Deep Clean: An Investigation into Machine Learning-based Approaches to Artefact Removal in Digitised Analog Photographic Images

深度清洁：对基于机器学习的数字化模拟摄影图像中的伪影去除方法的研究

• 批准号: 2448921
• 金额: --
• 依托单位: University of Glasgow
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2020
• 资助国家: 英国
• 起止时间: 2020 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2448921
• 关键词: Deep; Clean; Investigation; into; Machine

The goal of this research is to provide a reliable way to identify and correct artefacts in images which are introduced by the medium through which the data is captured. It aims to do so by exploiting the inner structure of pre-trained neural networks in order to fine tune them to decouple, i.e. "disentangle" image content information from artefacts introduced by the recording medium. As preserving useful, non-corrupted content information is an important requirement to the task of restoration, successfully separating content from artefacts requires an attention mechanism to be added to improve baseline deep learning approaches to generating restored data from corrupted input.

本研究的目的是提供一种可靠的方法来识别和纠正图像中的伪影，这些伪影是由捕获数据的介质引入的。它的目的是通过利用预训练神经网络的内部结构来微调它们以解耦，即。将图像内容信息从记录介质引入的人工制品中“解开”。由于保留有用的、未损坏的内容信息是恢复任务的重要要求，因此成功地将内容与工件分离需要添加注意机制，以改进基线深度学习方法，从而从损坏的输入中生成恢复的数据。