基于结构图像先验与ASM能量的服装图像去噪  

Structural image prior and ASM energy for clothing image denoising

作　　者：赵长乐 何利力[1] ZHAO Changle;HE Lili(School of Information,Zhejiang Sci-Tech University,Hangzhou 310018,China)

机构地区：[1]浙江理工大学信息学院,杭州310018

出　　处：《智能计算机与应用》2021年第6期133-138,共6页Intelligent Computer and Applications

摘　　要：服装图像在采集和传输过程中会受到噪声不同程度的影响,为更有效地去除服装图像中的噪声本文提出了一种基于ASM图像能量的深度学习图像去噪方法。该方法基于结构图像先验理论,以随机向量作为卷积神经网络的输入,含噪声的服装图像作为目标输出。网络通过反向传播进行迭代,根据噪声与自然图像对于网络的阻抗不同,迭代至输出图像的ASM能量极大值处进行截断,截断处的输出图像即为去噪后的服装图像。实验结果表明,该方法对服装图像去噪后的PSNR达到29.91,比NLM去噪提高了 0.74,比guided去噪提高了 1.97。与传统的图像滤波去噪算法相比,该方法能更有效地去除图像中的噪声,保留服装图像的纹理细节。Clothing images are affected by noise to different degrees during the process of collection and transmission.In order to remove noise in clothing images more effectively,a deep learning image denoising method based on ASM image energy is proposed.This method is based on the theory of structural image prior,using random vectors as the input of the convolutional neural network and noise-containing clothing images as the target output.The network is iterated through backpropagation,according to the different impedances of the noise and the natural image to the network,iterate to the maximum value of the ASM energy of the output image for truncation,and the output image at the truncation is the denoised clothing image.The experimental results show that the PSNR of this method after denoising clothing images reaches 29.91,which is 0.74 higher than NLM denoising and 1.97 higher than guided denoising.Compared with traditional image filtering and denoising algorithms,This method can more effectively remove the noise in the image and retain the texture details of the clothing image.

关 键 词：服装图像去噪 ASM图像能量 深度学习 结构图像先验 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]