用Python和PyTorch实现验证码识别

本文介绍如何使用PyTorch构建一个完整的验证码识别系统，包括数据生成、模型构建、训练与预测。

1. 安装必要的库
   pip install torch torchvision pillow captcha numpy2. 生成验证码图片数据集
   使用captcha库生成图片作为训练数据：
   from captcha.image import ImageCaptcha
   import random
   import string
   import os
   更多内容访问ttocr.com或联系1436423940
   characters = string.digits + string.ascii_uppercase
   width, height = 160, 60
   length = 4

def create_captcha_images(num_samples=10000, save_dir="captcha_images"):
os.makedirs(save_dir, exist_ok=True)
generator = ImageCaptcha(width=width, height=height)
for i in range(num_samples):
text = ''.join(random.choices(characters, k=length))
img = generator.generate_image(text)
img.save(os.path.join(save_dir, f"{text}_{i}.png"))

create_captcha_images()3. 定义数据集加载类
编写Dataset类处理图片和标签：
from torch.utils.data import Dataset, DataLoader
from torchvision import transforms
from PIL import Image
import torch

class CaptchaDataset(Dataset):
def init(self, folder):
self.folder = folder
self.files = os.listdir(folder)
self.transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.5,), (0.5,))
])
self.char2idx = {char: idx for idx, char in enumerate(characters)}

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

def __getitem__(self, idx):
    filename = self.files[idx]
    label = filename.split('_')[0]
    image = Image.open(os.path.join(self.folder, filename)).convert('RGB')
    image = self.transform(image)
    label = torch.tensor([self.char2idx[c] for c in label], dtype=torch.long)
    return image, label

dataset = CaptchaDataset("captcha_images")
train_loader = DataLoader(dataset, batch_size=64, shuffle=True)4. 建立CNN+LSTM模型
搭建提取特征和识别序列的混合模型：
import torch.nn as nn

class CaptchaModel(nn.Module):
def init(self):
super().init()
self.cnn = nn.Sequential(
nn.Conv2d(3, 32, 3, padding=1), nn.ReLU(), nn.MaxPool2d(2),
nn.Conv2d(32, 64, 3, padding=1), nn.ReLU(), nn.MaxPool2d(2),
nn.Conv2d(64, 128, 3, padding=1), nn.ReLU(), nn.MaxPool2d((2,1))
)
self.rnn = nn.LSTM(128*7, 128, batch_first=True, num_layers=2, bidirectional=True)
self.fc = nn.Linear(256, len(characters))

def forward(self, x):
    x = self.cnn(x)
    x = x.permute(0, 3, 1, 2)
    b, w, c, h = x.size()
    x = x.reshape(b, w, c*h)
    x, _ = self.rnn(x)
    x = self.fc(x)
    return x5. 训练模型

进行模型训练：
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = CaptchaModel().to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=0.001)
criterion = nn.CrossEntropyLoss()

for epoch in range(20):
model.train()
total_loss = 0
for images, labels in train_loader:
images, labels = images.to(device), labels.to(device)
outputs = model(images)
loss = sum(criterion(outputs[:, i, :], labels[:, i]) for i in range(length))
optimizer.zero_grad()
loss.backward()
optimizer.step()
total_loss += loss.item()
print(f"Epoch {epoch+1}, Loss: {total_loss:.4f}")6. 测试模型
编写验证码识别函数：
def recognize(model, img_path):
model.eval()
transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.5,), (0.5,))
])
img = Image.open(img_path).convert('RGB')
img = transform(img).unsqueeze(0).to(device)
with torch.no_grad():
output = model(img)
pred = torch.argmax(output, dim=2)
pred_text = ''.join([characters[i] for i in pred[0]])
return pred_text

test_img = "captcha_images/F7Q3_1.png"
print("Predicted:", recognize(model, test_img))7. 可选的优化方向
• 增加数据增强如随机旋转、加噪声
• 尝试使用更深的卷积网络
• 应用CTC Loss处理不定长验证码
• 融入注意力机制提升识别效果