Facial Matching with Age and Gender Prediction: AI Project

This project delves into the application of machine learning, specifically deep learning and convolutional neural networks (CNNs), for paired facial matching with age and gender prediction. The research investigates the use of algorithms to analyze facial attributes, aiming to predict age and gender from a single face image. The methodology involves utilizing libraries like OpenCV and Pafy to process images and videos, with the CNN model trained on captured data. The project addresses key research questions regarding dataset clarity, the suitability of CNNs, and the accuracy of age prediction. It also explores the impact of image quality and the use of video data. The project follows ethical research guidelines, using secondary data sources and ensuring data privacy. The analysis involves image pre-processing techniques, including converting images to grayscale and scaling, followed by the application of the CNN model for classification. The project also discusses the layers of the convolutional neural network and the functionality of different components of the model such as Conv layer, ReLu and Pool layer. The aim is to optimize the model for accurate facial attribute prediction, with considerations for future work involving video analysis and streaming applications.

Running head: PAIRED FACIAL MATCHING WITH AGE AND GENDER PREDICTION
Paired Facial Matching With Age and Gender Prediction
Name of the Student:
Student ID:
Name of the University:

2Paired Facial Matching With Age and Gender Prediction
Table of Contents
Abstract............................................................................................................................................3
Introduction......................................................................................................................................3
Research Questions..........................................................................................................................5
Research Methodology....................................................................................................................5
References........................................................................................................................................7

3Paired Facial Matching With Age and Gender Prediction
Abstract
Machine learning algorithms are used widely for various purposes. Machine learning algorithm
is considered to be a subpart of the Artificial intelligence. Simply it can be said that with the
recent evaluation of machine learning it has now become possible to predict future instances and
forecast data according to historical data. Machine learning models has the ability to learn from
historic data which helps to find hidden patterns and to make fruitful decisions with minimum
human interaction. Machine learning models are also known as predictive models.
One can detect age and gender much easily because with the increasing amount of social
platform and social media. Performing classification or detection is still significantly lacking
when performed on real world images which huge volume of data. Thus we can say
convolutional neural network is considered to be as a significant choice which can be
implemented using deep learning technique to detect age and gender together.
Introduction
For the analysis and prediction, age and gender are the two main facial attributes which
play a vital foundational place in the social media interaction through which the age and gender
can be identified using a single face image. It can be done using intelligent system or application
such as access control, human-computer interaction, law enforcement and marketing intelligence
(Antipov, Berrani and Dugelay, 2016). Moreover there were many works which earlier but faces
various issues and also the accuracy of the model did not turn up very well (Kingma & Adam,
2014).
Moreover it is said that many datasets use for such analysis doesn’t fit the benchmark for
age and gender detection (Eidinger, Enbar & Hassner, 2014). These type of images faces major
problems and challenges in the real world (Fu, Guo & Huang, 2010). Most of the times the
image may be too much blur in terms of low resolution then occlusions also the image may be
out-of-plane and many more (Golomb, Lawrence & Sejnowski, 1990). Thus the proposal is an
attempt to minimize the gap between the face recognition capabilities and those of age and
gender estimation methods (Jain & Learned-Miller, 2010).
We consider deep learning to be a subfield of machine learning where the algorithm is
used as the structure and the function of the brain which is basically called the artificial neural
network (Jia et al., 2014). Deep learning algorithm comprises multiple layers of different

4Paired Facial Matching With Age and Gender Prediction
components of the neural network which are interconnected. At the end the output layer
produces the predicted result with the results of all the other combined layers (Krizhevsky,
Sutskever & Hinton, 2012). Using deep learning image and video recognition, recommendation
system, natural language processing, media recreation, image analysis and many more are now
possible and are implemented widely (Kwon & Vitoria Lobo, 1999). The recent advancement in
the field of computer vision with the help of Deep learning has created far reaching impact on
technologies over the time. Especially it has create huge impact over a particular algorithm
which is mainly the Convolutional neural network (LeCun, Bengio & Hinton, 2015).
One of the most commonly used deep learning algorithm is i.e. CNN which takes the
input image and assign importance in the form of weights to every objects in the image and can
distinguish one from the other (LeCun et al., 1998). In convolutional neural network the pre-
processing step required is much less as compared to other classification or predictive algorithms
available (Li et al., 2015). The main advantage of these algorithm is that while in other
algorithms filters is placed by hand engineering, here the ConvNets can learn these
filters/characteristics (Nguyen et al., 2015). An assumption is being made that inputs form of
images by the ConvNet architectures is helpful to encode specific features into the architecture
(Levi & Hassner, 2015). Thus helps the model to use less parameters in the network (Makinen &
Raisamo, 2008). The major functionality of using ConvNets is that without losing feature the
ConvNets reduce the image into smaller forms which can be easier to process for getting a good
prediction.
The layers of the convolutional neural network is being discussed below-
 The input mainly comprise the raw pixel values of the images which will use for
classification and the height and width and the color channels will be the parameters of
the input network.
 For computing the outputs of all the neuron which are connected CONV layer is
responsible for this (Parkhi, Vedaldi & Zisserman, 2015). Also responsible for computing
the dot product between their weights and with a small region where it connects with the
input volume of the layer. According to the filters the volume changes.
 ReLu known for activation function and is applicable element wise, such as the max (0,
x) thresholding at zero. This this result in unchanged of volumes.

5Paired Facial Matching With Age and Gender Prediction
 Down sampling plays a crucial role along with spatial dimensions basically the width and
height which results in decreasing of the parameters and thus the layer responsible is the
Pool layer.
 At the end fully connected have implemented to compute the class score where each
vector represent a particular class. Thus, each neuron in this layer connects to all the
numbers in the previous volume.
Research Questions
 Whether the dataset used for the analysis is clear without distortion or not.
 Whether the background is bright enough to capture the parameters of the face?
 Whether CNN is the appropriate predictive model to predict age and gender and why
chosen.
 Whether the model predict the exact age or does it produced a ranges where the
predictive age lies and how the accuracy changes with change in layers of the
convolutional neural network.
 Whether detected faces are from still images or from video which impact a crucial factor
for classifying the required parameters correctly.
 Is the proposed model or the algorithm optimal?
Research Methodology
Pafy is a kind of library which can retrieve the content from the YouTube platform. We
will need OpenCV to work with images or videos thus OpenCV will provide the easiest and
reliable method to import Haar-cascades and will be used for detecting faces directly (Mathias et
al., 2014). Using the libraries available for image processing, deep learning models basically the
CNN model will be implemented to train the captured images. Thus the analysis will be done
using OpenCV and with the help of DNN which stands for Deep Neural Network (Moghaddam
& Yang, 2002).
The research will follow every ethical norms prescribed under the academic research
guidelines. We will collect data from the secondary sources or from any videos available online.
We will not disclose details of person involved in the research (Szegedy et al., 2014). There will
be no tampering of data during analysis of data. Different public data as well as data from other
sources are available which can be utilized for the same.

6Paired Facial Matching With Age and Gender Prediction
The age can be detected using probability layer which mainly depicts that CNN’s output
layer (probability layer) in this CNN comprises of 8 values for 8 age classes (“0–2”, “4–6”, “8–
13”, “15–20”, “25–32”, “38–43”, “48–53” and “60 and above”). Different pre-processing of
images performed which includes converting a RGB image to grey scale with different scaling
factors including scaleFactor which mainly use for scaling the size of the image and
minNeighbors is responsible for detection of objects.
OpenCV provides a function namely blobFromImage() which is specifically used for
image-processing and for deep learning classification problems. It takes mainly 5 parameters
usually the image which will be the input of the neural network and the scaling factor which
comprise floating point number which shows the size of the scaling factor (Tronick & Cohn,
1989). Then the size is being supplied to the Convolutional Neural Network and most of the time
the size given is 224×224, 227×227, or 299×299. Then the mean which holds the mean
subtraction values of the image and at the end swapRB assumes by default the images are in
BGR order.
The method is much advanced and is implementable as python provides a variety of
libraries which help the processes to run smoothly. There are various works which are related to
this topic hence further implementation and extension of work need to be carried out to get
optimize result. Using video can be a challenge as the objects are movable and will be difficult to
detect. Thus, using videos or streaming these concept need to be develop.

7Paired Facial Matching With Age and Gender Prediction
References
Antipov, G., Berrani, S. A., & Dugelay, J. L. (2016). Minimalistic CNN-based ensemble model
for gender prediction from face images. Pattern recognition letters, 70, 59-65.
D. P. Kingma and J. Ba. Adam: A method for stochastic optimization. CoRR, abs/1412.6980,
2014.
Eidinger, E., Enbar, R., & Hassner, T. (2014). Age and gender estimation of unfiltered faces.
IEEE Transactions on Information Forensics and Security, 9(12), 2170-2179.
Fu, Y., Guo, G., & Huang, T. S. (2010). Age synthesis and estimation via faces: A survey. IEEE
transactions on pattern analysis and machine intelligence, 32(11), 1955-1976.
Golomb, B. A., Lawrence, D. T., & Sejnowski, T. J. (1990, October). Sexnet: A neural network
identifies sex from human faces. In NIPS (Vol. 1, p. 2).
Jain, V., & Learned-Miller, E. (2010). Fddb: A benchmark for face detection in unconstrained
settings (Vol. 2, No. 6). UMass Amherst technical report.
Jia, Y., Shelhamer, E., Donahue, J., Karayev, S., Long, J., Girshick, R., ... & Darrell, T. (2014,
November). Caffe: Convolutional architecture for fast feature embedding. In Proceedings
of the 22nd ACM international conference on Multimedia (pp. 675-678).
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). Imagenet classification with deep
convolutional neural networks. In Advances in neural information processing systems
(pp. 1097-1105).
Kwon, Y. H., & da Vitoria Lobo, N. (1999). Age classification from facial images. Computer
vision and image understanding, 74(1), 1-21.
LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. nature, 521(7553), 436-444.

8Paired Facial Matching With Age and Gender Prediction
LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning applied to
document recognition. Proceedings of the IEEE, 86(11), 2278-2324.
Levi, G., & Hassner, T. (2015). Age and gender classification using convolutional neural
networks. In Proceedings of the IEEE conference on computer vision and pattern
recognition workshops (pp. 34-42).
Li, H., Lin, Z., Shen, X., Brandt, J., & Hua, G. (2015). A convolutional neural network cascade
for face detection. In Proceedings of the IEEE conference on computer vision and pattern
recognition (pp. 5325-5334).
Makinen, E., & Raisamo, R. (2008). Evaluation of gender classification methods with
automatically detected and aligned faces. IEEE transactions on pattern analysis and
machine intelligence, 30(3), 541-547.
Mathias, M., Benenson, R., Pedersoli, M., & Van Gool, L. (2014, September). Face detection
without bells and whistles. In European conference on computer vision (pp. 720-735).
Springer, Cham.
Moghaddam, B., & Yang, M. H. (2002). Learning gender with support faces. IEEE Transactions
on Pattern Analysis and Machine Intelligence, 24(5), 707-711.
Nguyen, D. T., Cho, S. R., Pham, T. D., & Park, K. R. (2015). Human age estimation method
robust to camera sensor and/or face movement. Sensors, 15(9), 21898-21930.
Parkhi, O. M., Vedaldi, A., & Zisserman, A. (2015). Deep face recognition.
Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S. E., Anguelov, D., ... & Rabinovich, A.
(2014). Going deeper with convolutions. CoRR, vol. abs/1409.4842.
Tronick, E. Z., & Cohn, J. F. (1989). Infant-mother face-to-face interaction: Age and gender
differences in coordination and the occurrence of miscoordination. Child development,
85-92.