In the following original picture, the objects are connected together, the purpose is to separate them, and then extract the outline and position
Original image: 
final effect:
The troublesome part is that the two connected target objects are separated from the lower right corner, as shown in the figure below: 
Basic methods: BoxFilter filtering, binarization, contour extraction, convex hull detection, image moment
code show as below:
/// <summary>
/// 获取分割点
/// </summary>
/// <param name="contours"></param>
/// <param name="contourCount"></param>
/// <param name="arcLength"></param>
/// <param name="farDistance"></param>
/// <returns></returns>
public List<Point> GetSplitPoints(Point[][] contours, List<int> contourCount, int arcLength, int farDistance)
{
#region 凸包检测
List<double> lArc = new List<double>();
//Mat src = srcImage.Clone();
List<Point[]> lpContours = new List<Point[]>();
List<int> hulls = new List<int>();
Point lastP = new Point();
Point firstP = new Point();
Point farLastP = new Point();
List<Point> lps = new List<Point>();
int dot = 1;
List<int> depth = new List<int>();
for (int i = 0; i < contourCount.Count; i++)
{
InputArray inputArray = InputArray.Create<Point>(contours[contourCount[i]]);
OutputArray outputArray = OutputArray.Create(hulls);
Cv2.ConvexHull(inputArray, outputArray, false, false);
if (Cv2.ArcLength(inputArray, true) < arcLength)
{
//lArc.Add(Cv2.ArcLength(inputArray, true));
continue;
}
//前三个值得含义分别为：凸缺陷的起始点，凸缺陷的终点，凸缺陷的最深点（即边缘点到凸包距离最大点）。
var defects = Cv2.ConvexityDefects(contours[contourCount[i]], hulls);
for (int j = 0; j < defects.Length; j++)
{
OpenCvSharp.Point start = contours[contourCount[i]][defects[j].Item0];
OpenCvSharp.Point end = contours[contourCount[i]][defects[j].Item1];
OpenCvSharp.Point far = contours[contourCount[i]][defects[j].Item2];
//OpenCvSharp.Point fart = contours[contourCount[i]][defects[j].Item3];
if (defects[j].Item3 > farDistance) //(4500 < defects[j].Item3 && defects[j].Item3 < 300000)
{
lps.Add(contours[contourCount[i]][defects[j].Item2]);
depth.Add(defects[j].Item3);
}
}
}
#endregion
return lps;
}
/// <summary>
/// 获取最小内接矩形
/// </summary>
/// <param name="contours"></param>
/// <param name="contourCount"></param>
/// <returns></returns>
public List<RotatedRect> GetMinRects(Point[][] contours, List<int> contourCount)
{
//Cv2.ImShow(",mmmm", morphImage);
//double rotateAngel = 0;
Point2f[] vertices = new Point2f[4];
//Point2f minRectcenterPoint = new Point2f();
List<RotatedRect> minRects = new List<RotatedRect>();
for (int i = 0; i < contourCount.Count; i++)
{
//获取轮廓点的矩形区域
//绘制Rio区域最小矩形
#region 绘制Rio区域最小矩形
RotatedRect minRect = Cv2.MinAreaRect(contours[contourCount[i]]);
minRects.Add(minRect);
#endregion
}
return minRects;
}
/// <summary>
/// 返回设置范围内的轮廓
/// </summary>
/// <param name="mat"></param>
/// <param name="range1"></param>
/// <param name="range2"></param>
/// <param name="contourCount"></param>
/// <returns></returns>
public Point[][] GetImageContours(Mat mat, int length, out List<int> contourCount)
{
List<double> arclength = new List<double>();
OpenCvSharp.Point[][] contours;
HierarchyIndex[] hierarchies;
//Cv2.ImShow(",mmmm", mat);
Cv2.FindContours(mat, out contours, out hierarchies, RetrievalModes.External, ContourApproximationModes.ApproxSimple, new Point());
Mat connImg = Mat.Zeros(mat.Size(), MatType.CV_8UC3);
Point2f[] vertices = new Point2f[4];
Mat drawOutline = Mat.Zeros(mat.Size(), mat.Type());
int sum = 0;
contourCount = new List<int>();
for (int i = 0; i < contours.Length; i++)
{
Rect rect1 = Cv2.BoundingRect(contours[i]);
if (Cv2.ArcLength(contours[i], true) > length)//(rect1.Width > range1 && rect1.Height < range2)
{
Cv2.DrawContours(drawOutline, contours, i, new Scalar(255, 0, 255), 2, LineTypes.Link8, hierarchies);
contourCount.Add(i);
arclength.Add(Cv2.ArcLength(contours[i], true));
sum++;
}
}
Cv2.ImShow("contours", drawOutline);
return contours;
}
/// <summary>
/// 图像灰度
/// 盒子滤波 保留边缘信息
/// 自适应阈值 效果不错 无需形态学降噪
/// 取反操作
/// 过滤不需要轮廓信息(面积 边长)
/// 轮廓提取
/// (以上每一步都很重要,否则,无法获取良好的轮廓)
/// 凸包检测
/// 根据轮廓信息,查找大凸包,获取分割点
/// 重新操作图像
/// 在二值化图像时,分割连接点位置
/// 绘制轮廓
/// 绘制最小内接矩形和质心点
/// 识别目标位置完成
/// 注意:不同大小的图像处理时,需要修改自适应阈值参数、轮廓过滤面积、凸包检测的分割点过滤
/// </summary>
/// <param name="srcImage"></param>
/// <returns></returns>
public Mat PreProcess(Mat srcImage)
{
Mat grayMat = new Mat();
Cv2.CvtColor(srcImage, grayMat, ColorConversionCodes.BGRA2GRAY);
//Cv2.ImShow("grayMat", grayMat);
Mat blurImg = BoxFilter(grayMat);
//Cv2.ImShow("blurImg", blurImg);
// 注意:不同大小的图像处理时,需要修改参数
Mat threshold = new Mat();
Cv2.AdaptiveThreshold(blurImg, threshold, 255, AdaptiveThresholdTypes.MeanC, ThresholdTypes.Binary, 15, 2);
//Cv2.Threshold(threshold, threshold, 0, 255, ThresholdTypes.BinaryInv);
Cv2.ImShow("threshold", threshold);
//Mat morphImg = MorphImage(threshold, MorphShapes.Ellipse, MorphTypes.Dilate, 1, new OpenCvSharp.Size(3, 3));
//Cv2.ImShow("morphImg", morphImg);
//Mat cannyImg = new Mat();
//Cv2.Laplacian(morphImg2, cannyImg, MatType.CV_8UC3, 5, 1);//Cv2.Canny(morphImg, cannyImg, 30, 90);//3和4参数的 最佳比例在1/3和1/2之间
//Cv2.ImShow("cannyImg", cannyImg);
Mat bitwiseMat = new Mat();
Cv2.BitwiseNot(threshold, bitwiseMat);
Cv2.ImShow("bitwiseMat", bitwiseMat);
List<int> contourCount;
//轮廓提取
Point[][] contours = GetImageContours(bitwiseMat, 600, out contourCount);
//凸包检测
List<Point> lps = GetSplitPoints(contours, contourCount, 800, 4500);
// 注意:不同大小的图像处理时,需要修改参数
//重新处理
Cv2.AdaptiveThreshold(blurImg, threshold, 255.0, AdaptiveThresholdTypes.MeanC, ThresholdTypes.Binary, 13, 2);
Cv2.ImShow("threshold1", threshold);
//MorphImage(threshold, MorphShapes.Ellipse, MorphTypes.Close, 1, new OpenCvSharp.Size(3, 3));
//Cv2.ImShow("morphImg1", morphImg);
Cv2.BitwiseNot(threshold, bitwiseMat);
Cv2.ImShow("bitwiseMat1", bitwiseMat);
//提取凸显点坐标
if (lps.Count > 1)
{
Cv2.Line(bitwiseMat, lps[0], lps[1], Scalar.Black, 2, LineTypes.Link8);
}
Cv2.ImShow("bitwiseMat2", bitwiseMat);
//轮廓提取
contourCount.Clear(); // 注意:不同大小的图像处理时,需要修改length参数
Point[][] newContours = GetImageContours(bitwiseMat, 550, out contourCount);
List<RotatedRect> rotatedRects = GetMinRects(newContours, contourCount);
for (int i = 0; i < rotatedRects.Count; i++)
{
#region 绘制Rio区域最小矩形
Point2f[] vertices = rotatedRects[i].Points();
#endregion
//绘制最小矩形
#region 绘制最小矩形
Cv2.Line(srcImage, Convert.ToInt32(vertices[0].X), Convert.ToInt32(vertices[0].Y), Convert.ToInt32(vertices[1].X), Convert.ToInt32(vertices[1].Y), new Scalar(0, 0, 255), 2);
Cv2.Line(srcImage, Convert.ToInt32(vertices[0].X), Convert.ToInt32(vertices[0].Y), Convert.ToInt32(vertices[3].X), Convert.ToInt32(vertices[3].Y), new Scalar(0, 0, 255), 2);
Cv2.Line(srcImage, Convert.ToInt32(vertices[1].X), Convert.ToInt32(vertices[1].Y), Convert.ToInt32(vertices[2].X), Convert.ToInt32(vertices[2].Y), new Scalar(0, 0, 255), 2);
Cv2.Line(srcImage, Convert.ToInt32(vertices[2].X), Convert.ToInt32(vertices[2].Y), Convert.ToInt32(vertices[3].X), Convert.ToInt32(vertices[3].Y), new Scalar(0, 0, 255), 2);
//获取重心点
Moments M;
M = Cv2.Moments(vertices);
double cX = M.M10 / M.M00;
double cY = M.M01 / M.M00;
//显示目标中心并提取坐标点
Cv2.Circle(srcImage, (int)cX, (int)cY, 2, Scalar.Yellow, 2);
//Console.WriteLine("AngleRect_angle: {0}", minRect.Angle);
#endregion
}
Cv2.ImShow("srcImage", srcImage);
return null;
}
Image binarization after grayscale image:
image inversion
draw outline
Convex hull detection, find the segmentation point, the location of the segmentation point found at the yellow point mark in the figure below
Put the found segmentation points in the binarized image, connect a line, and re-contour recognition can be segmented
Minimal contour rectangle extraction and drawing, and drawing centroid position
At this point, the connection has been separated
Note: To use the above method, some parameters need to be set according to the image size, such as binarization processing parameters, filter contour shape size, convex hull detection point acquisition and other positions, and parameters need to be set according to the actual situation;