无论是跨境电商还是制造业分拣设备，在包裹流转出入库的场景，为了保证包裹分拣计划和测量数据绑定真实性，经常会遇到面单扣取的需求，下面我就通过两种实现原理来实现这一功能。
一:OpenCVSharp 通过面单轮廓/颜色/边缘等组合检测实现
二:通过OCR识别面单内容,根据所有切割点坐标点最小外界矩形来定位面单位置(扣面单的场景需求是看清面单内容，当然想要扣取完整面单图片，可以添加面单尺寸,规则信息等维度计算或者直接用第三种方式)
三:YOLO+Labelme标定工具，通过模型训练定位扣取(这个抽时间单独展开一篇解释)
方式一:OpencvSharp 通过轮廓/颜色/边缘检测
这种方式对于包裹和面单颜色有明显差异的场景很友好，对于包裹颜色和面单颜色接近的效果一般（建议考虑第二种方式），虽然可以根据面单样式或者文字聚集密度等多重维度来组合分析，但是过于复杂，并且定制化程度很高，废话少说，先看看效果:
原图：
通过显示增强后的效果图:
废话少说，附上核心代码:- staticvoidProcessSingleImage(string imagePath){if (!File.Exists(imagePath)) { Console.WriteLine("文件不存在!"); Console.ReadKey();return; }try { Console.WriteLine($"处理: {Path.GetFileName(imagePath)}");var stopwatch = Stopwatch.StartNew();// 检测面单var results = _detector.DetectLabels(imagePath); stopwatch.Stop(); Console.WriteLine($"检测耗时: {stopwatch.ElapsedMilliseconds}ms"); Console.WriteLine($"找到 {results.Count} 个面单区域");if (results.Count == 0) { Console.WriteLine("未检测到面单!"); Console.ReadKey();return; }// 显示结果foreach (var result in results) { Console.WriteLine($"- {result.DetectionMethod}: 置信度 {result.Confidence:F2}, " +$"位置 [{result.BoundingBox.X}, {result.BoundingBox.Y}, " +$"{result.BoundingBox.Width}, {result.BoundingBox.Height}]"); }// 创建输出目录var outputDir = _config.OutputDirectory;if (!Directory.Exists(outputDir)) Directory.CreateDirectory(outputDir);var baseName = Path.GetFileNameWithoutExtension(imagePath);// 保存可视化结果if (_config.SaveVisualized) {using (var original = new Bitmap(imagePath)) { Bitmap bitResult = ImageProcessor.DrawBoundingBoxesSafe(original, results);var visPath = Path.Combine(outputDir, $"{baseName}_detected.png"); ImageProcessor.SaveImage(bitResult, visPath); Console.WriteLine($"可视化结果已保存: {visPath}"); } }// 保存抠图结果 if (_config.SaveCropped) {using (var mat = Cv2.ImRead(imagePath)) {for (int i = 0; i < results.Count; i++) {var cropped = _detector.CropLabel(mat, results[i].BoundingBox);if (cropped != null) {// 图像增强 _detector.EnhanceImage(ref cropped);var cropPath = Path.Combine(outputDir, $"{baseName}_label_{i + 1}.png"); Console.WriteLine(cropped); ImageProcessor.SaveImage(cropped, cropPath); Console.WriteLine($"抠图已保存: {cropPath}"); cropped.Dispose(); } } } }// 保存检测结果到JSON SaveResultsToJson(results, Path.Combine(outputDir, $"{baseName}_results.json")); Console.WriteLine("\n处理完成! 按任意键继续..."); }catch (Exception ex) { Console.WriteLine($"处理失败: {ex.Message}"); }}
- public List DetectLabels(string imagePath) {var results = new List();using (var mat = Cv2.ImRead(imagePath, OpenCvSharp.ImreadModes.Color)) {if (mat.Empty())thrownew FileNotFoundException($"无法加载图像: {imagePath}");// 方法1: 轮廓检测var contourResults = DetectByContours(mat); results.AddRange(contourResults);// 方法2: 颜色检测var colorResults = DetectByColor(mat); results.AddRange(colorResults);// 方法3: 边缘检测var edgeResults = DetectByEdges(mat); results.AddRange(edgeResults); }// 合并和筛选结果return FilterResults(results); }
- private List DetectByContours(OpenCvSharp.Mat src) {var results = new List(); using (var gray = new OpenCvSharp.Mat())using (var binary = new OpenCvSharp.Mat()) { Cv2.CvtColor(src, gray, ColorConversionCodes.BGR2GRAY);// 二值化 Cv2.Threshold(gray, binary, 0, 255, ThresholdTypes.Binary | ThresholdTypes.Otsu);// 形态学操作var kernel = Cv2.GetStructuringElement(MorphShapes.Rect, new OpenCvSharp.Size(3, 3)); Cv2.MorphologyEx(binary, binary, MorphTypes.Close, kernel);// 查找轮廓 Cv2.FindContours(binary, outvar contours, outvar hierarchy, RetrievalModes.External, ContourApproximationModes.ApproxSimple);foreach (var contour in contours) {var area = Cv2.ContourArea(contour); if (area < _minArea || area > _maxArea)continue; Console.WriteLine($"面积：{area}");var rect = Cv2.BoundingRect(contour);// 计算宽高比var aspectRatio = (double)rect.Width / rect.Height;// 面单通常为矩形，宽高比在一定范围内if (aspectRatio > 0.5 && aspectRatio < 3.0) {// 计算矩形度var rectArea = rect.Width * rect.Height;var rectangularity = area / rectArea;Console.WriteLine(rectangularity);if (rectangularity > 0.55) { results.Add(new DetectionResult { BoundingBox = rect.ToRectangle(), Confidence = rectangularity, DetectionMethod = "Contour" }); } } } }return results; }
- private List DetectByColor(OpenCvSharp.Mat src) {var results = new List();using (var hsv = new OpenCvSharp.Mat())using (var mask = new OpenCvSharp.Mat()) {// 转换到HSV色彩空间 Cv2.CvtColor(src, hsv, ColorConversionCodes.BGR2HSV);// 定义白色/浅色范围var lowerWhite1 = new Scalar(0, 0, 200);var upperWhite1 = new Scalar(180, 30, 255);var lowerWhite2 = new Scalar(0, 0, 180);var upperWhite2 = new Scalar(180, 80, 255);using (var mask1 = new OpenCvSharp.Mat())using (var mask2 = new OpenCvSharp.Mat()) { Cv2.InRange(hsv, lowerWhite1, upperWhite1, mask1); Cv2.InRange(hsv, lowerWhite2, upperWhite2, mask2); Cv2.BitwiseOr(mask1, mask2, mask); }// 形态学操作var kernel = Cv2.GetStructuringElement(MorphShapes.Rect, new OpenCvSharp.Size(5, 5)); Cv2.MorphologyEx(mask, mask, MorphTypes.Close, kernel); Cv2.MorphologyEx(mask, mask, MorphTypes.Open, kernel);// 查找轮廓 Cv2.FindContours(mask, outvar contours, outvar hierarchy, RetrievalModes.External, ContourApproximationModes.ApproxSimple);foreach (var contour in contours) {var area = Cv2.ContourArea(contour);if (area < _minArea || area > _maxArea)continue;var rect = Cv2.BoundingRect(contour);// 计算颜色均匀度var uniformity = CalculateColorUniformity(src, rect);if (uniformity > _confidenceThreshold) { results.Add(new DetectionResult { BoundingBox = rect.ToRectangle(), Confidence = uniformity, DetectionMethod = "Color" }); } } }return results; }3.边缘检测private List DetectByEdges(OpenCvSharp.Mat src) {var results = new List();using (var gray = new OpenCvSharp.Mat())using (var edges = new OpenCvSharp.Mat()) { Cv2.CvtColor(src, gray, ColorConversionCodes.BGR2GRAY);// 降噪 Cv2.GaussianBlur(gray, gray, new OpenCvSharp.Size(5, 5), 1.5);// 边缘检测 Cv2.Canny(gray, edges, 50, 150);// 膨胀var kernel = Cv2.GetStructuringElement(MorphShapes.Rect, new OpenCvSharp.Size(3, 3)); Cv2.Dilate(edges, edges, kernel, iterations: 2);// 查找轮廓 Cv2.FindContours(edges, outvar contours, outvar hierarchy, RetrievalModes.External, ContourApproximationModes.ApproxSimple);foreach (var contour in contours) {var area = Cv2.ContourArea(contour);if (area < _minArea || area > _maxArea)continue;var rect = Cv2.BoundingRect(contour);// 计算边缘密度using (var roi = new OpenCvSharp.Mat(edges, rect)) {var totalPixels = roi.Rows * roi.Cols;var edgePixels = Cv2.CountNonZero(roi);var edgeDensity = (double)edgePixels / totalPixels;if (edgeDensity > 0.1 && rect.Width > 100 && rect.Height > 100) { results.Add(new DetectionResult { BoundingBox = rect.ToRectangle(), Confidence = edgeDensity, DetectionMethod = "Edge" }); } } } }return results; }
OCR基础模型用的是SVTR-LCNet这个架构的网络模型，论文是公开的，我们在这个基础上做的复现与调优。话不多说，先看效果
相机拍照原始包裹图片
OCR识别切割效果(根据识别文字角度自动校正)
定位到每个识别内容的矩形坐标，获取所有当前图片所有切割矩形的最小外接矩形，然后裁切，就可以得到包含所有面单内容的图片
抠面单效果(实际会比面单小，但是满足客户需求，包含了所有面单内容)
废话不多说，附上代码
- ////// 返回面单图片//////异常信息///面单是否增强///是否本地保存///public Bitmap GetLabelImageByBitmap(outstring errorMsg, bool IsEnhanceImage = true, bool IsSaveLocl = true) { Bitmap croppedImage = null; errorMsg = string.Empty; try { if (!File.Exists(imagePath)) { ShellLine.WriteLine($"请确保 {imagePath} 存在"); errorMsg = $"请确保 {imagePath} 存在"; returnnew Bitmap(10, 10); } //图片目录 string imageDir = Path.GetDirectoryName(debugImagePath); if (Directory.Exists(imageDir)) { Directory.CreateDirectory(imageDir); } Bitmap bitmap1 = new Bitmap(imagePath); var rr = oCR.GetOCRDataStr(bitmap1, debugImagePath); // 读取JSON文件 string jsonFilePath = imageDir + "\\content.json"; if (!File.Exists(jsonFilePath)) { errorMsg = $"未找到JSON文件，请确保 {jsonFilePath} 存在"; ShellLine.WriteLine($"未找到JSON文件，请确保 {jsonFilePath} 存在"); returnnew Bitmap(imagePath); } string preRotatedImage = imageDir + "\\preRotatedImg.jpg"; if (!File.Exists(preRotatedImage)) { errorMsg = $"未找到面单文件，请确保包裹面单清晰且存在"; ShellLine.WriteLine($"未找到面单文件，请确保包裹面单清晰且存在"); returnnew Bitmap(imagePath); } // 解析矩形数据并计算最小外接矩形 List rectangles = ParseRectanglesFromJson(jsonFilePath); if (rectangles.Count == 0) { errorMsg = "未在JSON文件中找到有效的矩形数据"; ShellLine.WriteLine("未在JSON文件中找到有效的矩形数据"); returnnew Bitmap(imagePath); } Rectangle boundingRect = CalculateBoundingRectangle(rectangles); ShellLine.WriteLine($"最小外接矩形: X={boundingRect.X}, Y={boundingRect.Y}, Width={boundingRect.Width}, Height={boundingRect.Height}"); ShellLine.WriteLine($"包含 {rectangles.Count} 个元素"); // 加载图片并进行裁剪 using (Bitmap originalImage = new Bitmap(preRotatedImage)) { // 确保矩形在图片范围内 Rectangle safeRect = GetSafeRectangle(boundingRect, originalImage); // 裁剪图片 croppedImage = CropImage(originalImage, safeRect); if (IsEnhanceImage) { // 增强显示 EnhanceImage(ref croppedImage); } if (IsSaveLocl) { // 保存结果 string outputPath = Path.Combine( Path.GetDirectoryName(preRotatedImage), Path.GetFileNameWithoutExtension(preRotatedImage) + "_cropped_enhanced.jpg"); croppedImage.Save(outputPath, ImageFormat.Jpeg); ShellLine.WriteLine($"处理完成！结果已保存到: {outputPath}"); } // 显示裁剪区域信息 ShellLine.WriteLine($"\n裁剪区域信息:"); ShellLine.WriteLine($" 原始图片尺寸: {originalImage.Width}x{originalImage.Height}"); ShellLine.WriteLine($" 裁剪区域: {safeRect.X}, {safeRect.Y}, {safeRect.Width}x{safeRect.Height}"); ShellLine.WriteLine($" 增强后图片尺寸: {croppedImage.Width}x{croppedImage.Height}"); return croppedImage; } } catch (Exception ex) { errorMsg = $"处理过程中出现错误: {ex.Message}"; ShellLine.WriteLine($"处理过程中出现错误: {ex.Message}"); ShellLine.WriteLine($"堆栈跟踪: {ex.StackTrace}"); returnnew Bitmap(imagePath); } finally { // 释放资源 croppedImage?.Dispose(); } }
- ////// 图片增强显示//////publicvoidEnhanceImage(ref Bitmap image) { using (var mat = image.ToMat()) using (var lab = new OpenCvSharp.Mat()) { // 转换为Lab色彩空间 Cv2.CvtColor(mat, lab, ColorConversionCodes.BGR2Lab); Cv2.Split(lab, outvar labChannels); // 对亮度通道进行直方图均衡化 Cv2.EqualizeHist(labChannels[0], labChannels[0]); Cv2.Merge(labChannels, lab); Cv2.CvtColor(lab, mat, ColorConversionCodes.Lab2BGR); // 释放通道 foreach (var channel in labChannels) channel.Dispose(); // 更新图像 image.Dispose(); image = mat.ToBitmap(); } }
- // 计算包含所有矩形的最小外接矩形 static Rectangle CalculateBoundingRectangle(List rectangles) { if (rectangles.Count == 0) thrownew ArgumentException("矩形列表为空"); int minX = int.MaxValue; int minY = int.MaxValue; int maxX = int.MinValue; int maxY = int.MinValue; foreach (Rectangle rect in rectangles) { minX = Math.Min(minX, rect.X); minY = Math.Min(minY, rect.Y); maxX = Math.Max(maxX, rect.X + rect.Width); maxY = Math.Max(maxY, rect.Y + rect.Height); } // 添加一些边距，使裁剪更美观 int margin = 10; minX = Math.Max(0, minX - margin); minY = Math.Max(0, minY - margin); maxX = maxX + margin; maxY = maxY + margin; returnnew Rectangle(minX, minY, maxX - minX, maxY - minY); }// 确保矩形在图片范围内static Rectangle GetSafeRectangle(Rectangle rect, Bitmap image) { int x = Math.Max(0, Math.Min(rect.X, image.Width - 1)); int y = Math.Max(0, Math.Min(rect.Y, image.Height - 1)); int width = Math.Min(rect.Width, image.Width - x); int height = Math.Min(rect.Height, image.Height - y); returnnew Rectangle(x, y, width, height); }// 裁剪图片static Bitmap CropImage(Bitmap source, Rectangle cropArea) { Bitmap target = new Bitmap(cropArea.Width, cropArea.Height); using (Graphics g = Graphics.FromImage(target)) { g.DrawImage(source, new Rectangle(0, 0, cropArea.Width, cropArea.Height), cropArea, GraphicsUnit.Pixel); } return target; }
感谢各位耐心查阅！ 如果您有更好的想法欢迎一起交流，有不懂的也可以微信公众号联系博主，作者公众号会经常发一些实用的小工具和demo源码，需要的可以去看看！另外，如果觉得本篇博文对您或者身边朋友有帮助的，麻烦点个关注！赠人玫瑰，手留余香，您的支持就是我写作最大的动力，感谢您的关注，期待和您一起探讨！再会！
来源：程序园用户自行投稿发布，如果侵权，请联系站长删除
免责声明：如果侵犯了您的权益，请联系站长，我们会及时删除侵权内容，谢谢合作！ |
提升卡
置顶卡
沉默卡
喧嚣卡
变色卡
千斤顶
照妖镜
