使用 Java 和 Deeplearning4j 实现图像验证码识别

验证码识别是图像分类中的一个应用场景，常用于自动化测试、爬虫识别等领域。本文将演示如何用 Java 语言和深度学习框架 Deeplearning4j 实现一个简单的图像验证码识别系统。

1. 准备开发环境

使用 Maven 搭建项目，添加以下依赖到 pom.xml：

<dependencies>
  <dependency>
    <groupId>org.deeplearning4j</groupId>
    <artifactId>deeplearning4j-core</artifactId>
    <version>1.0.0-beta7</version>
  </dependency>
  <dependency>
    <groupId>org.nd4j</groupId>
    <artifactId>nd4j-native-platform</artifactId>
    <version>1.0.0-beta7</version>
  </dependency>
 更多内容访问ttocr.com或联系1436423940 <dependency>
    <groupId>org.datavec</groupId>
    <artifactId>datavec-api</artifactId>
    <version>1.0.0-beta7</version>
  </dependency>
</dependencies>

2. 加载验证码数据

假设你已经有一个图片文件夹 captcha_samples/，图片名形如 7K2B_1.png，其中 7K2B 是验证码内容。

我们使用 ParentPathLabelGenerator 来从文件名中提取标签信息。

File parentDir = new File("captcha_samples");
FileSplit fileSplit = new FileSplit(parentDir, NativeImageLoader.ALLOWED_FORMATS);
BalancedPathFilter pathFilter = new BalancedPathFilter(new Random(), NativeImageLoader.ALLOWED_FORMATS);
InputSplit[] inputSplits = fileSplit.sample(pathFilter, 0.8, 0.2);
InputSplit trainData = inputSplits[0];
InputSplit testData = inputSplits[1];

3. 图像转换与标签编码

int height = 60;
int width = 160;
int channels = 1;
int outputNum = 36;  // 0-9, A-Z
int labelLength = 4;

ImageTransform transform = new ResizeImageTransform(width, height);

ImageRecordReader recordReader = new ImageRecordReader(height, width, channels, new CaptchaLabelGenerator());
recordReader.initialize(trainData);

DataSetIterator trainIter = new RecordReaderDataSetIterator.Builder(recordReader, 32)
    .classification(1, outputNum * labelLength)
    .build();

自定义 CaptchaLabelGenerator 用于从文件名提取每个字符的索引。

4. 构建 CNN 模型

MultiLayerConfiguration conf = new NeuralNetConfiguration.Builder()
    .updater(new Adam(0.001))
    .list()
    .layer(new ConvolutionLayer.Builder(3, 3).nIn(channels).nOut(32).activation(Activation.RELU).build())
    .layer(new SubsamplingLayer.Builder(PoolingType.MAX, new int[]{2,2}).build())
    .layer(new ConvolutionLayer.Builder(3, 3).nOut(64).activation(Activation.RELU).build())
    .layer(new SubsamplingLayer.Builder(PoolingType.MAX, new int[]{2,2}).build())
    .layer(new DenseLayer.Builder().nOut(256).activation(Activation.RELU).build())
    .layer(new OutputLayer.Builder(LossFunctions.LossFunction.MCXENT)
        .activation(Activation.SOFTMAX)
        .nOut(outputNum * labelLength).build())
    .setInputType(InputType.convolutionalFlat(height, width, channels))
    .build();

MultiLayerNetwork model = new MultiLayerNetwork(conf);
model.init();
model.setListeners(new ScoreIterationListener(10));

5. 训练模型

int epochs = 10;
for (int i = 0; i < epochs; i++) {
    model.fit(trainIter);
}

6. 测试验证码识别

NativeImageLoader loader = new NativeImageLoader(height, width, channels);
INDArray image = loader.asMatrix(new File("captcha_samples/7K2B_1.png"));
image = image.divi(255);

INDArray output = model.output(image);
int[] predictedIndices = output.argMax(1).toIntVector();

String[] chars = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ".split("");
StringBuilder sb = new StringBuilder();
for (int i = 0; i < labelLength; i++) {
    int idx = predictedIndices[i];
    sb.append(chars[idx % outputNum]);
}

System.out.println("Predicted: " + sb.toString());