You are on page 1of 2

Group Project

Facial Recognition via Convolution neural network

Deep Learning-AI

GROUP 11

FT191094 Sumit Sharma


FT191101 Vibhor Jain
FT191028 Dip Joy Choudhury
FT191050 Neha Bhagchandani
FT192076 Sanchit Arora
FT192078 Sanjna V Gopal
Facial recognition using Convolution neural network

Procedure followed

Image--Convolution Maxpooling Flattening-Neural network

 We created two convolution layers of 32 feature maps of size 3*3 and used ‘Relu’ activation
function, the detection system captures a part of the face and separate detection systems or
detectors are created for different parts of the face. Images detected by the filters are
convoluted at the end.
 8000 images were not enough to run this model, we have taken existing images, flipped and
resized them to increase the total no. images.
 Maxpooling was applied using 2*2 matrix.
 We have added two hidden layers with 128 nodes.
 Transformation was carried out on training dataset.
 Post transformation we have done re-scaling of both train and test dataset by 1/255.
 Flip flopped training data set.
 Created the training and test dataset.
 Created the classifier.

Test Cases

Test Case Description Accuracy Image Detected


(Yes/No)
0 Base Code

1 Samples Per Epoch is


5000 and validation is
1000
2 Samples Per Epoch is
10000 and validation
is 4000
3 Using RMS Prop _
baseline
4 BatchSize is 25 in train
and test
5 Leaky Rely Twice -
One after Input and
One 2nd before
output
6 MaxPooling at 4 after
input
7 Relu leaky with 0.1

Relu leaky with 0.2


8

You might also like