[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.0.14.0.0\Base\Source\lib目录下的工程为源码工程，zxing.vs2012为源码工程Solution文件；
ZXing.Net.Source.0.14.0.0\Base\Clients\WindowsFormsDemo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：
