深度反卷积神经网络优化下的低质图像去模糊数学模型  

Deblurring mathematical model of low quality image under deep deconvolution neural network optimization

作　　者：亓金锋 Qi Jinfeng(Shandong Open University)

机构地区：[1]山东开放大学,山东济南250014

出　　处：《现代电影技术》2024年第12期56-61,共6页Advanced Motion Picture Technology

基　　金：山东省2024年度艺术科学重点课题“全面乡村振兴视域下山东省农村公共文化服务体系建设路径优化研究”(L2024Z05100824)。

摘　　要：随着时间的推移,电影胶片由于存放时间长、存储物理环境不一等原因会出现老化、褪色、划痕和模糊等问题,而现存经典资料影片胶片在转换为数字格式进行存储后,大多仍存在随机噪声多、图像质量不高等问题。为缓解上述问题,本文设计了基于深度反卷积神经网络优化下的低质图像去模糊数学模型,首先采用泊松分布法建立低质图像退化方程,分析低质图像的随机噪声;随后构建低质图像去模糊数学模型的初始框架,并利用深度反卷积神经网络对其进行优化,确定损失函数,完成低质图像去模糊数学模型的构建。实验结果显示,本文所提数学模型在实践应用中表现出良好的低质图像去模糊处理结果,峰值信噪比较高,可用于经典资料影片图像画面的修复,在低质图像去模糊领域具备良好的应用前景。Due to the long storage time,negative degradation,storage space physical environment and other reasons,aging,fading,scratches and blur and other problems would come across to motion picture film as time goes by.Followed with the conversion to digital format for storage,most of classic data motion picture film would appear to have random noise,low-quality image or other problems.In order to ease off problems listed above,this paper proposed a mathematical model of defuzzification of low quality images,which designed based on deep deconvolution neural network optimisation.The degradation equation of low quality images is primarily established by using Poisson distribution method to analyze the random noise of low quality images.Then the initial frame of the address image deblurring mathematical model is constructed by using the deep deconvolution neural network to optimize it.The method should confirm the loss function,and complete the construction of the low quality image deblurring mathematical model.The experimental results show that the mathematical model presented in this paper presents good deblurring effect for low quality images in practical application,and the peak signal-to-noise ratio is high.It can be applied in the restoration of classic data motion picture,and has a good application prospect in the field of deblurring low-quality images.

关 键 词：图像去模糊 低质图像 深度反卷积神经网络 模糊图像 图像处理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]