ELEC4630 Image Processing and Computer Vision
Assignment 3
(Due date: Friday 3/6/2022 at 4pm)
Assignment report should include coding, results, images, and a verbal description of
how you approached the problem. Some similar solutions can be found on the internet, but
you won’t learn anything by copying these verbatim, and you may be flagged for integrity
issues — none of us want this. However, in this assignment you are encouraged to find
internet solutions and modify the code significantly to match your problem. This is fine
and ethical as long as you cite your sources appropriately. Also, the tutors will help you
with the coding. Most importantly, have fun.
Ethics: It is considered to be unethical to disclose a person’s identity via photos and other
means unless you have their full knowing and voluntary consent. For this reason, many
recent large face recognition databases are considered to be unethical since faces were
harvested off the internet. In this assignment, do not show or use images of faces) unless
you have consent in writing, or it is part of a database where I indicate that consent has
been obtained. These assignments will be marked but not distributed or made public by the
teaching team.
Figure 1 Australian Smart Gate
Figure 2 READID Passport Reader
1. In this question I want you to build a bare bones face verification system similar to
what is deployed at Australian Airports for border control. Australia is a leader in
this area and Brisbane was the first city to deploy this technology at the border in
2007.
First, I want you to extract your ID photo from your passport using a mobile
phone reader and the app READID available on iPhone and Android Phones.
You can use this photo or another in your assignment — I suggest you could
use the first photo in the CalTech database as an alternative. This electronic
photo extraction from your passport occurs when your passport is scanned.
Photos are generally removed after the matching step. (Why are they
removed? Why have the photos in the passport chip instead of a central
database?)
Next take a series of similar selfies to simulate photo capture at the border.
Note that in current smartgates, the angle of capture, lighting, pose,
expression, wearing of glasses is tightly controlled (Why?).
Build a deep learning based face verification system to verify your identity
using Dlib or other libraries. There is no need to train the systems.
Plot the ROC curve of your verification system using the Caltech faces as
zero-effort imposters (on Blackboard as CalTechfaces.zip). Comment on
your results. Would this system be satisfactory for the application of border
control? Are the results realistic? Is the database sufficiently large? How
could you make the ROC curve results closer to real life results?
(15 marks)
2. Develop and test some method to distinguish cats from dogs. Feel free to use
existing source code as a basis for your work. Always cite your sources and explain
your modifications. Propose your own architecture and see how it compares to
others in the reference literature. Note that it is OK to use a subset of, say, 2000
images (see b below). It is probably easiest to develop this in the Google Collab
environment.
a. https://www.kaggle.com/c/dogs-vs-cats/overview
b. https://colab.research.google.com/github/google/eng-
edu/blob/master/ml/pc/exercises/image_classification_part1.ipynb#scrollTo
=651IgjLyo-Jx
c. https://towardsdatascience.com/image-classifier-cats-vs-dogs-with-
convolutional-neural-networks-cnns-and-google-colabs-4e9af21ae7a8
d. https://machinelearningmastery.com/how-to-develop-a-convolutional-
neural-network-to-classify-photos-of-dogs-and-cats/
Figure 3 Cats versus Dogs
(15 marks)
(Total 30 Marks)
Assignment 4 Marking Scheme and Criteria
Q1
Coding of a solution to Australian border control. Appropriate images, description
and explanation of method, issues of implementation, related images and graphs
o (8 marks)
Explanation of the requirements of a border control biometric solution regarding
privacy efficiency etc. Answers to inline questions.
o (7 marks)
Q2
Coding of solutions to perform cat and dog classification. Description and
explanation of architectures, performance metrics, appropriate citation of sources.
o (8 marks)
Evaluating and explaining your own unique architecture
o (7 marks)