随机模型，并通过数值模拟验证了该模型的正确性和有效性。针对含有人工闪烁的电影序列，提出了一种基于M-估计的快速闪烁抑制方法，并对其性能进行了测试。使用参考序列的方法可以比几种传统方法快19倍地执行闪变抑制。研究了一种基于图像马尔可夫随机场模型的快速有效的斑点检测方法。以含有人工斑点的胶片序列为例，测试了该方法的性能。实验结果表明，该方法检测斑点的速度是基于马尔可夫随机场模型的传统方法的5倍。研究了一种用于减少闪烁和斑点检测的参考图像的产生方法。实验结果表明，该方法可以产生足够低的闪烁和斑点噪声，并能产生很好的参考图像。

This research aims at restoring old film sequences with several damages due to aging, chemical effects and mechanical deformation, etc.1. Blotches in old film sequences can be removed by temporal digital signal processing because they are discontinuous in time. This research proposed a stochastic model for generation and detection in old film sequences and demonstrated its validity and effectiveness by numerical simulation.2. This research proposed a fast flicker reduction method using M-estimation for film sequences with artificial flickers and examined its performance. The proposed method using a reference sequence can execute flicker reduction 19 times faster than several conventional methods.3. This research studied a fast and effective blotch detection method based on MRF model of images. The performance of the proposed method was examined for film sequences with artificial blotches. It was confirmed that the proposed method can detect blotches 5 times faster than the conventional method based on MRF model.4. This research studied a producing method of reference images for flicker reduction and blotch detection. It was confirmed that the producing method proposed can generate very good reference images with sufficiently low flicker and blotch noises.