[开源]ZXING的.NET版本源码解析

[概述]

ZXing ("zebra crossing") is an open-source, multi-format 1D/2D barcode image processing library implemented in Java, with ports to other languages.

开源地址：

https://github.com/zxing/zxing

[工程结构]

以ZXing.Net.Source.0.14.0.0版本为例，此文件目录下对应两个工程：

Base和WinMD，我们主要分析Base工程，其中：

ZXing.Net.Source.0.14.0.0BaseSourcelib目录下的工程为源码工程，zxing.vs2012为源码工程Solution文件；

ZXing.Net.Source.0.14.0.0BaseClientsWindowsFormsDemo目录下的工程为ZXING输出类库的应用工程，WindowsFormsDemo为应用工程Solution文件。

[应用工程分析]

WindowsFormsDemo有三个Tab，分别为Decoder/Encoder/WebCam，分别实现图片读码/二维码生成/网络摄像头采样读码（主要调用了avicap32.dll，它是Windows API应用程序接口相关模块，用于对摄像头和其它视频硬件进行AⅥ电影和视频的截取，详见工程文件WebCam.cs）。

Decoder(图片读码)：

private void btnStartDecoding_Click(object sender, EventArgs e)
      {
         var fileName = txtBarcodeImageFile.Text;
         if (!File.Exists(fileName))
         {
            MessageBox.Show(this, String.Format("File not found: {0}", fileName), "Error", MessageBoxButtons.OK,
                            MessageBoxIcon.Error);
            return;
         }

         using (var bitmap = (Bitmap)Bitmap.FromFile(fileName))
         {
            if (TryOnlyMultipleQRCodes)
               Decode(bitmap, TryMultipleBarcodes, new List<BarcodeFormat> { BarcodeFormat.QR_CODE });
            else
               Decode(bitmap, TryMultipleBarcodes, null);
         }
      }

      private void Decode(Bitmap image, bool tryMultipleBarcodes, IList<BarcodeFormat> possibleFormats)
      {
         resultPoints.Clear();
         lastResults.Clear();
         txtContent.Text = String.Empty;

         var timerStart = DateTime.Now.Ticks;
         Result[] results = null;
         barcodeReader.Options.PossibleFormats = possibleFormats;
         if (tryMultipleBarcodes)
            results = barcodeReader.DecodeMultiple(image);
         else
         {
            var result = barcodeReader.Decode(image);
            if (result != null)
            {
               results = new[] {result};
            }
         }
         var timerStop = DateTime.Now.Ticks;

         if (results == null)
         {
            txtContent.Text = "No barcode recognized";
         }
         labDuration.Text = new TimeSpan(timerStop - timerStart).Milliseconds.ToString("0 ms");

         if (results != null)
         {
            foreach (var result in results)
            {
               if (result.ResultPoints.Length > 0)
               {
                  var rect = new Rectangle((int) result.ResultPoints[0].X, (int) result.ResultPoints[0].Y, 1, 1);
                  foreach (var point in result.ResultPoints)
                  {
                     if (point.X < rect.Left)
                        rect = new Rectangle((int) point.X, rect.Y, rect.Width + rect.X - (int) point.X, rect.Height);
                     if (point.X > rect.Right)
                        rect = new Rectangle(rect.X, rect.Y, rect.Width + (int) point.X - rect.X, rect.Height);
                     if (point.Y < rect.Top)
                        rect = new Rectangle(rect.X, (int) point.Y, rect.Width, rect.Height + rect.Y - (int) point.Y);
                     if (point.Y > rect.Bottom)
                        rect = new Rectangle(rect.X, rect.Y, rect.Width, rect.Height + (int) point.Y - rect.Y);
                  }
                  using (var g = picBarcode.CreateGraphics())
                  {
                     g.DrawRectangle(Pens.Green, rect);
                  }
               }
            }
         }
      }

Encoder(二维码生成)：

（待续）

WebCam(网络摄像头采样读码)：

private void btnDecodeWebCam_Click(object sender, EventArgs e)
      {
         if (wCam == null)
         {
            wCam = new WebCam {Container = picWebCam};

            wCam.OpenConnection();

            webCamTimer = new Timer();
            webCamTimer.Tick += webCamTimer_Tick;
            webCamTimer.Interval = 200; // Image derivation interval
            webCamTimer.Start();

            btnDecodeWebCam.Text = "Decoding..."; // Update UI
         }
         else
         {
            webCamTimer.Stop();
            webCamTimer = null;
            wCam.Dispose();
            wCam = null;

            btnDecodeWebCam.Text = "Decode"; // Update UI
         }
      }

      void webCamTimer_Tick(object sender, EventArgs e)
      {
         var bitmap = wCam.GetCurrentImage(); // Derive a imaghe
         if (bitmap == null)
            return;
         Console.WriteLine("Bitmap width is:{0}, height is{1}. Camera is: {2} mega-pixel.", bitmap.Width, bitmap.Height, bitmap.Width* bitmap.Height/10000);
         var reader = new BarcodeReader();
         var result = reader.Decode(bitmap); // Decode the image
         if (result != null)
         {
            txtTypeWebCam.Text = result.BarcodeFormat.ToString();
            txtContentWebCam.Text = result.Text;
         }
      }

其中WebCam对象定义的各类对摄像头的参数设置和操作详见WebCam.cs。

[源码工程分析]

1.图像解码（Qrcode为例）

Qrcode解码流程为检测定位->解码，涉及的几个主要文件为：BarcodeReader.cs(createBinarizer)->BarcodeReaderGeneric.cs(createBinarizer)->HybridBinarizer.cs(createBinarizer)、QRCodeReader.cs，Detector.cs和FinderPatternFinder.cs，Decoder.cs。

HybridBinarizer.cs(createBinarizer)类实现位图的二值化处理，核心代码段为：

/// <summary>
      /// Calculates the final BitMatrix once for all requests. This could be called once from the
      /// constructor instead, but there are some advantages to doing it lazily, such as making
      /// profiling easier, and not doing heavy lifting when callers don't expect it.
      /// </summary>
      private void binarizeEntireImage()
      {
         if (matrix == null)
         {
            LuminanceSource source = LuminanceSource;
            int width = source.Width;
            int height = source.Height;
            if (width >= MINIMUM_DIMENSION && height >= MINIMUM_DIMENSION)
            {
               byte[] luminances = source.Matrix;

               int subWidth = width >> BLOCK_SIZE_POWER;
               if ((width & BLOCK_SIZE_MASK) != 0)
               {
                  subWidth++;
               }
               int subHeight = height >> BLOCK_SIZE_POWER;
               if ((height & BLOCK_SIZE_MASK) != 0)
               {
                  subHeight++;
               }
               int[][] blackPoints = calculateBlackPoints(luminances, subWidth, subHeight, width, height);

               var newMatrix = new BitMatrix(width, height);
               calculateThresholdForBlock(luminances, subWidth, subHeight, width, height, blackPoints, newMatrix);
               matrix = newMatrix;
            }
            else
            {
               // If the image is too small, fall back to the global histogram approach.
               matrix = base.BlackMatrix;
            }
         }
      }

      /// <summary>
      /// For each 8x8 block in the image, calculate the average black point using a 5x5 grid
      /// of the blocks around it. Also handles the corner cases (fractional blocks are computed based
      /// on the last 8 pixels in the row/column which are also used in the previous block).
      /// PS(Jay):This algrithm has big issue!!! Should be enhanced!!!
      /// </summary>
      /// <param name="luminances">The luminances.</param>
      /// <param name="subWidth">Width of the sub.</param>
      /// <param name="subHeight">Height of the sub.</param>
      /// <param name="width">The width.</param>
      /// <param name="height">The height.</param>
      /// <param name="blackPoints">The black points.</param>
      /// <param name="matrix">The matrix.</param>
      private static void calculateThresholdForBlock(byte[] luminances, int subWidth, int subHeight, int width, int height, int[][] blackPoints, BitMatrix matrix)
      {
         for (int y = 0; y < subHeight; y++)
         {
            int yoffset = y << BLOCK_SIZE_POWER;
            int maxYOffset = height - BLOCK_SIZE;
            if (yoffset > maxYOffset)
            {
               yoffset = maxYOffset;
            }
            for (int x = 0; x < subWidth; x++)
            {
               int xoffset = x << BLOCK_SIZE_POWER;
               int maxXOffset = width - BLOCK_SIZE;
               if (xoffset > maxXOffset)
               {
                  xoffset = maxXOffset;
               }
               int left = cap(x, 2, subWidth - 3);
               int top = cap(y, 2, subHeight - 3);
               int sum = 0;
               for (int z = -2; z <= 2; z++)
               {
                  int[] blackRow = blackPoints[top + z];
                  sum += blackRow[left - 2];
                  sum += blackRow[left - 1];
                  sum += blackRow[left];
                  sum += blackRow[left + 1];
                  sum += blackRow[left + 2];
               }
               int average = sum / 25;
               thresholdBlock(luminances, xoffset, yoffset, average, width, matrix);
            }
         }
      }

      private static int cap(int value, int min, int max)
      {
         return value < min ? min : value > max ? max : value;
      }

      /// <summary>
      /// Applies a single threshold to an 8x8 block of pixels.
      /// </summary>
      /// <param name="luminances">The luminances.</param>
      /// <param name="xoffset">The xoffset.</param>
      /// <param name="yoffset">The yoffset.</param>
      /// <param name="threshold">The threshold.</param>
      /// <param name="stride">The stride.</param>
      /// <param name="matrix">The matrix.</param>
      private static void thresholdBlock(byte[] luminances, int xoffset, int yoffset, int threshold, int stride, BitMatrix matrix)
      {
         int offset = (yoffset * stride) + xoffset;
         for (int y = 0; y < BLOCK_SIZE; y++, offset += stride)
         {
            for (int x = 0; x < BLOCK_SIZE; x++)
            {
               int pixel = luminances[offset + x] & 0xff;
               // Comparison needs to be <=, so that black == 0 pixels are black, even if the threshold is 0.
               matrix[xoffset + x, yoffset + y] = (pixel <= threshold);
            }
         }
      }

      /// <summary>
      /// Calculates a single black point for each 8x8 block of pixels and saves it away.
      /// See the following thread for a discussion of this algorithm:
      /// http://groups.google.com/group/zxing/browse_thread/thread/d06efa2c35a7ddc0
      /// </summary>
      /// <param name="luminances">The luminances.</param>
      /// <param name="subWidth">Width of the sub.</param>
      /// <param name="subHeight">Height of the sub.</param>
      /// <param name="width">The width.</param>
      /// <param name="height">The height.</param>
      /// <returns></returns>
      private static int[][] calculateBlackPoints(byte[] luminances, int subWidth, int subHeight, int width, int height)
      {
         int[][] blackPoints = new int[subHeight][];
         for (int i = 0; i < subHeight; i++)
         {
            blackPoints[i] = new int[subWidth];
         }

         for (int y = 0; y < subHeight; y++)
         {
            int yoffset = y << BLOCK_SIZE_POWER;
            int maxYOffset = height - BLOCK_SIZE;
            if (yoffset > maxYOffset)
            {
               yoffset = maxYOffset;
            }
            for (int x = 0; x < subWidth; x++)
            {
               int xoffset = x << BLOCK_SIZE_POWER;
               int maxXOffset = width - BLOCK_SIZE;
               if (xoffset > maxXOffset)
               {
                  xoffset = maxXOffset;
               }
               int sum = 0;
               int min = 0xFF;
               int max = 0;
               for (int yy = 0, offset = yoffset * width + xoffset; yy < BLOCK_SIZE; yy++, offset += width)
               {
                  for (int xx = 0; xx < BLOCK_SIZE; xx++)
                  {
                     int pixel = luminances[offset + xx] & 0xFF;
                     // still looking for good contrast
                     sum += pixel;
                     if (pixel < min)
                     {
                        min = pixel;
                     }
                     if (pixel > max)
                     {
                        max = pixel;
                     }
                  }
                  // short-circuit min/max tests once dynamic range is met
                  if (max - min > MIN_DYNAMIC_RANGE)
                  {
                     // finish the rest of the rows quickly
                     for (yy++, offset += width; yy < BLOCK_SIZE; yy++, offset += width)
                     {
                        for (int xx = 0; xx < BLOCK_SIZE; xx++)
                        {
                           sum += luminances[offset + xx] & 0xFF;
                        }
                     }
                  }
               }

               // The default estimate is the average of the values in the block.
               int average = sum >> (BLOCK_SIZE_POWER * 2);
               if (max - min <= MIN_DYNAMIC_RANGE)
               {
                  // If variation within the block is low, assume this is a block with only light or only
                  // dark pixels. In that case we do not want to use the average, as it would divide this
                  // low contrast area into black and white pixels, essentially creating data out of noise.
                  //
                  // The default assumption is that the block is light/background. Since no estimate for
                  // the level of dark pixels exists locally, use half the min for the block.
                  average = min >> 1;

                  if (y > 0 && x > 0)
                  {
                     // Correct the "white background" assumption for blocks that have neighbors by comparing
                     // the pixels in this block to the previously calculated black points. This is based on
                     // the fact that dark barcode symbology is always surrounded by some amount of light
                     // background for which reasonable black point estimates were made. The bp estimated at
                     // the boundaries is used for the interior.

                     // The (min < bp) is arbitrary but works better than other heuristics that were tried.
                     int averageNeighborBlackPoint = (blackPoints[y - 1][x] + (2 * blackPoints[y][x - 1]) +
                         blackPoints[y - 1][x - 1]) >> 2;
                     if (min < averageNeighborBlackPoint)
                     {
                        average = averageNeighborBlackPoint;
                     }
                  }
               }
               blackPoints[y][x] = average;
            }
         }
         return blackPoints;
      }

这一段算法有存在改进的必要。在HybridBinarizer继承的GlobalHistogramBinarizer类中，是从图像中均匀取5行（覆盖整个图像高度），每行取中间五分之四作为样本；以灰度值为X轴，每个灰度值的像素个数为Y轴建立一个直方图，从直方图中取点数最多的一个灰度值，然后再去给其他的灰度值进行分数计算，按照点数乘以与最多点数灰度值的距离的平方来进行打分，选分数最高的一个灰度值。接下来在这两个灰度值中间选取一个区分界限（这两个点灰度值大的是偏白色的点，灰度值小的是偏黑色的点），取的原则是尽量靠近灰度值大的点（偏白色的点）、并且要点数越少越好。界限有了以后就容易了，与整幅图像的每个点进行比较，如果灰度值比界限小的就是黑，在新的矩阵中将该点置1，其余的就是白，为0。此部分具体代码见GlobalHistogramBinarizer类的BlackMatrix()重写方法。这个算法的劣势是由于是全局计算阈值点，所以应对局部阴影不太理想（However, because it picks a global black point, it cannot handle difficult shadows and gradients.）。

QRCodeReader类实现了接口Reader，核心段代码为：

/// <summary>
      /// Locates and decodes a barcode in some format within an image. This method also accepts
      /// hints, each possibly associated to some data, which may help the implementation decode.
      /// </summary>
      /// <param name="image">image of barcode to decode</param>
      /// <param name="hints">passed as a <see cref="IDictionary{TKey, TValue}"/> from <see cref="DecodeHintType"/>
      /// to arbitrary data. The
      /// meaning of the data depends upon the hint type. The implementation may or may not do
      /// anything with these hints.</param>
      /// <returns>
      /// String which the barcode encodes
      /// </returns>
      public Result decode(BinaryBitmap image, IDictionary<DecodeHintType, object> hints)
      {
         DecoderResult decoderResult;
         ResultPoint[] points;
         if (image == null || image.BlackMatrix == null)
         {
            // something is wrong with the image
            return null;
         }
         if (hints != null && hints.ContainsKey(DecodeHintType.PURE_BARCODE)) // 纯barcode图片
         {
            var bits = extractPureBits(image.BlackMatrix);
            if (bits == null)
               return null;
            decoderResult = decoder.decode(bits, hints);
            points = NO_POINTS;
         }
         else
         {
            var detectorResult = new Detector(image.BlackMatrix).detect(hints); // 检测barcode
            if (detectorResult == null)
               return null;
            decoderResult = decoder.decode(detectorResult.Bits, hints); // 解码barcode
            points = detectorResult.Points;
         }
         if (decoderResult == null)
            return null;

         // If the code was mirrored: swap the bottom-left and the top-right points.
         var data = decoderResult.Other as QRCodeDecoderMetaData;
         if (data != null)
         {
            data.applyMirroredCorrection(points);
         }

         var result = new Result(decoderResult.Text, decoderResult.RawBytes, points, BarcodeFormat.QR_CODE);
         var byteSegments = decoderResult.ByteSegments;
         if (byteSegments != null)
         {
            result.putMetadata(ResultMetadataType.BYTE_SEGMENTS, byteSegments);
         }
         var ecLevel = decoderResult.ECLevel;
         if (ecLevel != null)
         {
            result.putMetadata(ResultMetadataType.ERROR_CORRECTION_LEVEL, ecLevel);
         }
         if (decoderResult.StructuredAppend)
         {
            result.putMetadata(ResultMetadataType.STRUCTURED_APPEND_SEQUENCE, decoderResult.StructuredAppendSequenceNumber);
            result.putMetadata(ResultMetadataType.STRUCTURED_APPEND_PARITY, decoderResult.StructuredAppendParity);
         }
         return result;
      }

qrcode->detector目录下的Detector类：

namespace ZXing.QrCode.Internal
{
   /// <summary>
   /// <p>Encapsulates logic that can detect a QR Code in an image, even if the QR Code
   /// is rotated or skewed, or partially obscured.</p>
   /// </summary>
   /// <author>Sean Owen</author>
   public class Detector
   {
      private readonly BitMatrix image;
      private ResultPointCallback resultPointCallback;

      /// <summary>
      /// Initializes a new instance of the <see cref="Detector"/> class.
      /// </summary>
      /// <param name="image">The image.</param>
      public Detector(BitMatrix image)
      {
         this.image = image;
      }

      /// <summary>
      /// Gets the image.
      /// </summary>
      virtual protected internal BitMatrix Image
      {
         get
         {
            return image;
         }
      }

      /// <summary>
      /// Gets the result point callback.
      /// </summary>
      virtual protected internal ResultPointCallback ResultPointCallback
      {
         get
         {
            return resultPointCallback;
         }
      }

      /// <summary>
      ///   <p>Detects a QR Code in an image, simply.</p>
      /// </summary>
      /// <returns>
      ///   <see cref="DetectorResult"/> encapsulating results of detecting a QR Code
      /// </returns>
      public virtual DetectorResult detect()
      {
         return detect(null);
      }

      /// <summary>
      ///   <p>Detects a QR Code in an image, simply.</p>
      /// </summary>
      /// <param name="hints">optional hints to detector</param>
      /// <returns>
      ///   <see cref="DetectorResult"/> encapsulating results of detecting a QR Code
      /// </returns>
      public virtual DetectorResult detect(IDictionary<DecodeHintType, object> hints)
      {
         resultPointCallback = hints == null || !hints.ContainsKey(DecodeHintType.NEED_RESULT_POINT_CALLBACK) ? null : (ResultPointCallback)hints[DecodeHintType.NEED_RESULT_POINT_CALLBACK];

         FinderPatternFinder finder = new FinderPatternFinder(image, resultPointCallback);
         FinderPatternInfo info = finder.find(hints);
         if (info == null)
            return null;

         return processFinderPatternInfo(info);
      }

      /// <summary>
      /// Processes the finder pattern info.
      /// </summary>
      /// <param name="info">The info.</param>
      /// <returns></returns>
      protected internal virtual DetectorResult processFinderPatternInfo(FinderPatternInfo info)
      {
         FinderPattern topLeft = info.TopLeft;
         FinderPattern topRight = info.TopRight;
         FinderPattern bottomLeft = info.BottomLeft;

         float moduleSize = calculateModuleSize(topLeft, topRight, bottomLeft);
         if (moduleSize < 1.0f)
         {
            return null;
         }
         int dimension;
         if (!computeDimension(topLeft, topRight, bottomLeft, moduleSize, out dimension))
            return null;
         Internal.Version provisionalVersion = Internal.Version.getProvisionalVersionForDimension(dimension);
         if (provisionalVersion == null)
            return null;
         int modulesBetweenFPCenters = provisionalVersion.DimensionForVersion - 7;

         AlignmentPattern alignmentPattern = null;
         // Anything above version 1 has an alignment pattern
         if (provisionalVersion.AlignmentPatternCenters.Length > 0)
         {

            // Guess where a "bottom right" finder pattern would have been
            float bottomRightX = topRight.X - topLeft.X + bottomLeft.X;
            float bottomRightY = topRight.Y - topLeft.Y + bottomLeft.Y;

            // Estimate that alignment pattern is closer by 3 modules
            // from "bottom right" to known top left location
            //UPGRADE_WARNING: Data types in Visual C# might be different.  Verify the accuracy of narrowing conversions. "ms-help://MS.VSCC.v80/dv_commoner/local/redirect.htm?index='!DefaultContextWindowIndex'&keyword='jlca1042'"
            float correctionToTopLeft = 1.0f - 3.0f / (float)modulesBetweenFPCenters;
            //UPGRADE_WARNING: Data types in Visual C# might be different.  Verify the accuracy of narrowing conversions. "ms-help://MS.VSCC.v80/dv_commoner/local/redirect.htm?index='!DefaultContextWindowIndex'&keyword='jlca1042'"
            int estAlignmentX = (int)(topLeft.X + correctionToTopLeft * (bottomRightX - topLeft.X));
            //UPGRADE_WARNING: Data types in Visual C# might be different.  Verify the accuracy of narrowing conversions. "ms-help://MS.VSCC.v80/dv_commoner/local/redirect.htm?index='!DefaultContextWindowIndex'&keyword='jlca1042'"
            int estAlignmentY = (int)(topLeft.Y + correctionToTopLeft * (bottomRightY - topLeft.Y));

            // Kind of arbitrary -- expand search radius before giving up
            for (int i = 4; i <= 16; i <<= 1)
            {
               alignmentPattern = findAlignmentInRegion(moduleSize, estAlignmentX, estAlignmentY, (float)i);
               if (alignmentPattern == null)
                  continue;
               break;
            }
            // If we didn't find alignment pattern... well try anyway without it
         }

         PerspectiveTransform transform = createTransform(topLeft, topRight, bottomLeft, alignmentPattern, dimension);

         BitMatrix bits = sampleGrid(image, transform, dimension);
         if (bits == null)
            return null;

         ResultPoint[] points;
         if (alignmentPattern == null)
         {
            points = new ResultPoint[] { bottomLeft, topLeft, topRight };
         }
         else
         {
            points = new ResultPoint[] { bottomLeft, topLeft, topRight, alignmentPattern };
         }
         return new DetectorResult(bits, points);
      }

      private static PerspectiveTransform createTransform(ResultPoint topLeft, ResultPoint topRight, ResultPoint bottomLeft, ResultPoint alignmentPattern, int dimension)
      {
         //UPGRADE_WARNING: Data types in Visual C# might be different.  Verify the accuracy of narrowing conversions. "ms-help://MS.VSCC.v80/dv_commoner/local/redirect.htm?index='!DefaultContextWindowIndex'&keyword='jlca1042'"
         float dimMinusThree = (float)dimension - 3.5f;
         float bottomRightX;
         float bottomRightY;
         float sourceBottomRightX;
         float sourceBottomRightY;
         if (alignmentPattern != null)
         {
            bottomRightX = alignmentPattern.X;
            bottomRightY = alignmentPattern.Y;
            sourceBottomRightX = sourceBottomRightY = dimMinusThree - 3.0f;
         }
         else
         {
            // Don't have an alignment pattern, just make up the bottom-right point
            bottomRightX = (topRight.X - topLeft.X) + bottomLeft.X;
            bottomRightY = (topRight.Y - topLeft.Y) + bottomLeft.Y;
            sourceBottomRightX = sourceBottomRightY = dimMinusThree;
         }

         return PerspectiveTransform.quadrilateralToQuadrilateral(
            3.5f,
            3.5f,
            dimMinusThree,
            3.5f,
            sourceBottomRightX,
            sourceBottomRightY,
            3.5f,
            dimMinusThree,
            topLeft.X,
            topLeft.Y,
            topRight.X,
            topRight.Y,
            bottomRightX,
            bottomRightY,
            bottomLeft.X,
            bottomLeft.Y);
      }

      private static BitMatrix sampleGrid(BitMatrix image, PerspectiveTransform transform, int dimension)
      {
         GridSampler sampler = GridSampler.Instance;
         return sampler.sampleGrid(image, dimension, dimension, transform);
      }

      /// <summary> <p>Computes the dimension (number of modules on a size) of the QR Code based on the position
      /// of the finder patterns and estimated module size.</p>
      /// </summary>
      private static bool computeDimension(ResultPoint topLeft, ResultPoint topRight, ResultPoint bottomLeft, float moduleSize, out int dimension)
      {
         int tltrCentersDimension = MathUtils.round(ResultPoint.distance(topLeft, topRight) / moduleSize);
         int tlblCentersDimension = MathUtils.round(ResultPoint.distance(topLeft, bottomLeft) / moduleSize);
         dimension = ((tltrCentersDimension + tlblCentersDimension) >> 1) + 7;
         switch (dimension & 0x03)
         {
            // mod 4
            case 0:
               dimension++;
               break;
            // 1? do nothing
            case 2:
               dimension--;
               break;
            case 3:
               return true;
         }
         return true;
      }

      /// <summary> <p>Computes an average estimated module size based on estimated derived from the positions
      /// of the three finder patterns.</p>
      /// </summary>
      protected internal virtual float calculateModuleSize(ResultPoint topLeft, ResultPoint topRight, ResultPoint bottomLeft)
      {
         // Take the average
         return (calculateModuleSizeOneWay(topLeft, topRight) + calculateModuleSizeOneWay(topLeft, bottomLeft)) / 2.0f;
      }

      /// <summary> <p>Estimates module size based on two finder patterns -- it uses
      /// {@link #sizeOfBlackWhiteBlackRunBothWays(int, int, int, int)} to figure the
      /// width of each, measuring along the axis between their centers.</p>
      /// </summary>
      private float calculateModuleSizeOneWay(ResultPoint pattern, ResultPoint otherPattern)
      {
         //UPGRADE_WARNING: Data types in Visual C# might be different.  Verify the accuracy of narrowing conversions. "ms-help://MS.VSCC.v80/dv_commoner/local/redirect.htm?index='!DefaultContextWindowIndex'&keyword='jlca1042'"
         float moduleSizeEst1 = sizeOfBlackWhiteBlackRunBothWays((int)pattern.X, (int)pattern.Y, (int)otherPattern.X, (int)otherPattern.Y);
         //UPGRADE_WARNING: Data types in Visual C# might be different.  Verify the accuracy of narrowing conversions. "ms-help://MS.VSCC.v80/dv_commoner/local/redirect.htm?index='!DefaultContextWindowIndex'&keyword='jlca1042'"
         float moduleSizeEst2 = sizeOfBlackWhiteBlackRunBothWays((int)otherPattern.X, (int)otherPattern.Y, (int)pattern.X, (int)pattern.Y);
         if (Single.IsNaN(moduleSizeEst1))
         {
            return moduleSizeEst2 / 7.0f;
         }
         if (Single.IsNaN(moduleSizeEst2))
         {
            return moduleSizeEst1 / 7.0f;
         }
         // Average them, and divide by 7 since we've counted the width of 3 black modules,
         // and 1 white and 1 black module on either side. Ergo, divide sum by 14.
         return (moduleSizeEst1 + moduleSizeEst2) / 14.0f;
      }

      /// <summary> See {@link #sizeOfBlackWhiteBlackRun(int, int, int, int)}; computes the total width of
      /// a finder pattern by looking for a black-white-black run from the center in the direction
      /// of another point (another finder pattern center), and in the opposite direction too.
      /// </summary>
      private float sizeOfBlackWhiteBlackRunBothWays(int fromX, int fromY, int toX, int toY)
      {

         float result = sizeOfBlackWhiteBlackRun(fromX, fromY, toX, toY);

         // Now count other way -- don't run off image though of course
         float scale = 1.0f;
         int otherToX = fromX - (toX - fromX);
         if (otherToX < 0)
         {
            //UPGRADE_WARNING: Data types in Visual C# might be different.  Verify the accuracy of narrowing conversions. "ms-help://MS.VSCC.v80/dv_commoner/local/redirect.htm?index='!DefaultContextWindowIndex'&keyword='jlca1042'"
            scale = (float)fromX / (float)(fromX - otherToX);
            otherToX = 0;
         }
         else if (otherToX >= image.Width)
         {
            //UPGRADE_WARNING: Data types in Visual C# might be different.  Verify the accuracy of narrowing conversions. "ms-help://MS.VSCC.v80/dv_commoner/local/redirect.htm?index='!DefaultContextWindowIndex'&keyword='jlca1042'"
            scale = (float)(image.Width - 1 - fromX) / (float)(otherToX - fromX);
            otherToX = image.Width - 1;
         }
         //UPGRADE_WARNING: Data types in Visual C# might be different.  Verify the accuracy of narrowing conversions. "ms-help://MS.VSCC.v80/dv_commoner/local/redirect.htm?index='!DefaultContextWindowIndex'&keyword='jlca1042'"
         int otherToY = (int)(fromY - (toY - fromY) * scale);

         scale = 1.0f;
         if (otherToY < 0)
         {
            //UPGRADE_WARNING: Data types in Visual C# might be different.  Verify the accuracy of narrowing conversions. "ms-help://MS.VSCC.v80/dv_commoner/local/redirect.htm?index='!DefaultContextWindowIndex'&keyword='jlca1042'"
            scale = (float)fromY / (float)(fromY - otherToY);
            otherToY = 0;
         }
         else if (otherToY >= image.Height)
         {
            //UPGRADE_WARNING: Data types in Visual C# might be different.  Verify the accuracy of narrowing conversions. "ms-help://MS.VSCC.v80/dv_commoner/local/redirect.htm?index='!DefaultContextWindowIndex'&keyword='jlca1042'"
            scale = (float)(image.Height - 1 - fromY) / (float)(otherToY - fromY);
            otherToY = image.Height - 1;
         }
         //UPGRADE_WARNING: Data types in Visual C# might be different.  Verify the accuracy of narrowing conversions. "ms-help://MS.VSCC.v80/dv_commoner/local/redirect.htm?index='!DefaultContextWindowIndex'&keyword='jlca1042'"
         otherToX = (int)(fromX + (otherToX - fromX) * scale);

         result += sizeOfBlackWhiteBlackRun(fromX, fromY, otherToX, otherToY);
         return result - 1.0f; // -1 because we counted the middle pixel twice
      }

      /// <summary> <p>This method traces a line from a point in the image, in the direction towards another point.
      /// It begins in a black region, and keeps going until it finds white, then black, then white again.
      /// It reports the distance from the start to this point.</p>
      /// 
      /// <p>This is used when figuring out how wide a finder pattern is, when the finder pattern
      /// may be skewed or rotated.</p>
      /// </summary>
      private float sizeOfBlackWhiteBlackRun(int fromX, int fromY, int toX, int toY)
      {
         // Mild variant of Bresenham's algorithm;
         // see http://en.wikipedia.org/wiki/Bresenham's_line_algorithm
         bool steep = Math.Abs(toY - fromY) > Math.Abs(toX - fromX);
         if (steep)
         {
            int temp = fromX;
            fromX = fromY;
            fromY = temp;
            temp = toX;
            toX = toY;
            toY = temp;
         }

         int dx = Math.Abs(toX - fromX);
         int dy = Math.Abs(toY - fromY);
         int error = -dx >> 1;
         int xstep = fromX < toX ? 1 : -1;
         int ystep = fromY < toY ? 1 : -1;

         // In black pixels, looking for white, first or second time.
         int state = 0;
         // Loop up until x == toX, but not beyond
         int xLimit = toX + xstep;
         for (int x = fromX, y = fromY; x != xLimit; x += xstep)
         {
            int realX = steep ? y : x;
            int realY = steep ? x : y;

            // Does current pixel mean we have moved white to black or vice versa?
            // Scanning black in state 0,2 and white in state 1, so if we find the wrong
            // color, advance to next state or end if we are in state 2 already
            if ((state == 1) == image[realX, realY])
            {
               if (state == 2)
               {
                  return MathUtils.distance(x, y, fromX, fromY);
               }
               state++;
            }
            error += dy;
            if (error > 0)
            {
               if (y == toY)
               {


                  break;
               }
               y += ystep;
               error -= dx;
            }
         }
         // Found black-white-black; give the benefit of the doubt that the next pixel outside the image
         // is "white" so this last point at (toX+xStep,toY) is the right ending. This is really a
         // small approximation; (toX+xStep,toY+yStep) might be really correct. Ignore this.
         if (state == 2)
         {
            return MathUtils.distance(toX + xstep, toY, fromX, fromY);
         }
         // else we didn't find even black-white-black; no estimate is really possible
         return Single.NaN;

      }

      /// <summary>
      ///   <p>Attempts to locate an alignment pattern in a limited region of the image, which is
      /// guessed to contain it. This method uses {@link AlignmentPattern}.</p>
      /// </summary>
      /// <param name="overallEstModuleSize">estimated module size so far</param>
      /// <param name="estAlignmentX">x coordinate of center of area probably containing alignment pattern</param>
      /// <param name="estAlignmentY">y coordinate of above</param>
      /// <param name="allowanceFactor">number of pixels in all directions to search from the center</param>
      /// <returns>
      ///   <see cref="AlignmentPattern"/> if found, or null otherwise
      /// </returns>
      protected AlignmentPattern findAlignmentInRegion(float overallEstModuleSize, int estAlignmentX, int estAlignmentY, float allowanceFactor)
      {
         // Look for an alignment pattern (3 modules in size) around where it
         // should be
         //UPGRADE_WARNING: Data types in Visual C# might be different.  Verify the accuracy of narrowing conversions. "ms-help://MS.VSCC.v80/dv_commoner/local/redirect.htm?index='!DefaultContextWindowIndex'&keyword='jlca1042'"
         int allowance = (int)(allowanceFactor * overallEstModuleSize);
         int alignmentAreaLeftX = Math.Max(0, estAlignmentX - allowance);
         int alignmentAreaRightX = Math.Min(image.Width - 1, estAlignmentX + allowance);
         if (alignmentAreaRightX - alignmentAreaLeftX < overallEstModuleSize * 3)
         {
            return null;
         }

         int alignmentAreaTopY = Math.Max(0, estAlignmentY - allowance);
         int alignmentAreaBottomY = Math.Min(image.Height - 1, estAlignmentY + allowance);

         var alignmentFinder = new AlignmentPatternFinder(
            image,
            alignmentAreaLeftX,
            alignmentAreaTopY,
            alignmentAreaRightX - alignmentAreaLeftX,
            alignmentAreaBottomY - alignmentAreaTopY,
            overallEstModuleSize,
            resultPointCallback);

         return alignmentFinder.find();
      }
   }
}

qrcode->detector目录下的FinderPatternFinder类：