How does the vision.Cas​cadeObject​Detector detect left and right eyes separately? It is constantly detecting left and right as the same eye.

I am using the vision.CascadeObjectDetector to find eyes in facial images.

I am reading in the image, converting to gray scale and performing the detection similar to the face detection tutorial here: http://www.mathworks.com.au/help/vision/ref/vision.cascadeobjectdetector-class.html

and using 'LeftEyeCART' and 'RightEyeCART' as classification models.

The for some images in my database the detector is finding the left and right eyes to be the same eyes, with no notable variation in image causing the false detection. See the images below:

How does it work? Does it scan the image from the right hand side to find eyes and simply missed the left eye? Therefore will I have to do some pre - processing on the images to aid the detection? Will I have to train a Cascade Object Detector like this example: http://www.mathworks.com.au/help/vision/ug/train-a-cascade-object-detector.html

for my dataset?

Answer 1

[Dima Lisin]

If both eyes are visible, you would be better off trying to detect the pair of eyes using 'EyePairBig' and 'EyePairSmall' models. Single eye detectors are more useful in cases when only one eye is visible.

[Ushna Asif]

I want to detect left and right separately and I also having the same problem kindly help.

[Dima Lisin]

Why not detect both eyes, and then split the resulting bounding box in half?

Answer 2

[ibrahim thamary]

try this:

    I= imread('img.jpg');
     EyeDetect = vision.CascadeObjectDetector('EyePairBig');
     BB=step(EyeDetect,I);

    x=length(BB(:,4));

   %To detect Eyes
I2 = imcrop(I,BB(x,:));
    figure,imshow(I2);
   rectangle('Position',BB,'LineWidth',4,'LineStyle','-','EdgeColor','b');
  %To detect the left 
  na1 = I2(:, 1 : floor(end/2.3));
  figure,imshow(na1);
  %To detect the right
  na2 = I2(:, floor(end/(2.24))+1 : end );
  figure,imshow(na2);

[Luis Orlando Mendoza Guerrero]

hi,

when i copy, line 8 has error... 'Value must be a 4 element vector'

Answer 3

[Karbala'a Unvi. Science]

Thank you for this explanation