Synopsis For Face Recognition Algorithm

DEPARTMENT OF ELECTRONICS AND COMMUNICATION ENGINEERING
IMPLEMENTATION OF SYMMETRY MEASURES IN FACE RECOGNITION ALGORITHM FOR EFFICIENT COMPUTATION
Abstract
Face recognition is an important area of computer vision research and has gained signicant interest in recent years. Recent research in the area of automatic machine recognition of human faces has shown that there may be an advantage in utilizing face symmetry to improve recognition accuracy. Our project aims to find if there is a statistical significance between face symmetry and face recognition. Also, we aim to highlight its significant advantage over the currently existing systems.
Introduction
It is well-known that the face exhibits reectional symmetry about a bilateral symmetry axis. Additionally, research into using the symmetry of the face to assist with face detection and face recognition has been previously studied. The average-half-face exploits facial symmetry by dividing the frontal full face into two halves about the bilateral symmetry axis, mirroring one of the halves across the symmetry axis, and then averaging the two resulting images. The use of the average-half-face in face recognition research has shown a potential increase in accuracy and decrease in storage and computation time as compared to using the original full face.
Implementation Average-Half-Face
The average-half-face is an average of the two halves of the face along a bilateral symmetry axis of the face, which has proven to produce better results than the original full face. Algorithm Outline: 1. Center and orient the face in the image. 2. Partition the image into two equal left and right halves, Il and Ir. 3. Reverse of the ordering of (or mirror) the columns of one of the images (in this case the right image), producing Ir0. 4. Average the resulting mirror right half image (Ir0) with the left half image (Il) to produce the average-half-face. The average-half-face is then used in place of the full face image for face recognition.
Face Recognition Algorithm In our project, eigenfaces is used as the face recognition algorithm. We choose to use eigenfaces because of its simplicity and because it highlights the difference in face recognition accuracy between using the full face and the average-half-face. Eigenfaces is based on Principal Components Analysis (PCA) and is a common face recognition algorithm for benchmarking. PCA captures as much of the variance in the data as possible in as few principal components as possible. All images are projected into the same subspace dened by a chosen number of principal components and test images are classied in the subspace using nearest neighbors.
Implementation in MATLAB The 2D Image for which the computation is to be done is opened using the MATLAB application. Using existing Face Detection Algorithms, the facial part of the subject is extracted. In the next part, average-half-face of the image is obtained using the existing techniques. Using Principle Component Analysis (PCM) algorithm, the average-half-face is compared against the 2D images present in the database. By the application of the symmetry measure as defined, the computation efficiency for subjects of different symmetry measure is calculated.
Flowchart
START
OPEN 2D IMAGE IN MATLAB
FACE DETECTION ALGORITHM
COMPUTATION OF AVERAGE-HALF-FACE
COMPARE AVERAGE-HALFFACE WITH 2D IMAGE DATABASE
TABULATE RESULTS FOR DIFFERENT SYMMETRY MEASURES
END
Measuring Symmetry
We have to note that, it is difcult to nd a measure that encapsulates the symmetry of the face in a single number that is used to easily compare the symmetry between different faces. Although, some symmetry measures can be based on feature points on the face and the relationships between these points. However, these feature points are not readily available on every face database and require manual supervision for reliable accuracy. We have adapted one previous method for measuring symmetry, known as the density difference (D). The measure is formulated by: D(i, j) = I(i, j) I(i, j) (1)
Where I(i, j) is a pixel from one half of the image and I(i, j) is a pixel from the mirror of the other half of the image. The resulting density difference D is itself an image which displays the
asymmetry present in the face. However, we desire a single value, or score, for the symmetry of each individual face for comparing the symmetry of many faces. Therefore, we dene the following scores: Sum of absolute differences (s-score) Symmetry proportion (p-score) s-score applied to Gaussian smoothed image (sg-score) p-score applied to Gaussian smoothed image (pg-score)
The s-score is a simple extension of the density difference and is dened as: s = i, j |D(i, j)| (2)
In addition to the s-score, we introduce a symmetry proportion score (p-score) that is bounded between 0 and 1 and may give a better intuition for the symmetry of the face. The p-score is dened as: p = 1 - i, j T(i, j) / N (3)
where T(i, j) is 0 if the absolute difference of the pixels is less than a certain threshold and 1 otherwise and N is the total number of pixels used in the symmetry score. It is apparent that faces that are highly symmetric will give a p-score that is close to 1.
Face Recognition Results

We have partitioned the database into most symmetric and least symmetric subjects according to the average s, p, sg and pg-score values of each of the subjects. Now we will investigate the face recognition results of every image independently and compare their results using both the original full face and the average-half-face.
Conclusions and Future Enhancement

We aim to utilize the relationship between the symmetry of the face and face recognition in current techniques. We have used symmetry scores to compare most symmetric and least symmetric subgroups. Future work will include introducing additional symmetry measures as well as extending this analysis to more databases of 2D and 3D faces. The ultimate goal would be to create a correlation between the symmetry of the face and face recognition that could be used to improve the overall face recognition accuracy.

Synopsis For Face Recognition Algorithm

Uploaded by

Copyright:

Available Formats

Synopsis For Face Recognition Algorithm

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Synopsis For Face Recognition Algorithm

Uploaded by

Copyright:

Available Formats

DEPARTMENT OF ELECTRONICS AND COMMUNICATION ENGINEERING

IMPLEMENTATION OF SYMMETRY MEASURES IN FACE RECOGNITION ALGORITHM FOR EFFICIENT COMPUTATION

OPEN 2D IMAGE IN MATLAB

FACE DETECTION ALGORITHM

COMPARE AVERAGE-HALFFACE WITH 2D IMAGE DATABASE

TABULATE RESULTS FOR DIFFERENT SYMMETRY MEASURES

Face Recognition Results

Conclusions and Future Enhancement

You might also like