Why to use generic face-rec algorithms if we already have few images in database?
62 views
Skip to first unread message
Dairu
May 18, 2012, 6:43:18 AM5/18/12
to Face Recognition Research Community
Hi,
I wanted to ask a very basic doubt and possibly discuss my solution
here, if you permit.
The main advantage of face recognition algorithms like PCA lies in the
domain of dimension reduction. For eg., if we have a collection of
images of say 20000 persons, then we dont want to store it and make
recognition process much more complex. Instead, we reduce this huge
number to a very smaller number (say 200) by the use of eigenfaces.
My point is that if we have only 100 images in database and we want to
compare a new image, the USP of PCA vanishes (though not in terms of
working procedure).
PCA (and other algorithms) aimed at recognizing the person, i.e. WHO
IS THIS? whereas with such small no. of images in my database, i want
to focus on IS THIS THE CORRECT PERSON OR NOT? I am trying to think
as answer to my query in YES or NO rather that 'HE IS MICHEAL'.
Do we have any algorithms which aims at this problem?
I know that I can modify PCA and apply a small condition to check but
why to use such mathematically complex algorithm.
POSSIBLE SOLUTION PROVIDED BY Mr. Sahil in of the posts : As all
images are basically matrices, we take matrices of 2 images and try to
calculate the co-relation factor of the these matrices. If it is found
to be greater than a threshold value we will say that 2 images matched
and this is the right person.
Please comment and if possible, discuss.
Thank You.
I wanted to ask a very basic doubt and possibly discuss my solution
here, if you permit.
The main advantage of face recognition algorithms like PCA lies in the
domain of dimension reduction. For eg., if we have a collection of
images of say 20000 persons, then we dont want to store it and make
recognition process much more complex. Instead, we reduce this huge
number to a very smaller number (say 200) by the use of eigenfaces.
My point is that if we have only 100 images in database and we want to
compare a new image, the USP of PCA vanishes (though not in terms of
working procedure).
PCA (and other algorithms) aimed at recognizing the person, i.e. WHO
IS THIS? whereas with such small no. of images in my database, i want
to focus on IS THIS THE CORRECT PERSON OR NOT? I am trying to think
as answer to my query in YES or NO rather that 'HE IS MICHEAL'.
Do we have any algorithms which aims at this problem?
I know that I can modify PCA and apply a small condition to check but
why to use such mathematically complex algorithm.
POSSIBLE SOLUTION PROVIDED BY Mr. Sahil in of the posts : As all
images are basically matrices, we take matrices of 2 images and try to
calculate the co-relation factor of the these matrices. If it is found
to be greater than a threshold value we will say that 2 images matched
and this is the right person.
Please comment and if possible, discuss.
Thank You.
santiago everardo
May 18, 2012, 10:51:37 AM5/18/12
to face...@googlegroups.com
Hi Dairu, have you tried using correlation filters?. You require a threshold value to recognize a person. This techniques are more discriminative than PCA.
--
M. C. Everardo Santiago Ramírez
http://www.sciencedirect.com/science/article/pii/S1877705812018152
2012/5/18 Dairu <ahir...@gmail.com>
--
You received this message because you are subscribed to the Google Groups "Face Recognition Research Community" group.
To post to this group, send email to face...@googlegroups.com.
To unsubscribe from this group, send email to face-rec+u...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/face-rec?hl=en.
M. C. Everardo Santiago Ramírez
Reply all
Reply to author
Forward
0 new messages