一种基于随机退化的模糊图像修复方法

• Format: Patent
• Language: Chinese
• Published: 26.07.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

本发明公开了一种基于随机退化的模糊图像修复方法,包括以下步骤：S1.给定模糊的物理或者退化模型；S2.在深度模型训练的每一次迭代中，根据物理或者退化模型随机选定一组参数，获取模糊核；S3.将该模糊核与整个batch的图像进行图像卷积，获取模糊退化图像；S4.将获得的模糊图像与原始的清晰图像用于本次迭代的深度模型训练；S5.重复步骤S2~S4，直至模型训练完毕。本发明基于相关模糊的物理模型或退化模型已知的情况下，通过随机参数迭代，完成深度网络的训练，使训练出的模型具有更强的泛化性，可用于模糊图像修复过程中对数据集生成进行随机退化参数的迭代自适应选取。 The invention discloses a blurred image restoration method based on random degradation. The method comprises the following steps: S1, giving a blurred physical or degradation model; s2, in each iteration of deep model training, randomly selecting a group of parameters according to a physical or degradation model, and obtaining a fuzzy kernel; s3, carrying out image convolution on the blurred kernel and the image of the whole batch to obtain a blurred degraded image; s4, using the obtained blurred image and the original clear image for the deep model training of the iteration; and S5, repeating the steps S2-S4 until the model training is completed. Under the condition that a related fuzzy physical model or a degradation model is known, t