深度生成模型下缺损图像修复方法仿真研究

Simulation Research on Defect Image Repair Method Based on Depth Generation Model

图像在存储或传输过程中容易产生缺损,为获取全面的图像信息,提出一种基于深度生成模型的缺损图像修复方法。利用小波线性变换特征,在小波逆变换过程中选取合适的阈值去除图像噪声,得到初始图像,利用深度生成模型中的生成对抗网络增强图像质量,通过对抗训练增强缺损图像质量,将图像修复问题转换成像素填充问题,缩短结构部分与破损区域的距离,生成缺损图像预填充结果,利用PDE有限差分修复缺损图像中心点信息,利用人工复原法修改等照度线方向权重,实现缺损图像修复。实验结果表明,所提方法修复效果较好,能最大程度保留原始图像信息。

At present,the image is prone to defects during storage or transmission.In order to obtain complete image information,this paper put forward a method of repairing defective images based on deep generative model.At first,wavelet linear transformation features were utilized to select an appropriate threshold during the wavelet inverse transformation and remove image noise thus obtaining the initial image.Then,a generative adversarial network in deep generative models was used to enhance the quality of the defective image.Meanwhile,the problem of image restoration was transformed into a pixel-flling problem,thus shortening the distance between the structural part and the damaged area.After that,a preliminary flled result for the defective image was generated.Finally,the PDE finite difference method was adopted to repair the central point of the defective image.At the same time,the artificial restoration method was used to modify the weight of the isophote direction,thus achieving the restoration of the defective image.Experimental results show that the proposed method has good restoration effect and can retain the original image information to the greatest extent.