import torch<\/p>\nimport torch.nn as nn<\/p>\n
import torch.optim as optim<\/p>\n
from torchvision import datasets, transforms<\/p>\n
from torch.utils.data import DataLoader<\/p>\n
class DeblurCNN(nn.Module):<\/p>\n
def __init__(self):<\/p>\n
super(DeblurCNN, self).__init__()<\/p>\n
self.conv1 = nn.Conv2d(3, 64, kernel_size=5, padding=2)<\/p>\n
self.conv2 = nn.Conv2d(64, 64, kernel_size=5, padding=2)<\/p>\n
self.conv3 = nn.Conv2d(64, 3, kernel_size=5, padding=2)<\/p>\n
self.relu = nn.ReLU(inplace=True)<\/p>\n
def forward(self, x):<\/p>\n
x = self.relu(self.conv1(x))<\/p>\n
x = self.relu(self.conv2(x))<\/p>\n
x = self.conv3(x)<\/p>\n
return x<\/p>\n
def train_model(train_loader, model, criterion, optimizer, num_epochs=25):<\/p>\n
for epoch in range(num_epochs):<\/p>\n
model.train()<\/p>\n
running_loss = 0.0<\/p>\n
for inputs, targets in train_loader:<\/p>\n
inputs, targets = inputs.to(device), targets.to(device)<\/p>\n
optimizer.zero_grad()<\/p>\n
outputs = model(inputs)<\/p>\n
loss = criterion(outputs, targets)<\/p>\n
loss.backward()<\/p>\n
optimizer.step()<\/p>\n
running_loss += loss.item() * inputs.size(0)<\/p>\n
epoch_loss = running_loss \/ len(train_loader.dataset)<\/p>\n
print(f'Epoch {epoch}\/{num_epochs - 1}, Loss: {epoch_loss:.4f}')<\/p>\n
return model<\/p>\n
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")<\/p>\n
model = DeblurCNN().to(device)<\/p>\n
criterion = nn.MSELoss()<\/p>\n
optimizer = optim.Adam(model.parameters(), lr=0.001)<\/p>\n
train_dataset = datasets.ImageFolder('path_to_train_data', transform=transforms.ToTensor())<\/p>\n
train_loader = DataLoader(train_dataset, batch_size=32, shuffle=True)<\/p>\n
model = train_model(train_loader, model, criterion, optimizer, num_epochs=25)<\/p>\n
torch.save(model.state_dict(), 'deblur_cnn.pth')<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u4e09\u3001\u56fe\u50cf\u589e\u5f3a\u6280\u672f<\/h2>\n<\/p>\n
\u56fe\u50cf\u589e\u5f3a\u6280\u672f\u53ef\u4ee5\u901a\u8fc7\u8c03\u6574\u56fe\u50cf\u7684\u5bf9\u6bd4\u5ea6\u3001\u4eae\u5ea6\u548c\u9510\u5ea6\u7b49\u53c2\u6570\u6765\u63d0\u9ad8\u56fe\u50cf\u7684\u6e05\u6670\u5ea6\u3002\u8fd9\u4e9b\u6280\u672f\u53ef\u4ee5\u4f5c\u4e3a\u56fe\u50cf\u53bb\u6a21\u7cca\u7684\u8f85\u52a9\u65b9\u6cd5\uff0c\u8fdb\u4e00\u6b65\u6539\u5584\u56fe\u50cf\u7684\u89c6\u89c9\u6548\u679c\u3002<\/p>\n<\/p>\n
1\u3001\u5bf9\u6bd4\u5ea6\u589e\u5f3a<\/h3>\n<\/p>\n
\u901a\u8fc7\u8c03\u6574\u56fe\u50cf\u7684\u5bf9\u6bd4\u5ea6\uff0c\u53ef\u4ee5\u7a81\u51fa\u56fe\u50cf\u7684\u7ec6\u8282\uff0c\u589e\u5f3a\u89c6\u89c9\u6548\u679c\u3002OpenCV\u63d0\u4f9b\u4e86\u591a\u79cd\u65b9\u6cd5\u6765\u8c03\u6574\u56fe\u50cf\u7684\u5bf9\u6bd4\u5ea6\uff0c\u5305\u62ec\u76f4\u65b9\u56fe\u5747\u8861\u5316\u548c\u81ea\u9002\u5e94\u76f4\u65b9\u56fe\u5747\u8861\u5316\u3002<\/p>\n<\/p>\n
import cv2<\/p>\ndef enhance_contrast(image):<\/p>\n
lab = cv2.cvtColor(image, cv2.COLOR_BGR2LAB)<\/p>\n
l, a, b = cv2.split(lab)<\/p>\n
clahe = cv2.createCLAHE(clipLimit=3.0, tileGridSize=(8, 8))<\/p>\n
cl = clahe.apply(l)<\/p>\n
limg = cv2.merge((cl, a, b))<\/p>\n
return cv2.cvtColor(limg, cv2.COLOR_LAB2BGR)<\/p>\n
img = cv2.imread('blurred_image.jpg')<\/p>\n
enhanced_img = enhance_contrast(img)<\/p>\n
cv2.imwrite('enhanced_image.jpg', enhanced_img)<\/p>\n
<\/code><\/pre>\n<\/p>\n
2\u3001\u9510\u5ea6\u589e\u5f3a<\/h3>\n<\/p>\n
\u901a\u8fc7\u9510\u5316\u6ee4\u6ce2\u5668\uff0c\u53ef\u4ee5\u589e\u5f3a\u56fe\u50cf\u7684\u8fb9\u7f18\u548c\u7ec6\u8282\uff0c\u63d0\u9ad8\u56fe\u50cf\u7684\u6e05\u6670\u5ea6\u3002\u5e38\u7528\u7684\u9510\u5316\u6ee4\u6ce2\u5668\u5305\u62ec\u62c9\u666e\u62c9\u65af\u6ee4\u6ce2\u5668\u548c\u9ad8\u901a\u6ee4\u6ce2\u5668\u3002<\/p>\n<\/p>\n
import cv2<\/p>\nimport numpy as np<\/p>\n
def sharpen_image(image):<\/p>\n
kernel = np.array([[0, -1, 0], [-1, 5, -1], [0, -1, 0]])<\/p>\n
return cv2.filter2D(image, -1, kernel)<\/p>\n
img = cv2.imread('blurred_image.jpg')<\/p>\n
sharpened_img = sharpen_image(img)<\/p>\n
cv2.imwrite('sharpened_image.jpg', sharpened_img)<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u56db\u3001\u7efc\u5408\u5e94\u7528<\/h2>\n<\/p>\n
\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\uff0c\u5e38\u5e38\u9700\u8981\u7efc\u5408\u591a\u79cd\u65b9\u6cd5\u6765\u8fbe\u5230\u6700\u4f73\u7684\u56fe\u50cf\u53bb\u6a21\u7cca\u6548\u679c\u3002\u4e0b\u9762\uff0c\u6211\u4eec\u5c06\u7ed9\u51fa\u4e00\u4e2a\u7efc\u5408\u5e94\u7528\u7684\u793a\u4f8b\uff0c\u7ed3\u5408\u56fe\u50cf\u53bb\u6a21\u7cca\u7b97\u6cd5\u3001\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u548c\u56fe\u50cf\u589e\u5f3a\u6280\u672f\u6765\u5904\u7406\u6a21\u7cca\u56fe\u50cf\u3002<\/p>\n<\/p>\n
import cv2<\/p>\nimport numpy as np<\/p>\n
import torch<\/p>\n
from PIL import Image<\/p>\n
from torchvision import transforms<\/p>\n
from deblurgan import generator<\/p>\n
def combined_deblur(image_path, deblurgan_model_path):<\/p>\n
# Step 1: Initial deblurring using DeblurGAN<\/p>\n
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")<\/p>\n
model = generator.DeBlurNet()<\/p>\n
model.load_state_dict(torch.load(deblurgan_model_path, map_location=device))<\/p>\n
model.to(device)<\/p>\n
model.eval()<\/p>\n
image = Image.open(image_path).convert("RGB")<\/p>\n
transform = transforms.Compose([transforms.ToTensor()])<\/p>\n
image = transform(image).unsqueeze(0).to(device)<\/p>\n
with torch.no_grad():<\/p>\n
deblurred = model(image)<\/p>\n
deblurred_image = deblurred.squeeze().cpu().numpy().transpose(1, 2, 0)<\/p>\n
deblurred_image = (deblurred_image * 255).astype(np.uint8)<\/p>\n
# Step 2: Contrast enhancement<\/p>\n
lab = cv2.cvtColor(deblurred_image, cv2.COLOR_BGR2LAB)<\/p>\n
l, a, b = cv2.split(lab)<\/p>\n
clahe = cv2.createCLAHE(clipLimit=3.0, tileGridSize=(8, 8))<\/p>\n
cl = clahe.apply(l)<\/p>\n
limg = cv2.merge((cl, a, b))<\/p>\n
contrast_enhanced = cv2.cvtColor(limg, cv2.COLOR_LAB2BGR)<\/p>\n
# Step 3: Sharpening<\/p>\n
kernel = np.array([[0, -1, 0], [-1, 5, -1], [0, -1, 0]])<\/p>\n
final_image = cv2.filter2D(contrast_enhanced, -1, kernel)<\/p>\n
return final_image<\/p>\n
combined_result = combined_deblur('blurred_image.jpg', 'deblurgan_model.pth')<\/p>\n
cv2.imwrite('final_restored_image.jpg', combined_result)<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u901a\u8fc7\u8fd9\u79cd\u7efc\u5408\u65b9\u6cd5\uff0c\u6211\u4eec\u53ef\u4ee5\u6709\u6548\u5730\u6062\u590d\u6a21\u7cca\u56fe\u50cf\u7684\u6e05\u6670\u5ea6\uff0c\u63d0\u5347\u56fe\u50cf\u7684\u89c6\u89c9\u8d28\u91cf\u3002<\/p>\n<\/p>\n
\u4e94\u3001\u603b\u7ed3<\/h2>\n<\/p>\n
\u56fe\u50cf\u53bb\u6a21\u7cca\u662f\u4e00\u4e2a\u590d\u6742\u7684\u4efb\u52a1\uff0c\u6d89\u53ca\u591a\u79cd\u6280\u672f\u548c\u65b9\u6cd5\u3002\u901a\u8fc7\u56fe\u50cf\u53bb\u6a21\u7cca\u7b97\u6cd5\u3001\u5377\u79ef\u795e\u7ecf\u7f51\u7edc<\/strong>\u548c\u56fe\u50cf\u589e\u5f3a\u6280\u672f<\/strong>\u7b49\u65b9\u6cd5\uff0c\u6211\u4eec\u53ef\u4ee5\u6709\u6548\u5730\u5904\u7406\u6a21\u7cca\u56fe\u50cf\uff0c\u63d0\u9ad8\u56fe\u50cf\u7684\u6e05\u6670\u5ea6\u3002\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\uff0c\u53ef\u4ee5\u6839\u636e\u5177\u4f53\u60c5\u51b5\u9009\u62e9\u5408\u9002\u7684\u65b9\u6cd5\uff0c\u751a\u81f3\u7ed3\u5408\u591a\u79cd\u65b9\u6cd5\u6765\u8fbe\u5230\u6700\u4f73\u6548\u679c\u3002\u5e0c\u671b\u672c\u6587\u7684\u4ecb\u7ecd\u80fd\u5e2e\u52a9\u4f60\u5728\u56fe\u50cf\u53bb\u6a21\u7cca\u7684\u4efb\u52a1\u4e2d\u53d6\u5f97\u66f4\u597d\u7684\u6548\u679c\u3002<\/p>\n<\/p>\n\u76f8\u5173\u95ee\u7b54FAQs\uff1a<\/strong><\/h2>\n \u5982\u4f55\u4f7f\u7528Python\u4fee\u590d\u6a21\u7cca\u56fe\u7247\uff1f<\/strong>
\u4fee\u590d\u6a21\u7cca\u56fe\u7247\u7684\u5e38\u7528\u65b9\u6cd5\u662f\u4f7f\u7528\u56fe\u50cf\u5904\u7406\u5e93\uff0c\u4f8b\u5982OpenCV\u6216PIL\uff08Pillow\uff09\u3002\u53ef\u4ee5\u901a\u8fc7\u5e94\u7528\u9510\u5316\u6ee4\u955c\u3001\u53bb\u6a21\u7cca\u7b97\u6cd5\u6216\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u6765\u63d0\u5347\u56fe\u50cf\u7684\u6e05\u6670\u5ea6\u3002\u5177\u4f53\u6b65\u9aa4\u5305\u62ec\uff1a\u52a0\u8f7d\u56fe\u50cf\u3001\u9009\u62e9\u5408\u9002\u7684\u4fee\u590d\u7b97\u6cd5\u5e76\u6267\u884c\u5904\u7406\uff0c\u6700\u540e\u4fdd\u5b58\u6216\u5c55\u793a\u4fee\u590d\u540e\u7684\u56fe\u50cf\u3002<\/p>\n\u4f7f\u7528\u54ea\u4e9bPython\u5e93\u53ef\u4ee5\u5904\u7406\u6a21\u7cca\u56fe\u7247\uff1f<\/strong>
\u5728Python\u4e2d\uff0c\u6709\u51e0\u4e2a\u6d41\u884c\u7684\u56fe\u50cf\u5904\u7406\u5e93\u53ef\u4ee5\u7528\u6765\u4fee\u590d\u6a21\u7cca\u56fe\u7247\u3002OpenCV\u662f\u4e00\u4e2a\u529f\u80fd\u5f3a\u5927\u7684\u8ba1\u7b97\u673a\u89c6\u89c9\u5e93\uff0c\u9002\u5408\u5904\u7406\u590d\u6742\u7684\u56fe\u50cf\u4fee\u590d\u4efb\u52a1\u3002PIL\uff08Pillow\uff09\u63d0\u4f9b\u4e86\u7b80\u5355\u7684\u63a5\u53e3\uff0c\u9002\u7528\u4e8e\u57fa\u672c\u7684\u56fe\u50cf\u64cd\u4f5c\u3002\u5176\u4ed6\u5e93\u5982scikit-image\u548cimageio\u4e5f\u53ef\u4ee5\u7528\u4e8e\u56fe\u50cf\u589e\u5f3a\u548c\u5904\u7406\u3002<\/p>\n\u5982\u4f55\u8bc4\u4f30\u4fee\u590d\u540e\u7684\u56fe\u7247\u6548\u679c\uff1f<\/strong>
\u8bc4\u4f30\u4fee\u590d\u6548\u679c\u53ef\u4ee5\u901a\u8fc7\u591a\u79cd\u65b9\u5f0f\u8fdb\u884c\u3002\u5e38\u89c1\u7684\u65b9\u6cd5\u5305\u62ec\u89c6\u89c9\u68c0\u67e5\u3001\u8ba1\u7b97\u56fe\u50cf\u7684\u5cf0\u503c\u4fe1\u566a\u6bd4\uff08PSNR\uff09\u548c\u7ed3\u6784\u76f8\u4f3c\u6027\u6307\u6570\uff08SSIM\uff09\u3002PSNR\u7528\u4e8e\u91cf\u5316\u56fe\u50cf\u8d28\u91cf\uff0cSSIM\u5219\u8003\u8651\u4e86\u4eba\u773c\u5bf9\u56fe\u50cf\u611f\u77e5\u7684\u7279\u5f81\u3002\u901a\u8fc7\u6bd4\u8f83\u4fee\u590d\u524d\u540e\u7684\u56fe\u50cf\u8d28\u91cf\u6307\u6807\uff0c\u53ef\u4ee5\u5ba2\u89c2\u5730\u8bc4\u4f30\u4fee\u590d\u6548\u679c\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"\u56fe\u7247\u6a21\u7cca\u53ef\u4ee5\u901a\u8fc7\u591a\u79cd\u65b9\u6cd5\u4fee\u590d\u6e05\u6670\uff0c\u5305\u62ec\u4f7f\u7528\u56fe\u50cf\u53bb\u6a21\u7cca\u7b97\u6cd5\u3001\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u3001\u56fe\u50cf\u589e\u5f3a\u6280\u672f\u7b49\u3002 \u5728\u8fd9\u7bc7\u6587\u7ae0\u4e2d\uff0c\u6211\u4eec\u5c06 […]","protected":false},"author":3,"featured_media":1174056,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[37],"tags":[],"acf":[],"_links":{"self":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/1174046"}],"collection":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/comments?post=1174046"}],"version-history":[{"count":"1","href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/1174046\/revisions"}],"predecessor-version":[{"id":1174057,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/1174046\/revisions\/1174057"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/media\/1174056"}],"wp:attachment":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/media?parent=1174046"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/categories?post=1174046"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/tags?post=1174046"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
![\"\u56fe\u7247\u6a21\u7cca\u5982\u4f55\u4fee\u590d\u6e05\u6670Python\"](\"https:\/\/cdn-kb.worktile.com\/kb\/wp-content\/uploads\/2024\/04\/26075654\/617ff4d0-cf83-4da6-9c3c-9a7444c83459.webp\")
<\/p>\n