基于Pytorch的图像降噪处理方法设计与实现

1.下载数据集

!wget -N https://raw.githubusercontent.com/eebowen/Transfer-Learning-and-Deep-Neural-Network-Acceleration-for-Image-Classification/master/nntools.py
!wget -N https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/bsds/BSDS300-images.tgz
!tar -zxvf BSDS300-images.tgz
dataset_root_dir = './BSDS300/images/'

%matplotlib inline

%matplotlib inline

import os
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.data as td
import torchvision as tv
from PIL import Image
import matplotlib.pyplot as plt
import time

2.训练集加噪

标准差为30的高斯噪声180*180彩铅（左上角或者随机位置）

class NoisyBSDSDataset(td.Dataset):

    def __init__(self, root_dir, mode='train', image_size=(180, 180), sigma=30):
        super(NoisyBSDSDataset, self).__init__()
        self.mode = mode
        self.image_size = image_size
        self.sigma = sigma
        self.images_dir = os.path.join(root_dir, mode)
        self.files = os.listdir(self.images_dir)

    def __len__(self):
        return len(self.files)

    def __repr__(self):
        return "NoisyBSDSDataset(mode={}, image_size={}, sigma={})". \
            format(self.mode, self.image_size, self.sigma)

    def __getitem__(self, idx):
        img_path = os.path.join(self.images_dir, self.files[idx])
        clean = Image.open(img_path).convert('RGB')   
        # 随机裁剪
        #i = np.random.randint(clean.size[0] - self.image_size[0])
        #j = np.random.randint(clean.size[1] - self.image_size[1])
        i=0
        j=0
        clean = clean.crop([i, j, i+self.image_size[0], j+self.image_size[1]])
        transform = tv.transforms.Compose([
            # 转换张量
            tv.transforms.ToTensor(),
            # [−1, 1]
            tv.transforms.Normalize((.5, .5, .5), (.5, .5, .5))
            ])
        clean = transform(clean)
        
        noisy = clean + 2 / 255 * self.sigma * torch.randn(clean.shape)
        return noisy, clean

def myimshow(image,ax = plt):
  image = image.to('cpu').numpy()
  image = np.moveaxis(image,[0,1,2],[2,0,1])
  image = (image + 1) / 2
  image[image < 0] = 0
  image[image > 1] = 1
  h = ax.imshow(image)
  ax.axis('off')
  return h

导入训练集和测试集进入

train_set= NoisyBSDSDataset(dataset_root_dir)
test_set = NoisyBSDSDataset(dataset_root_d