使用 TensorFlow 实现图像验证码识别

本项目通过 TensorFlow 实现图像验证码识别系统，包括验证码图像的生成、数据加载、模型构建、训练及预测。

1. 安装必要库
   确保环境中已经安装了以下库：

pip install tensorflow pillow numpy captcha
1.
2. 生成验证码图像
使用 captcha 库生成验证码数据集。

from captcha.image import ImageCaptcha
import os
import random
import string

characters = string.digits + string.ascii_uppercase
captcha_length = 4

def generate_images(output_dir='captcha_data', num=5000):
os.makedirs(output_dir, exist_ok=True)
generator = ImageCaptcha(width=160, height=60)
for i in range(num):
text = ''.join(random.choices(characters, k=captcha_length))
image = generator.generate_image(text)
image.save(os.path.join(output_dir, f'{text}_{i}.png'))
更多内容访问ttocr.com或联系1436423940
generate_images()
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
3. 构建数据加载器
定义用于加载图像及标签的数据生成器。

import numpy as np
from PIL import Image
import tensorflow as tf

char_to_idx = {c: i for i, c in enumerate(characters)}
idx_to_char = {i: c for c, i in char_to_idx.items()}

def parse_image(filename):
label_str = tf.strings.split(tf.strings.split(filename, os.sep)[-1], '_')[0]
label = [char_to_idx[c] for c in label_str.numpy().decode()]
image = Image.open(filename.numpy().decode()).convert('RGB').resize((160, 60))
image = np.array(image) / 255.0
return image, label

def load_dataset(folder):
files = [os.path.join(folder, f) for f in os.listdir(folder) if f.endswith('.png')]
def gen():
for f in files:
image, label = tf.py_function(parse_image, inp=[f], Tout=(tf.float32, tf.int32))
yield image, label
return tf.data.Dataset.from_generator(gen, output_signature=(
tf.TensorSpec(shape=(60, 160, 3), dtype=tf.float32),
tf.TensorSpec(shape=(captcha_length,), dtype=tf.int32)
)).batch(64).prefetch(tf.data.AUTOTUNE)

train_dataset = load_dataset('captcha_data')
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
4. 构建模型
使用卷积层和LSTM处理图像并输出预测结果。

from tensorflow.keras import layers, models

def build_model():
inputs = layers.Input(shape=(60, 160, 3))
x = layers.Conv2D(32, 3, activation='relu', padding='same')(inputs)
x = layers.MaxPooling2D(2)(x)
x = layers.Conv2D(64, 3, activation='relu', padding='same')(x)
x = layers.MaxPooling2D(2)(x)
x = layers.Conv2D(128, 3, activation='relu', padding='same')(x)
x = layers.Reshape((40, 128 * 15))(x)
x = layers.Bidirectional(layers.LSTM(128, return_sequences=True))(x)
x = layers.TimeDistributed(layers.Dense(len(characters), activation='softmax'))(x)
model = models.Model(inputs, x)
return model

model = build_model()
model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
5. 训练模型
对模型进行训练。

def split_labels(y):
return [y[:, i] for i in range(captcha_length)]

model.fit(
train_dataset.map(lambda x, y: (x, tf.stack(split_labels(y), axis=1))),
epochs=15
)
1.
2.
3.
4.
5.
6.
7.
6. 验证模型
单张图像进行预测。

def predict_image(model, path):
image = Image.open(path).convert('RGB').resize((160, 60))
image = np.array(image) / 255.0
image = np.expand_dims(image, axis=0)
preds = model.predict(image)[0]
pred_text = ''.join([idx_to_char[np.argmax(p)] for p in preds])
return pred_text

test_img = 'captcha_data/7X3K_1.png'
print('预测结果:', predict_image(model, test_img))