基于分数阶全变分和扩散模型的图像去模糊方法

Image deblurring based on fractional-order total variation and diffusion models

图像去模糊是数字图像处理领域的重要研究方向之一.在许多实际应用中,由于成像设备和物体之间发生相对运动,产生的模糊会降低图像质量和视觉效果.本文提出了一种结合分数阶全变分(FTV)损失函数和去噪扩散概率模型(DDPM)的图像去模糊方法 .首先通过基于概率建模的DDPM实现对图像结构信息的增强,然后利用FTV损失函数作为正则项,进一步恢复图像细节.与传统的图像去模糊方法相比,本文方法能够在保持图像整体清晰度的同时,还原更多的图像细节信息.实验结果验证了该方法在恢复受运动模糊影响的图像上具有显著优越性,为图像去模糊领域的进一步发展提供了新方向.

Image deblurring is a significant area of study in the field of digital image processing.In various practical applications,due to relative motion between imaging devices and objects,the resulting blurring will degrade the image quality and visual effect.This paper proposes an image deblurring method combining Fractional-order Total Variation(FTV)loss function and The enhancement of image structure information through probability modeling by DDPM,followed by the incorporation of the FTV loss function as a regular‐ization term,facilitates the restoration of image details.Compared with traditional image deblurring methods,this method can restore more image detail while maintaining the overall image clarity.The experimental re‐sults have verified that the method has the ability to restore images affected by motion blur.The significant su‐periority provides a new direction for the further development of image deblurring.