0
$\begingroup$

To my understanding Viola jones algorithm is a combination of weak classifiers built on training imageset. Lets say we have training images of 24x24 and algorithm selected 100 haar features and built a classifier. Assume there is a test image which is of size 512x512 and there two persons(just faces) in the image one person(100x100 size in 512x512) closer to the camera and the other(10x10 size in 512x512) at a distance.

  1. How does viola jones detect faces detect both persons having learnt features features for faces that fit in 24x24?
  2. Is the algorithm implemented in libraries in such a way that haar features vary depending on the size of the image
|improve this question
$\endgroup$
  • $\begingroup$ Seriously, I don't know the Viola-Jones algorithm, but a person whose complete size is 10x10 pixels will have a face of maybe 1.5x2 pixels. No classifier in this world can classify that. $\endgroup$ – Marcus Müller Nov 5 '16 at 10:15
1
$\begingroup$

I think you are talking about multiscale object detection. It will rescale the images to different resolution and then use trained classifier over it.

|improve this answer
$\endgroup$
0
$\begingroup$

Adding to what Navin Prashath already mentioned, similarly you can rescale the learned featuremaps. Applying a specific featuremap at varying sizes will enable the algorithm to detect i.e. faces of varying sizes within the same image.

|improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.