Giters

sendroid-laonan / opencv

OpenCV -ver 3.30

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Java + OpenCV 答题卡识别

Demo介绍

  • 识别静态图片答题卡，对图片要求比较高，必须A4大小，内容铺满窗口；
  • 印刷质量太差影响答案精度（已测）
  • 扫描质量太差可能影响答案精度（未测）

准备工具

  • OpenCV 3.30
  • IntelliJ IDEA 2017.2.5
  • JDK 1.8_151
  • A4公务员考试答题卡

环境配置

 1. 打开IDEA 选择 File -> Project Structure -> Modules -> Add -> JARs or directories
 2. 选择 opencv 目录下的 build > java > opencv-330.jar
 3. File -> Project Structure -> Libraries -> New Project Library
 4. 选择 opencv 目录下的 build > java > x64 > opencv_java330.dll

主要步骤

1. 读取图片
2. 图片转化为灰度图
3. 图片设定阈值
4. 开运算(先腐蚀，后膨胀)
5. 指定答题区域
6. 找到涂选框
7. 根据涂选框包含的坐标确定所涂选的选项及题号

具体实现

  1. 加载Library
 static {
           System.loadLibrary(Core.NATIVE_LIBRARY_NAME);
       }

2. 设定答题卡图片path

 String sheet = "E:/picpool/A4.jpg";

3. 读取图片

Mat img = Imgcodecs.imread(oriImg);
Mat srcImage2 = new Mat();
Mat srcImage3 = new Mat();
Mat srcImage4 = new Mat();
Mat srcImage5 = new Mat();
Mat colorfulImage = new Mat();

4. 图片变成灰度图片

Imgproc.cvtColor(img, srcImage2, Imgproc.COLOR_RGB2GRAY);

5. 图片二值化

Imgproc.adaptiveThreshold(srcImage2, srcImage3, 255, Imgproc.ADAPTIVE_THRESH_MEAN_C, Imgproc.THRESH_BINARY_INV, 255, 1);

6. 腐蚀膨胀操作（图片印刷质量太差影响精度）

Mat element = Imgproc.getStructuringElement(MORPH_RECT, new Size(1, 6));
Imgproc.erode(srcImage3, srcImage4, element);
Imgproc.dilate(srcImage4, srcImage5, element);

7. 指定答题区域（根据矩形覆盖面积选取，可用多个对象对多块区域进行操作）

Mat imag_ch1 = srcImage4.submat(new Rect(200, 1065, 1930, 2210));
Mat imag_ch2 = srcImage4.submat(new Rect(100, 500, 300, 1000));

8. 识别提取轮廓，根据宽度范围找到涂选框

Vector<MatOfPoint> chapter1 = new Vector<>();
Imgproc.findContours(imag_ch1, chapter1, new Mat(), 2, 3);
Mat result = new Mat(imag_ch1.size(), CV_8U, new Scalar(255));
Imgproc.drawContours(colorfulImage, chapter1, -1, new Scalar(0), 2);

List<RectComp> RectCompList = new ArrayList<>();
    double startX, startY, endX, endY;
    for (MatOfPoint mop : chapter1) {
        Rect rm = Imgproc.boundingRect(mop);
        RectComp ti = new RectComp(rm);
        if (ti.rm.width > 60 && ti.rm.width < 85) {
            RectCompList.add(ti);
            Imgproc.rectangle(colorfulImage, new Point(startX, startY), new Point(endX, endY), new Scalar(0, 0, 255), 2);
        }
    }

9. 根据涂选框包含的坐标确定所涂选的选项及题号

for (RectComp rc : RectCompList) {
    for (int h = 0; h < 7; h++) {
        if ((rc.rm.contains(new Point(rc.rm.x + 20, 115 + (320 * h))))) {
            for (int w = 0; w < 4; w++) {
                if (rc.rm.contains(new Point(55 + (500 * w), rc.rm.y))) {
                    listenAnswer.put(1 + (20 * h) + (5 * w), "A");
                } else if (rc.rm.contains(new Point(135 + (500 * w), rc.rm.y))) {
                    listenAnswer.put(2 + (20 * h) + (5 * w), "A");
                } else if (rc.rm.contains(new Point(215 + (500 * w), rc.rm.y))) {
                    listenAnswer.put(3 + (20 * h) + (5 * w), "A");
                } else if (rc.rm.contains(new Point(300 + (500 * w), rc.rm.y))) {
                    listenAnswer.put(4 + (20 * h) + (5 * w), "A");
                } else if (rc.rm.contains(new Point(380 + (500 * w), rc.rm.y))) {
                    listenAnswer.put(5 + (20 * h) + (5 * w), "A");
                }
            }
        } 
    }
 }

10. 输出结果

Iterator iter = listenAnswer.entrySet().iterator();
    while (iter.hasNext()) {
        Map.Entry entry = (Map.Entry) iter.next();
        Object key = entry.getKey();
        Object val = entry.getValue();
        System.out.println("第" + key + "题,选项:" + val);
    }

About

OpenCV -ver 3.30