This document discusses gender and age classification using facial features. It explains that while humans can easily determine gender and estimate age from faces, it remains challenging for computers. The document outlines an approach that uses a convolutional neural network (CNN) trained on facial images and metadata scraped from LinkedIn profiles. Baseline classifiers using logistic regression are compared to the CNN, showing the CNN achieves 81% accuracy for gender classification and 68% for age estimation. Future work to improve accuracy is proposed, such as collecting more training data.

1. Gender and Age
Classification using Facial
Features
DR. Aron Culotta
Harsh Parikh

2. What is Face Detection & Recognition??
The difference between face detection and recognition is that in detection
we just need to determine if there is some face in the image, but in
recognition we want to determine whose face it is.

3. Face Detection
In Face Detection, we need to be able to define the general
structure of a face.

4. Face Recognition
Identify person

5. Motivation
Automatic age and gender
classification has become relevant
to an increasing amount of
applications, particularly since the
rise of social platforms and social
media. Nevertheless, performance
of existing methods on real-world
images is still significantly lacking,
especially when compared to the
tremendous leaps in performance
recently reported for the related
task of face recognition.

6. Problem
 Age and Gender Classification using Facial
Features. A human can easily make these estimates
from faces. Yet, it is still a challenging task for a
computer.
 Even just with facial images, humans would use
some high-level information, like the skin tone, the
hairstyle, the facial hair. That can be hard for the
prototype system to extract. On the other hand the
system will use computer vision techniques that
humans are not able to use.

7. Approach
{ 'Full_Name': 'Eryn Olson',
'Gender': 'Female',
'Recommended_Ids': ['erynolson', 'eryn-olson-62639510a',
'eryn-olson-432679b1', 'eryn-olson-1351aa74', 'courtney-
tillman-8b62a664', 'chance-cozby-13235224', 'bill-gates-
b1a606b0', 'sarah-eves-2937a719', 'andrew-solheim-
810a3517', 'andrea-cundiff-76020535', 'mollie-harper-
6b850947', 'tyler-shaw-934b7', 'donna-conroy-66987340',
'paul-wood-73455920'],
'User_ID': 'eryn-olson-50328143',
'age': 20 }
 Scrap LinkedIn Profiles
 Get recommended ids list
 Find Age and predict gender
 Store all images and labelize them

8. Baseline classifier
 Feature Extraction
o Face detection from picture
o Edge Detection
 Logistic Regression
o Basic Classification method
o Predicts binary response

9. Baseline Classifier Results (Gender)
With edge
Without edge

10. Baseline Classifier Results (Age)
With edge
Without edge

11. CNN or ConvNet
A Convolutional Neural Network (CNN) is comprised of one or more convolutional layers and then
followed by one or more fully connected layers as in a standard multilayer neural network.

12. Contd.

13. CNN Results for Gender
Accuracy on 1000 epochs
Accuracy on 200 epochs
Gender classification for whole image
Testing Accuracy = 0.81

14. CNN Results for Age
Accuracy on 1000 epochs
Accuracy on 200 epochs
Testing Accuracy = 0.68

15. Cause of accuracy failure

16. Conclusion / Future Work
 Keep collecting more data to get better accuracy.
 Try to diminish testing error and remove noise from
data.
 Use PCA for feature reduction
 Use more age group, and predict the age.