Calculate percentage of a face in an image

Asked

Viewed 390 times

3

How would the calculation of porcetangem of a face in an image? Suppose I had a 3x4 photo of a user, but he had a lot of his hair covering his face. Is there any way to detect that the image does not have a minimum percentage of the user’s face?

ios image-processing opencv
  • 1

    With shape recognition you can search for elements such as mouths, eyes and hair. Hence you can count the amount of elements present. But percent of exposed area cannot be done accurately, because people have faces with different shapes.

  • The DMV-MG System has a mechanism when the photo for CNH is taken. It calculates 70% face in an image. I’m going to run a facial recognition test and do some calculations. I think it might solve.

  • Renan and Karlphillip are right. You can even do something: an idea that occurs to me is to use Haar Cascade to detect the region of the face, and find it using its own classifier to predict whether each pixel is skin or not; finally, the ratio between pixels classified as "skin" and the total pixels in the face region results in an estimate of the displayed face percentage. However, this estimate should vary greatly with the size of the region detected (mainly because of rotation), and with errors of skin classifier for hair close to skin tone, for example.

  • 1

    Is your goal to validate that the person is sufficiently visible in the photo? Is the application serious (like the CNH you mention) and the photo taking environment more or less controlled? Because if the possible estimation errors are not severe in the domain, or the user can be instructed to position the face correctly, the idea I suggested might be useful.

  • May be useful for future reference (yours or others): http://answall.com/a/118221/73

1 answer

5


Facial detection algorithms work from Features () facial characteristics. Thus, certain Features must be present in the photo so that the algorithm can extract the necessary data and determine whether the object being evaluated is a face or not.

It is completely possible to have a high percentage of a face exposure in the photo and yet the detection fails if the Features are hidden/manipulated, such as pose, facial expression, lighting, etc.

Therefore, thinking of the success of facial detection as if it depended on a minimum percentage of face exposure in the photo is not a realistic approach.

I recommend the article Eigenfaces vs Fishfaces, from Columbia University (1997), which discusses a more robust technique for solving these problems and makes an interesting comparison with the two most popular facial detection/recognition algorithms.

Browser other questions tagged ios image-processing opencv

You are not signed in. Login or sign up in order to post.