OpenCV & Python : Face Detection using haarcascades is detecting many boxes as eyes.

Question

I am using Haarcascades for detecting faces and eyes. My problem is, its bounding many boxes as eyes. My syntax is

face_cascade = cv2.CascadeClassifier('haarcascades\haarcascade_frontalface_default.xml')
eye_cascade = cv2.CascadeClassifier('haarcascades\haarcascade_eye.xml')
img = cv2.imread('SAM7.jpg')
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
faces = face_cascade.detectMultiScale(gray,1.2,6)

I am currently using 1.2 and 6. What should be the value of the parameters in faces(5 line) like scaleFactor, minNeighbors ??

you should reduce eye detection to regions where faces were previously detected. — Micka, Feb 18 '16 at 16:48

score 1 · Accepted Answer · answered Feb 18 '16 at 21:16

You really need to play with the parameters and find the ones works fine for you. Always there is a better way to do it but remember you'll never achieve 100% accuracy. You can learn about the parameters here.

An example of face and eyes detection in python that works for me:

    import cv2

    face_cascade = cv2.CascadeClassifier("../haarcascades/haarcascade_frontalface_default.xml")
    eye_cascade = cv2.CascadeClassifier("../haarcascades/haarcascade_eye.xml") 

    cap = cv2.VideoCapture(0)

        while cap.isOpened():
            ret, frame = cap.read()
            if ret:
                gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
                faces = face_cascade.detectMultiScale(
                        gray,
                        scaleFactor=1.3,
                        minNeighbors=5,
                        minSize=(50, 50)
                       )

                for (x, y, w, h) in faces:
                    cv2.rectangle(frame, (x, y), (x + w, y + h), (0, 255, 0),2)
                    roi_gray = gray[y:y + h, x:x + w]
                    roi_color = frame[y:y + h, x:x + w]
                    eyes = eye_cascade.detectMultiScale(
                           roi_gray,
                           scaleFactor=1.2,
                           minNeighbors=5,
                           minSize=(10, 10)
                           )

                    for (ex, ey, ew, eh) in eyes:
                        cv2.rectangle(roi_color, (ex, ey), (ex + ew, ey + eh), (255, 0, 0), 2)

            cv2.imshow("Faces found", frame)

            k = cv2.waitKey(10) & 0xff
            if k == 27:
                break

    cv2.destroyAllWindows()
    cap.release()

I hope this helps you. If you need help with the code let me know.

score 0 · Answer 2 · answered Feb 18 '16 at 19:12

To find parameters, a very good real-time implementation has been implemented here for face and eye detection. In this project, he first detects face region and then eyes. If you go through the code, you will find the section you can also draw the bounding box for face:

main.cpp : Detecting Face and Eyes

  //-- Detect faces
  face_cascade.detectMultiScale( frame_gray, faces, 1.1, 2, 0|CV_HAAR_SCALE_IMAGE|CV_HAAR_FIND_BIGGEST_OBJECT, cv::Size(150, 150) );
//  findSkin(debugImage);

  for( int i = 0; i < faces.size(); i++ )
  {
    rectangle(debugImage, faces[i], 1234);
  }
  //-- Show what you got
  if (faces.size() > 0) {
    findEyes(frame_gray, faces[0]);
  }

It uses face ROI to detect eyes:

void findEyes(cv::Mat frame_gray, cv::Rect face) {
  cv::Mat faceROI = frame_gray(face);
  cv::Mat debugFace = faceROI;

  if (kSmoothFaceImage) {
    double sigma = kSmoothFaceFactor * face.width;
    GaussianBlur( faceROI, faceROI, cv::Size( 0, 0 ), sigma);
  }
  //-- Find eye regions and draw them
  int eye_region_width = face.width * (kEyePercentWidth/100.0);
  int eye_region_height = face.width * (kEyePercentHeight/100.0);
  int eye_region_top = face.height * (kEyePercentTop/100.0);
  cv::Rect leftEyeRegion(face.width*(kEyePercentSide/100.0),
                         eye_region_top,eye_region_width,eye_region_height);
  cv::Rect rightEyeRegion(face.width - eye_region_width - face.width*(kEyePercentSide/100.0),
                          eye_region_top,eye_region_width,eye_region_height);

  //-- Find Eye Centers
  cv::Point leftPupil = findEyeCenter(faceROI,leftEyeRegion,"Left Eye");
  cv::Point rightPupil = findEyeCenter(faceROI,rightEyeRegion,"Right Eye");
  // get corner regions
  cv::Rect leftRightCornerRegion(leftEyeRegion);
  leftRightCornerRegion.width -= leftPupil.x;
  leftRightCornerRegion.x += leftPupil.x;
  leftRightCornerRegion.height /= 2;
  leftRightCornerRegion.y += leftRightCornerRegion.height / 2;
  cv::Rect leftLeftCornerRegion(leftEyeRegion);
  leftLeftCornerRegion.width = leftPupil.x;
  leftLeftCornerRegion.height /= 2;
  leftLeftCornerRegion.y += leftLeftCornerRegion.height / 2;
  cv::Rect rightLeftCornerRegion(rightEyeRegion);
  rightLeftCornerRegion.width = rightPupil.x;
  rightLeftCornerRegion.height /= 2;
  rightLeftCornerRegion.y += rightLeftCornerRegion.height / 2;
  cv::Rect rightRightCornerRegion(rightEyeRegion);
  rightRightCornerRegion.width -= rightPupil.x;
  rightRightCornerRegion.x += rightPupil.x;
  rightRightCornerRegion.height /= 2;
  rightRightCornerRegion.y += rightRightCornerRegion.height / 2;
  rectangle(debugFace,leftRightCornerRegion,200);
  rectangle(debugFace,leftLeftCornerRegion,200);
  rectangle(debugFace,rightLeftCornerRegion,200);
  rectangle(debugFace,rightRightCornerRegion,200);
  // change eye centers to face coordinates
  rightPupil.x += rightEyeRegion.x;
  rightPupil.y += rightEyeRegion.y;
  leftPupil.x += leftEyeRegion.x;
  leftPupil.y += leftEyeRegion.y;
  // draw eye centers
  circle(debugFace, rightPupil, 3, 1234);
  circle(debugFace, leftPupil, 3, 1234);

  //-- Find Eye Corners
  if (kEnableEyeCorner) {
    cv::Point2f leftRightCorner = findEyeCorner(faceROI(leftRightCornerRegion), true, false);
    leftRightCorner.x += leftRightCornerRegion.x;
    leftRightCorner.y += leftRightCornerRegion.y;
    cv::Point2f leftLeftCorner = findEyeCorner(faceROI(leftLeftCornerRegion), true, true);
    leftLeftCorner.x += leftLeftCornerRegion.x;
    leftLeftCorner.y += leftLeftCornerRegion.y;
    cv::Point2f rightLeftCorner = findEyeCorner(faceROI(rightLeftCornerRegion), false, true);
    rightLeftCorner.x += rightLeftCornerRegion.x;
    rightLeftCorner.y += rightLeftCornerRegion.y;
    cv::Point2f rightRightCorner = findEyeCorner(faceROI(rightRightCornerRegion), false, false);
    rightRightCorner.x += rightRightCornerRegion.x;
    rightRightCorner.y += rightRightCornerRegion.y;
    circle(faceROI, leftRightCorner, 3, 200);
    circle(faceROI, leftLeftCorner, 3, 200);
    circle(faceROI, rightLeftCorner, 3, 200);
    circle(faceROI, rightRightCorner, 3, 200);
  }

OpenCV & Python : Face Detection using haarcascades is detecting many boxes as eyes.

2 Answers2