Data Matrix扫描过慢：Java应用扫码性能优化求助

我正在开发一款Java应用，需扫描Data Matrix二维码以提取并展示其中的XML数据（对应德国医保处方BMP）。测试使用某文档中的示例码，运行设备为Microsoft Surface Go 4 Business，后置摄像头参数：800万像素、1080P高清视频，日志为德语。当前实现代码如下：

import java.util.concurrent.BlockingQueue;
import java.util.concurrent.LinkedBlockingQueue;

import com.github.sarxos.webcam.Webcam;
import com.github.sarxos.webcam.WebcamPanel;
import com.github.sarxos.webcam.WebcamResolution;
import com.google.zxing.*;
import com.google.zxing.common.HybridBinarizer;
import com.google.zxing.client.j2se.BufferedImageLuminanceSource;

import javax.swing.*;
import java.awt.*;
import java.awt.Dimension;
import java.awt.image.BufferedImage;
import java.util.concurrent.Executor;
import java.util.concurrent.Executors;
import java.util.concurrent.ThreadFactory;
import java.util.logging.Logger;

public class QRCodeScanner extends JFrame implements Runnable, ThreadFactory {

    private static final Logger LOGGER = Logger.getLogger(QRCodeScanner.class.getName());
    
    private static final long serialVersionUID = 1L;

    private BlockingQueue<String> decodedResultsQueue = new LinkedBlockingQueue<>();

    
    private Executor executor = Executors.newSingleThreadExecutor(this);
    private Webcam webcam = null;
    private WebcamPanel panel = null;
    private JTextArea textarea = null;
    private JScrollPane scrollPane = null;

    public QRCodeScanner() {
        super();

        setLayout(new BorderLayout());
        setTitle("QR Code Scanner");
        setDefaultCloseOperation(JFrame.EXIT_ON_CLOSE);

        Dimension size = WebcamResolution.VGA.getSize();

        webcam = Webcam.getWebcams().get(1);
        webcam.setViewSize(size);

        panel = new WebcamPanel(webcam);
        panel.setPreferredSize(size);

        textarea = new JTextArea();
        textarea.setEditable(false);
        scrollPane = new JScrollPane(textarea);
        scrollPane.setPreferredSize(size);

        add(panel, BorderLayout.WEST);
        add(scrollPane, BorderLayout.CENTER);

        pack();
        setVisible(true);

        executor.execute(this);
    }

    @Override
    public void run() {
        do {
            try {
                Thread.sleep(100);
            } catch (InterruptedException e) {
                LOGGER.severe("Thread interrupted: " + e.getMessage());
            }

            BufferedImage image = null;

            if (webcam.isOpen()) {
                LOGGER.info("Webcam is open.");
                if ((image = webcam.getImage()) == null) {
                    LOGGER.warning("Webcam image is null.");
                    continue;
                }
            } else {
                LOGGER.warning("Webcam is not open.");
            }

            Result result = processImageAndDecodeQRCode(image);

            if (result != null) {
                if (result.getBarcodeFormat() == BarcodeFormat.UPC_E) {
                    LOGGER.info("Detected UPC_E format, rescanning...");
                    continue;
                }
                result.getBarcodeFormat();
                LOGGER.info("QR code format: " + result.getBarcodeFormat());
                result.getNumBits();
                LOGGER.info("QR code num bits: " + result.getNumBits());
                result.getResultMetadata();
                LOGGER.info("QR code result metadata: " + result.getResultMetadata());
                result.getRawBytes();
                LOGGER.info("QR code raw bytes: " + result.getRawBytes());
                String text = result.getText();
                LOGGER.info("QR code text: " + text);
                decodedResultsQueue.add(text);
            }
        } while (true);
    }

    private Result processImageAndDecodeQRCode(BufferedImage image) {
        // Preprocess image: convert to grayscale and apply thresholding
        BufferedImage grayscaleImage = new BufferedImage(
            image.getWidth(), image.getHeight(), BufferedImage.TYPE_BYTE_GRAY);
        Graphics g = grayscaleImage.getGraphics();
        g.drawImage(image, 0, 0, null);
        g.dispose();

        LuminanceSource source = new BufferedImageLuminanceSource(grayscaleImage);
        BinaryBitmap bitmap = new BinaryBitmap(new HybridBinarizer(source));

        Result result = null;
        try {
            result = new MultiFormatReader().decode(bitmap);
        } catch (NotFoundException e) {
            LOGGER.warning("No QR code found.");
            e.getCause();
        }

        return result;
    }

    public void startUIUpdateThread() {
        new Thread(() -> {
            while (true) {
                try {
                    // Take the next decoded result from the queue and update the UI
                    String text = decodedResultsQueue.take();
                    textarea.append(text + "\n");
                } catch (InterruptedException e) {
                    LOGGER.severe("UI update thread interrupted: " + e.getMessage());
                }
            }
        }).start();
    }
    
    
    
    @Override
    public Thread newThread(Runnable r) {
        Thread t = new Thread(r, "example-runner");
        t.setDaemon(true);
        return t;
    }

    public static void main(String[] args) {
        QRCodeScanner scanner = new QRCodeScanner();
        scanner.startUIUpdateThread();  // Start the UI update thread
    }
}

在线生成的Data Matrix二维码扫描效果良好，但扫描PDF中的示例码时常耗时过长甚至失败，希望优化扫描性能，达到应用商店扫码工具的水平，寻求计算机视觉方向的优化方案或高效实现思路。

一、Zxing解码精准配置优化 #

• 指定解码格式，减少无效检测：当前MultiFormatReader会遍历所有条码格式，浪费计算资源。明确只扫描DataMatrix：

  Map<DecodeHintType, Object> hints = new HashMap<>();
  hints.put(DecodeHintType.POSSIBLE_FORMATS, Collections.singletonList(BarcodeFormat.DATA_MATRIX));
  result = new MultiFormatReader().decode(bitmap, hints);
  

• 替换二值化策略：HybridBinarizer在低对比度（如PDF截图）场景表现差，改用GlobalHistogramBinarizer或自定义局部阈值处理：

  BinaryBitmap bitmap = new BinaryBitmap(new GlobalHistogramBinarizer(source));
  

• 多尺度解码重试：对同一图像进行缩放（0.8x、1.2x等）后重试，覆盖不同大小的Data Matrix码：

  List<BufferedImage> scaledImages = Arrays.asList(
      scaleImage(image, 0.8f),
      image,
      scaleImage(image, 1.2f)
  );
  for (BufferedImage img : scaledImages) {
      // 执行解码逻辑
      if (result != null) break;
  }
  

  其中scaleImage可通过AffineTransform实现图像缩放。

二、图像预处理优化 #

• 针对性降噪：PDF二维码易有扫描噪声，用高斯模糊或中值滤波预处理：

  BufferedImage blurred = new BufferedImage(image.getWidth(), image.getHeight(), BufferedImage.TYPE_INT_ARGB);
  Graphics2D g2d = blurred.createGraphics();
  g2d.setRenderingHint(RenderingHints.KEY_INTERPOLATION, RenderingHints.VALUE_INTERPOLATION_BILINEAR);
  g2d.setRenderingHint(RenderingHints.KEY_RENDERING, RenderingHints.VALUE_RENDER_QUALITY);
  g2d.drawImage(image, 0, 0, null);
  g2d.dispose();
  

• 增强对比度：调整亮度和对比度，强化二维码与背景差异：

  RescaleOp rescaleOp = new RescaleOp(1.5f, -50, null); // 1.5为对比度系数，-50为亮度偏移
  BufferedImage enhanced = rescaleOp.filter(grayscaleImage, null);
  

• 裁剪感兴趣区域（ROI）：通过边缘检测、轮廓识别定位二维码区域，仅对该区域解码，减少处理范围：

  // 示例：用Canny边缘检测找到高对比度轮廓，筛选符合Data Matrix比例的区域后裁剪
  

三、性能与线程优化 #

• 降低采集分辨率：当前用VGA（640x480），可尝试480x360减少单帧计算量；若需高分辨率，先缩放再处理。
• 调整扫描帧率：Thread.sleep(100)对应10fps，可调整为15fps（sleep 66ms），或添加帧跳过逻辑（每2帧处理一次），避免重复计算。
• 异步解码优化：用多线程池并行处理多帧/多尺度图像，注意线程安全，限制并发数避免CPU过载。

四、替代方案与硬件加速 #

• 尝试性能更优的库：如通过JNI调用ZXing-CPP，利用C++的性能优势；或集成Google ML Kit条码扫描API，其针对移动设备做了深度优化。
• 集成OpenCV：借助OpenCV的专业算法（阈值处理、透视变换、轮廓检测）处理畸变、模糊的PDF二维码，提升识别率。

内容的提问来源于stack exchange，提问作者SmiSmiSmi42