Muhammed Hassan M (91617011)

AI based facial recognition for unique identification of animals - MSC MI 2017-2019

I am part of a project that has the overall scope as Animal face detection
and recognition using deep learning techniques.Animal Facial Recognition is
a fairly complex task as it would require using a unique identier that distinguishes
one animal in a species from another. This will involve identifying
unique characteristics for an animal that makes it identiable. The possibilities
being explored include colour, texture, face geometry and birth patterns
to uniquely identify an animal from all four sides (front, back and two sides).
The animals being considered are cows, cats, dogs and rabbits.The scope of
my project work is to extract features to determine the unique geometry of
the frontal face. To extract the face geometry, identifying facial landmarks of
the selected animals was the rst step. This is something that has to be built
from ground up as facial landmark denitions of these animals are not available
in public domain.I started by trying out a face landmark detection with
simple neural net and annotated landmarks in a CSV le. While this technique
works well with human face detection, the results with animals were
not as expected. I then tried Mask RCNN based approach. M-RCNN is a
state-of-the-art model for object detection. It is a deep neural network aimed
to solve instance segmentation problem in machine learning or computer vision.I
have now created a model - Mask-Region based Convolutional Neural
Network(M-RCNN) for identifying the facial landmarks like eyes, muzzle etc.
of cattle.The methodology included creating an annotated landmark data for
training and building a model for testing.It only takes the region of interest
within an object and these regions will be passed to the further neural network
to give the masked region of the object. For pre-processing of images
and feature extraction, several methods including Template Matching, Canny
Edge Detection, transformation etc. were employed.

Subjects--Topical Terms:
ANIMAL FACIAL RECOGNITION
DEEP LEARNING
M-RCNN
ARTIFICIAL INTELLIGENCE