Scribd
Upload
31 views

Face Recognition:Application of Image Processing

This document presents a method for face recognition using eigenfaces. The method uses discrete wavelet transform and principal component analysis. Faces are preprocessed using DWT to extract time-frequency information. PCA is then applied to generate eigenfaces from a set of training face images. Test images are compared to the eigenfaces to find the closest match. The method reduces dimensionality while retaining important information, and works for both frontal faces in the training set and new faces, achieving accurate recognition results.

Uploaded by

Arka Prava Lahiri
Copyright
© © All Rights Reserved
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
31 views

Face Recognition:Application of Image Processing

This document presents a method for face recognition using eigenfaces. The method uses discrete wavelet transform and principal component analysis. Faces are preprocessed using DWT to extract time-frequency information. PCA is then applied to generate eigenfaces from a set of training face images. Test images are compared to the eigenfaces to find the closest match. The method reduces dimensionality while retaining important information, and works for both frontal faces in the training set and new faces, achieving accurate recognition results.

Uploaded by

Arka Prava Lahiri
Copyright
© © All Rights Reserved
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
You are on page 1/ 6

FACE RECOGNITION:APPLICATION OF

IMAGE PROCESSING
INTRODUCTION
There are many techniques for face detection and recognition. It
can be done using PCA(Principal Component Analysis) or KLT.
Also recognition can be done in the spatial domain or in temporal
domain. Here in this paper , face recognition technique by
eigenface method is presented.
The method is done in the temporal domain using discrete wavelet
transform (dwt) so that we can get information about the time as
well as frequency of any part of the signal simultaneously. Also by
using Daubechies wavelet, first the image is reduced to a smaller
size so that the redundant information in the image and the noise is
eliminated. After these two pre-processing stages are over the main
algorithm is applied over the sample images and the test image for
face-recognition.

METHOD DESCRIPTION
1. Read the M test images.

Here M=7.

2. Do pre-processing on each image.(filtering and


dwt2 transformation).
After transformation-

3. Convert each NxN image into a column vector


(N^2 x 1).
4. Calculate the mean face from the mean of these M
vectors.
5. Subtract the mean face from each image. Let each
such subtracted image be S(i) (i=1,2,...M).
6. Compute the covariance matrix A of the vector got
by concatenating these mean subtracted images.
A=[S(1),S(2),S(3),.....S(M)].
7. Calculate the transpose of A . Let it be denoted y

A(t).
8. Get the M eigenvectors(v(i)) and corresponding
eigenvalues of the matrix A(t)A.
i=1,2,...M.
NOTE- 1.Here we could have calculated
eigenvectors of AA(t) but then number of such
vectors would have been high resulting in high
dimensionality which is a problem.
So again we would have had to reduce the
number to M.
So we A(t)A is computed and its eigenvectors
are taken so that we can get the best M
eigenvectors . Thus both time and dimensionality
is reduced.
2.We can easily get the M eigenvectors of AA(t)
by using u(i)=A*v(i).
9. Take each mean-subtracted face(S) , compute u(i)
(t)*S*u(i) add them together . Thus we represent each
face as a combination of eigenfaces.Let each such
face be F(i) (i=1,2,.....M).
[u(i)(t)=transpose of u(i)].

the eigenfaces for the sample faces are-

10. Get the test image. Here test image is-

11. Repeat steps 2 ,3 ,5 and 9 with test image and get


the approximated face(F)
12.Calculate the distance of this F with each of the
previously calculated faces ,i.e ,F(i) and the one with
lowest distance(also lesser than a threshold distance)
gives the best match of the test image.
Here result is-

This method has many advantages1. Dimensionality is reduced keeping par with
information.
2.It is done in temporal domain so method is not very
elaborate or time-consuming as compared to creating
a feature vector and comparing features(in spatial
domain).
3. It also works if the test image is not among the
images in database because in that case the distance
between the test image and all the sample images will
be more than the threshold.
CONCLUSION
There are certain limitations to this method such as
the face image should be frontal to get better results.
A noisy image causes recognition performance to be
degraded. But considering time and dimensionality
this method is of good use for face recognition and

gives accurate results in most cases.


REFERENCES
[1]. M.A.Turk and A.P. Pentland, Face Recognition
Using Eigenfaces, IEEE Conf. on Computer Vision
and Pattern Recognition., pp586-591, 1991.
[2] Two-dimensional PCA:a new approach to
appearance based face representation and recognition
.http://ieeexplore.ieee.org/xpls/abs_all.jsp?
arnumber=1261097
[3].Human face recogniton using PCA on wavelet
subband .
http://electronicimaging.spiedigitallibrary.org/article.
aspx?articleid=709788

You might also like