Tutorial: Color Image Processing

Tutorial: Color Image Processing

​

Tutorial: Color Image Segmentation

I. Introduction

In this tutorial, you are learn how to segment a colored object using a web-cam. We will use inRange() algorithm to segment the moving colored object and draw contour or boundary boxes around the tracked object.

II. Tutorial

OpenCV: inRange()

First, read the OpenCV documentation read here

void cv::inRange	(	
    InputArray 	src,
    InputArray 	lowerb,
    InputArray 	upperb,
    OutputArray 	dst 
)	
    
// # python   
    dst= cv2.inRange(src, lowerb, upperb, dst=None)
	dst1 = cv2.inRange(src, (0, 128, 0), (100, 255, 100))

Parameters

• src first input array.

• lowerb inclusive lower boundary array or a scalar.

• upperb inclusive upper boundary array or a scalar.

• dst output array of the same size as src and CV_8U type.

Example: Color Segmentation

One way of choosing the appropriate InRange conditions is analyzing the pixel statistic of a small sub-window within the targeted colored area.

Download the tutorial source code file

• tutorial code
• test image

Choose the target window to analyze with mouse click and drag.

See Appendix for the MouseEvent code

Analyze for the standard deviation and mean of the targeted color within the window.

  Mat roi_RGB(image, selection);           // Set ROI by the selection box       
  Mat roi_HSV;  
  cvtColor(roi_RGB, roi_HSV, CV_BGR2HSV);  
  Scalar means, stddev;  meanStdDev(roi_HSV, means, stddev);  
  cout << "\n  Selected ROI Means= " <<  means << " \n  stddev= " <<  stddev;  

Add slidebars to change the InRange values of each R, G, B or H, S, V and segment each colored ball.

/// set dst as the output of InRange
inRange(hsv, Scalar(MIN(hmin, hmax), MIN(smin, smax), MIN(vmin, vmax)),
	Scalar(MAX(hmin, hmax), MAX(smin, smax), MAX(vmin, vmax)), dst);

Apply appropriate morphology (i.e. dilation/erosion/fill holes) to the output binary images to cluster the detected objects into meaningful blobs.

Find all contours and select the contour with the largest area

Mat image_disp, hsv, hue, mask, dst;
vector<vector<Point> > contours;
vector<Vec4i> hierarchy;
…
findContours(dst, contours, hierarchy, CV_RETR_EXTERNAL, CV_CHAIN_APPROX_SIMPLE);
/// Find the Contour with the largest area ///
int idx = 0, largestComp = 0;
double maxArea = 0;
for (; idx >= 0; idx = hierarchy[idx][0])
{
	const vector<Point>& c = contours[idx];
	double area = fabs(contourArea(Mat(c)));		
	if (area > maxArea)
	{
		maxArea = area;
		largestComp = idx;
	}
}

Draw the contour and a box over the target object

/// Draw the Contour Box on Original Image ///
drawContours(image_disp, contours, largestComp, Scalar(255, 255, 255), 4, 8, hierarchy);
Rect boxPoint = boundingRect(contours[largestComp]);
rectangle(image_disp, boxPoint, Scalar(255, 0, 255), 3);

Now, segment other color balls

III. Exercise

Drawing the trajectory of a colored object with Webcam

Modify your tutorial program to keep drawing the contour on a white or black background image.

• Use your webcam for the source image data.

• Use any colored object as the target.

• Draw rectangles or circles for the output display.

Appendix

Sample code: On Mouse Event

/// On mouse event 
static void onMouse(int event, int x, int y, int, void*)
{
   if (selectObject)  // for any mouse motion
   {
   	selection.x = MIN(x, origin.x);
   	selection.y = MIN(y, origin.y);
   	selection.width = abs(x - origin.x) + 1;
   	selection.height = abs(y - origin.y) + 1;
   	selection &= Rect(0, 0, image.cols, image.rows);  
// Bitwise AND  check selection is within the image coordinate
   }

   switch (event)
   {
   case CV_EVENT_LBUTTONDOWN:
   	selectObject = true;
   	origin = Point(x, y);
   	break;
   case CV_EVENT_LBUTTONUP:
   	selectObject = false;
   	if (selection.area())
   		trackObject = true;
   	break;
   }
}