C# OpenCvSharp DNN 卡证检测矫正

说明

源码地址：https://modelscope.cn/models/iic/cv_resnet_carddetection_scrfd34gkps

在实人认证、文档电子化等场景中需要自动化提取卡证的信息，以便进一步做录入处理。这类场景通常存在两类问题，一是识别卡证类型时易受背景干扰，二是卡证拍摄角度造成的文字畸变影响OCR准确率。鉴于证件类数据的敏感性，我们采用大量合成卡证数据做训练(参见：SyntheticCards), 并改造人脸检测SOTA方法SCRFD(论文地址, 代码地址)训练了卡证检测矫正模型，可以对各类国际常见卡证（如，身份证、护照、驾照等）进行检测、定位及矫正，得到去除背景的正视角卡证图像，便于后续卡证分类或OCR内容提取。

效果

模型

Model Properties
-------------------------
---------------------------------------------------------------

Inputs
-------------------------
name：input.1
tensor：Float[1, 3, 640, 640]
---------------------------------------------------------------

Outputs
-------------------------
name：1401
tensor：Float[1, 25600, 1]
name：1455
tensor：Float[1, 6400, 1]
name：1507
tensor：Float[1, 1600, 1]
name：1408
tensor：Float[1, 25600, 4]
name：1461
tensor：Float[1, 6400, 4]
name：1513
tensor：Float[1, 1600, 4]
name：1415
tensor：Float[1, 25600, 8]
name：1467
tensor：Float[1, 6400, 8]
name：1519
tensor：Float[1, 1600, 8]
---------------------------------------------------------------

项目

代码

using OpenCvSharp;
using OpenCvSharp.Dnn;
using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.Drawing;
using System.Drawing.Imaging;
using System.Windows.Forms;

namespace OpenCvSharp_Demo
{
 public partial class Form1 : Form
 {
 public Form1()
 {
 InitializeComponent();
 }

 string fileFilter = "*.*|*.bmp;*.jpg;*.jpeg;*.tiff;*.tiff;*.png";
 string startupPath;
 string image_path;

 private void Form1_Load(object sender, EventArgs e)
 {
 startupPath = System.Windows.Forms.Application.StartupPath;

 image_path = "1.jpg";
 pictureBox1.Image = new Bitmap(image_path);
 image = new Mat(image_path);

 opencv_net = CvDnn.ReadNetFromOnnx("carddetection_scrfd34gkps.onnx");
 }

 private void button1_Click(object sender, EventArgs e)
 {
 OpenFileDialog ofd = new OpenFileDialog();
 ofd.Filter = fileFilter;
 if (ofd.ShowDialog() != DialogResult.OK) return;

 pictureBox1.Image = ;
 pictureBox2.Image = ;
 textBox1.Text = "";

 image_path = ofd.FileName;
 pictureBox1.Image = new Bitmap(image_path);
 image = new Mat(image_path);
 }

 Stopwatch stopwatch = new Stopwatch();
 Net opencv_net;
 Mat BN_image;
 Mat image;
 Mat result_image;

 float[] stride = { 8.0f, 16.0f, 32.0f };
 int inpWidth = 640;
 int inpHeight = 640;
 float confThreshold = 0.5f;
 float nmsThreshold = 0.5f;
 bool keep_ratio = true;

 Mat resize_image(Mat srcimg, ref int newh, ref int neww, ref int top, ref int left)
 {
 int srch = srcimg.Rows, srcw = srcimg.Cols;
 newh = inpHeight;
 neww = inpWidth;
 Mat dstimg = new Mat();
 if (keep_ratio && srch != srcw)
 {
 float hw_scale = (float)srch / srcw;
 if (hw_scale > 1)
 {
 newh = inpHeight;
 neww = (int)(inpWidth / hw_scale);
 Cv2.Resize(srcimg, dstimg, new OpenCvSharp.Size(neww, newh));
 left = (int)((inpWidth - neww) * 0.5);
 Cv2.CopyMakeBorder(dstimg, dstimg, 0, 0, left, inpWidth - neww - left, BorderTypes.Constant, 0);
 }
 else
 {
 newh = (int)(inpHeight * hw_scale);
 neww = inpWidth;
 Cv2.Resize(srcimg, dstimg, new OpenCvSharp.Size(neww, newh));
 top = (int)((inpHeight - newh) * 0.5);
 Cv2.CopyMakeBorder(dstimg, dstimg, top, inpHeight - newh - top, 0, 0, BorderTypes.Constant, 0);
 }
 }
 else
 {
 Cv2.Resize(srcimg, dstimg, new OpenCvSharp.Size(neww, newh));
 }
 return dstimg;
 }

 unsafe private void button2_Click(object sender, EventArgs e)
 {
 if (image_path == "")
 {
 return;
 }

 stopwatch.Restart();
 image = new Mat(image_path);
 result_image = image.Clone();

 int newh = 0, neww = 0, padh = 0, padw = 0;
 Mat img = resize_image(image, ref newh, ref neww, ref padh, ref padw);
 Mat blob = new Mat();

 BN_image = CvDnn.BlobFromImage(img, 1 / 128.0, new OpenCvSharp.Size(inpWidth, inpHeight), new Scalar(127.5, 127.5, 127.5), true, false);

 opencv_net.SetInput(BN_image);

 Mat[] outs = new Mat[9] { new Mat(), new Mat(), new Mat(), new Mat(), new Mat(), new Mat(), new Mat(), new Mat(), new Mat() };
 string[] outBlobNames = opencv_net.GetUnconnectedOutLayersNames();
 opencv_net.Forward(outs, outBlobNames);

 //generate proposals
 List<float> confidences = new List<float>();
 List boxes = new List();
 List> landmarks = new List>();
 float ratioh = (float)image.Rows / newh, ratiow = (float)image.Cols / neww;
 int n = 0, i = 0, j = 0, k = 0, l = 0;
 for (n = 0; n < 3; n++)
 {
 int num_grid_x = (int)(inpWidth / stride[n]);
 int num_grid_y = (int)(inpHeight / stride[n]);
 float* pdata_score = (float*)outs[n * 3].Data; //score
 float* pdata_bbox = (float*)outs[n * 3 + 1].Data; //bounding box
 float* pdata_kps = (float*)outs[n * 3 + 2].Data; //face landmark
 for (i = 0; i < num_grid_y; i++)
 {
 for (j = 0; j < num_grid_x; j++)
 {
 for (k = 0; k < 4; k++)
 {
 if (pdata_score[0] > confThreshold)
 {
 int xmin = (int)(((j - pdata_bbox[0]) * stride[n] - padw) * ratiow);
 int ymin = (int)(((i - pdata_bbox[1]) * stride[n] - padh) * ratioh);
 int width = (int)((pdata_bbox[2] + pdata_bbox[0]) * stride[n] * ratiow);
 int height = (int)((pdata_bbox[3] + pdata_bbox[1]) * stride[n] * ratioh);
 confidences.Add(pdata_score[0]);
 boxes.Add(new Rect(xmin, ymin, width, height));
 List landmark = new List();
 for (l = 0; l < 8; l += 2)
 {
 landmark.Add((int)(((j + pdata_kps[l]) * stride[n] - padw) * ratiow));
 landmark.Add((int)(((i + pdata_kps[l + 1]) * stride[n] - padh) * ratioh));
 }
 landmarks.Add(landmark);
 }
 pdata_score++;
 pdata_bbox += 4;
 pdata_kps += 8;
 }
 }
 }
 }
 int[] indices;
 CvDnn.NMSBoxes(boxes, confidences, confThreshold, nmsThreshold, out indices);
 //draw bbox and kps
 for (i = 0; i < indices.Length; ++i)
 {
 int idx = indices[i];
 Rect box = boxes[idx];
 Cv2.Rectangle(result_image, new OpenCvSharp.Point(box.X, box.Y), new OpenCvSharp.Point(box.X + box.Width, box.Y + box.Height), new Scalar(0, 0, 255), 2);
 for (k = 0; k < 8; k += 2)
 {
 Cv2.Circle(result_image, new OpenCvSharp.Point(landmarks[idx][k], landmarks[idx][k + 1]), 10, new Scalar(0, 255, 0), -1);
 }
 //Get the label for the class name and its confidence
 string label = confidences[idx].ToString("P2");
 //Display the label at the top of the bounding box
 int baseLine;
 OpenCvSharp.Size labelSize = Cv2.GetTextSize(label, HersheyFonts.HersheySimplex, 0.5, 2, out baseLine);
 int top = Math.Max(box.Y, labelSize.Height);
 Cv2.PutText(result_image, label, new OpenCvSharp.Point(box.X, top - 10), HersheyFonts.HersheySimplex, 1, new Scalar(0, 255, 0), 2);
 }

 stopwatch.Stop();
 double costTime = stopwatch.Elapsed.TotalMilliseconds;

 textBox1.Text = $"耗时:{costTime:F2}ms";
 pictureBox2.Image = new Bitmap(result_image.ToMemoryStream());
 }

 private void button3_Click(object sender, EventArgs e)
 {
 if (pictureBox2.Image == )
 {
 return;
 }
 Bitmap output = new Bitmap(pictureBox2.Image);
 var sdf = new SaveFileDialog();
 sdf.Title = "保存";
 sdf.Filter = "Images (*.jpg)|*.jpg|Images (*.png)|*.png|Images (*.bmp)|*.bmp";
 if (sdf.ShowDialog() == DialogResult.OK)
 {
 switch (sdf.FilterIndex)
 {
 case 1:
 {
 output.Save(sdf.FileName, ImageFormat.Jpeg);
 break;
 }
 case 2:
 {
 output.Save(sdf.FileName, ImageFormat.Png);
 break;
 }
 case 3:
 {
 output.Save(sdf.FileName, ImageFormat.Bmp);
 break;
 }
 }
 MessageBox.Show("保存成功，位置：" + sdf.FileName);
 }
 }
 }
}

参考

https://github.com/hpc203/cv_resnet_carddetection_scrfd34gkps-opencv-dnn