Abstract Title

Comparative Graph Model

Abstract

Comparative Graph Model is a facial recognition algorithm that takes several images of the same face, and identifies common features of that face. The algorithm looks at the closeness of features along with how many times each exists, to decide which features are the best representation of a face. The algorithm takes features from several images and condenses them down to one image, which is then used for comparison to other images. This algorithm uses a different approach than many facial recognition algorithms, which use a template to seek out facial features and make associations. CGM uses only data that is present to form associations. This makes the algorithm unsupervised, and can easily be used for other domains like object recognition. In this experiment, CGM is used in conjunction with a perceptron neural network to compare positive and negative images. Ten training images were condensed to one image via CGM, and compared against one positive image (the same person), and one negative image (a different person). In the experiment, each trial was run for 100 iterations. The algorithm recognized the same person 71% correctly, did not recognize a different person 79% correctly, with an overall correctness of 61%. Using an unsupervised algorithm is beneficial across many disciplines, especially where a pattern can be too hard to find or represent. The algorithm does not need to know what it is looking for to make associations, which makes it a good candidate for finding complex patterns and associations in data.

Modified Abstract

Many facial recognition algorithms use template-based approaches to find features in a facial image that are used for authentication. This works by associating certain features with commonly known features of a human face. This can miss or inaccurately identify features. Comparative Graph Model does not use a template, but instead looks for closeness of features across images, along with how many times a feature is found, in order to decide what features accurately represent a person’s face. This experiment uses Comparative Graph Model along with a perceptron neural network to create feature maps that are used for comparison to other images. Comparative Graph Model offers the advantage of versatility as it allows for features of any type of data, not just human faces.

Research Category

Computer Science/Mathematics

Primary Author's Major

Computer Science

Mentor #1 Information

Mr. Mehdi Ghayoumi

Presentation Format

Poster

Start Date

21-3-2017 1:00 PM

Research Area

Artificial Intelligence and Robotics

Share

COinS
 
Mar 21st, 1:00 PM

Comparative Graph Model

Comparative Graph Model is a facial recognition algorithm that takes several images of the same face, and identifies common features of that face. The algorithm looks at the closeness of features along with how many times each exists, to decide which features are the best representation of a face. The algorithm takes features from several images and condenses them down to one image, which is then used for comparison to other images. This algorithm uses a different approach than many facial recognition algorithms, which use a template to seek out facial features and make associations. CGM uses only data that is present to form associations. This makes the algorithm unsupervised, and can easily be used for other domains like object recognition. In this experiment, CGM is used in conjunction with a perceptron neural network to compare positive and negative images. Ten training images were condensed to one image via CGM, and compared against one positive image (the same person), and one negative image (a different person). In the experiment, each trial was run for 100 iterations. The algorithm recognized the same person 71% correctly, did not recognize a different person 79% correctly, with an overall correctness of 61%. Using an unsupervised algorithm is beneficial across many disciplines, especially where a pattern can be too hard to find or represent. The algorithm does not need to know what it is looking for to make associations, which makes it a good candidate for finding complex patterns and associations in data.