美文网首页
Android扫码优化(二)-openCV+ZXing+Zbar

Android扫码优化(二)-openCV+ZXing+Zbar

作者: A代码搬运工 | 来源:发表于2019-07-10 07:51 被阅读0次

    接上篇文章Android扫码优化(一)-openCV 官方Demo配置,只是简单讲解了在Android Studio怎么配置openCV sdk Demo。对于上文提到的扫码不好扫的特殊场景,该怎么优化呢?

    相机优化

    二维码解码的数据是通过硬件Camera获取到的,相机的优化策略非常重要。

    相机一般优化设置:

    • 选择最佳预览尺寸/图片尺寸
    • 设置适合的相机放大倍数
    • 调整聚焦时间
    • 设置自动对焦区域
    • 调整合理扫描区域

    但相机的硬件属性咱们改变不了,比如低端机器的Camera性能不行,按照以上策略再怎么优化也不行。设备性能好,聚焦快,扫码速度就快。相机做了一般处理,解码优化就很关键了。

    解码优化

    ZXing解码
    • 将Camera预览的byte数据转换成YUV数据
    • 将转换后的YUV数据转成BinaryBitmap
    • 通过预先设置好的解码器对BinaryBitmap进行解码
            byte[] rotatedData = new byte[data.length];
            for (int y = 0; y < height; y++) {
                for (int x = 0; x < width; x++)
                    rotatedData[x * height + height - y - 1] = data[x + y * width];
            }
            int tmp = width; // Here we are swapping, that's the difference to #11
            width = height;
            height = tmp;
    
            PlanarYUVLuminanceSource source = CameraManager.get().buildLuminanceSource(rotatedData, width, height);
            BinaryBitmap bitmap = new BinaryBitmap(new HybridBinarizer(source));
            try {
                rawResult = multiFormatReader.decodeWithState(bitmap);
            } catch (ReaderException re) {
            } finally {
                multiFormatReader.reset();
            }
    
    ZBar解码

    相对来说简单,直接将Camera预览数据转换成定义好的Image对象,通过预先设置的ImageScanner对Image进行解码。

            Image image = new Image(width, height, "Y800");
            image.setData(data);
    
            int result = scanner.scanImage(image);
    
            if (result != 0) {
                SymbolSet syms = scanner.getResults();
                for (Symbol sym : syms) {
                    return sym.getData();
                }
            }
    

    针对以上的ZXing和ZBar解码常见的优化如下:

    • 减少解码方式

      将multiFormatReader中不需要的解码器直接剔除

    • 解码算法优化

      ZBar比ZXing快,Zbar直接通过JNI调取C代码解码

    • 减少解码数据

      解码预览数据 不选择整个屏幕而是取扫码框区域

    • 解码库Zxing和Zbar结合

    OpenCV
    • 通过openCV将预览框的PlanarYUVLuminanceSource转换成RGB截取

      void getCropRect(unsigned char *nv21, int width, int height,
                       unsigned char *dest, int rectLeft, int rectTop, int rectWidth, int rectHeight) {
          Mat imgMat(height * 3 / 2, width, CV_8UC1, nv21);
          Mat rgbMat(height, width, CV_8UC3);
          Mat resultMat(rectHeight * 3 / 2, rectWidth, CV_8UC1, dest);
      
          cvtColor(imgMat, rgbMat, CV_YUV2RGB_NV21);
          Rect cropRect(rectLeft, rectTop, rectWidth, rectHeight);
          cvtColor(rgbMat(cropRect), resultMat, CV_RGB2YUV_I420);
      }
      
    • 通过openCV对预览数据进行降噪

          cvtColor(imgMat, innerMat, CV_YUV2GRAY_NV21);
          medianBlur(innerMat, innerMat, 7);
      
    • 缩小二值化阀值计算范围

      灰色二维码之所以难以识别是因为背景的色调过暗,拉低了二值化阈值的计算,导致整个二维码的难以识别。通过将二值化阈值的计算区域缩小到预览框的2/5,然后再通过计算的阈值对整个预览区域的数据进行二值化。

        int rectWidth = width / 5;
          int rectHeight = height / 5;
          int centerX = width / 2, centerY = height / 2;
      
          Rect cropLTRect(centerX - rectWidth, centerY - rectHeight, rectWidth, rectHeight);
          Rect cropRTRect(centerX, centerY - rectHeight, rectWidth, rectHeight);
          Rect cropLBRect(centerX - rectWidth, centerY, rectWidth, rectHeight);
          Rect cropRBRect(centerX, centerY, rectWidth, rectHeight);
      
          double threshLT = threshold(innerMat(cropLTRect), binMat(cropLTRect), 0, 255, CV_THRESH_OTSU);
          double threshRT = threshold(innerMat(cropRTRect), binMat(cropRTRect), 0, 255, CV_THRESH_OTSU);
          double threshLB = threshold(innerMat(cropLBRect), binMat(cropLBRect), 0, 255, CV_THRESH_OTSU);
          double threshRB = threshold(innerMat(cropRBRect), binMat(cropRBRect), 0, 255, CV_THRESH_OTSU);
      
    • 分块对预览区进行二值化

      分块进行计算阈值,是考虑到不同区域的亮度是不同的,如果整体进行计算的话,会丢失部分有效信息。而之所以选取靠近中心的小部分进行阈值化计算,是因为用户行为通常会将二维码对准预览框的中心,因此中心部分,包含有效亮度信息更为精确,减少背景亮度对整体二值化的影响。

          Rect LTRect(0, 0, centerX, centerY);
          Rect RTRect(centerX - 1, 0, centerX, centerY);
          Rect LBRect(0, centerY - 1, centerX, centerY);
          Rect RBRect(centerX - 1, centerY - 1, centerX, centerY);
      
          threshold(innerMat(LTRect), innerMat(LTRect), threshLT, 255, CV_THRESH_BINARY);
          threshold(innerMat(RTRect), innerMat(RTRect), threshRT, 255, CV_THRESH_BINARY);
          threshold(innerMat(LBRect), innerMat(LBRect), threshLB, 255, CV_THRESH_BINARY);
          threshold(innerMat(RBRect), innerMat(RBRect), threshRB, 255, CV_THRESH_BINARY);
      

    最终就形成了openCV+Zxing+ZBar的解码策略:

         // 1.使用Zbar
            Result rawResult = zbarDecode(processSrc,processWidth,processWidth);
    
            //2.使用ZXing
            if (rawResult == null) {
                rawResult = zxingDecode(processSrc,processWidth,processHeight);
            }
    
            if (rawResult == null) {
                // OpenCV预处理
                byte[] processData = new byte[processWidth * processHeight * 3 / 2];
                ImagePreProcess.preProcess(processSrc, processWidth, processHeight, processData);
    
                //3. opencv+Zbar
                rawResult = zbarDecode(processSrc,processWidth,processHeight);
    
                //4. opencv+Zxing
                if (rawResult == null) {
                    rawResult = zxingDecode(processSrc,processWidth,processHeight);
                }
            }
    
    

    总结

    经过上述opencv预处理尝试之后,大大提升了特殊情景的二维码识别。

    GitHub地址:https://github.com/gongchaobin/ScanMaster

    参考文章

    Android二维码扫描优化:http://blog.jostey.com/2018/04/27/Android二维码扫描优化/

    二维码扫码优化-字节跳动: https://zhuanlan.zhihu.com/p/44845942

    知乎扫码优化专栏:https://www.zhihu.com/question/53129607

    相关文章

      网友评论

          本文标题:Android扫码优化(二)-openCV+ZXing+Zbar

          本文链接:https://www.haomeiwen.com/subject/dxzqkctx.html