目录
中文内容摘要
Abstract
第一章 绪论及背景
1.1背景
1.2 图像去噪的发展状况
1.3 深度卷积神经网络
1.4 本文章节安排
第二章 相关工作
2.1图像的噪声模型
2.2深度神经网络
2.3卷积神经网络
2.4 卷积神经网络的训练
2.5 卷积神经网络应对图像处理
第三章 深度卷积网络模型
3.1 卷积神经网络基础
3.2 深度卷积网络应用于图像降噪
3.3 功能性拓展
第四章 实验结果及分析
4.1 训练数据
4.2 测试数据及对比
4.3 测试结果及分析
第五章 全文总结及展望
参考文献
附录
中文内容摘要
在这个信息时代,随着互联网的发展以及人工智能的发展,每个人都在习惯了利用自己的移动终端(手机、平板等)分享自己的生活,分享数字图像是其中一项很重要的内容。然而受到各种不可避免的外界影响，使得图像受到了噪声的干扰。图像的噪声不仅在一定程度上破坏了视觉感受体验，同时也对计算机对图像做进一步处理造成很大干扰。对于目前已经存在的一些去噪效果较好的算法，比如BM3D、LSSC，则又会有运算量较大的的问题。本文基于深度卷积网络，对图像进行去噪处理，并加以批量归一化等方法，以加速网络的训练过程。以高斯噪声为例，该网络对固定噪声水平的噪声以及不固定噪声水平的噪声，均有较好的去噪效果，同时拥有几十倍于BM3D的性能。同时，使用GPU加速将获得更高的性能。非常适合运算性能有限的移动端设备以及运算任务较多的服务器。
关键词：深度卷积神经网络（CNN），高斯噪声，图像去噪，批量归一化。
Abstract
In this information age, with the development of the Internet and the development of artificial intelligence, everyone is using their own mobile terminal to share their own lives, sharing digital images is one of the very important content. At the same time, the image processing is also a hot topic in artificial intelligence research. But by various unavoidable external influences, making the image subject to noise interference. Image noise not only affects the visual perception, but also on the computer to do further processing of the image caused great interference. There are some algorithms that have a good denoising effect, such as BM3D, LSSC, but they need a long time to run. In this paper, the image is denoised based on the deep convolution network, and the method of batch normalization is adopted to accelerate the training process of the network. Taking Gaussian noise as an example, the network has good denoising effect for noise at fixed noise level and noise with no fixed noise level, and it is several times faster than BM3D. Performance gains will be greater when using GPU acceleration. Very suitable for computing performance limited mobile terminal equipment and busy servers.
Keyword: Deep convolution network, Gaussian noise, Image denoising, Batch normalization。
