COMP3007 Computer Vision Coursework Description

Xin Chen & Andy French, 2019-20
1 Introduction
Face recognition has always been one of the hottest topics in computer vision for
decades. It is extremely useful in real-world applications, such as security, surveillance,
robotics, etc. With the advanced algorithm development in computer vision, more and
better methods have been proposed to address challenging face recognition problems,
such as poor lighting, different facial poses, occlusions, etc.
In this coursework, you will be provided with a public face database that contains
multiple face images from 100 subjects. The face images were captured in different
poses and different lighting conditions. You will use one face image from each subject
to train/build your computer program and recognise the remaining face images of
these subjects.
2 Key dates
Submission deadline of Matlab code and report: 5th May 2020.
3 Detailed requirements
Dataset:
You will be provided a face database from 100 subjects for developing and
evaluating your face recognition computer program. You are only allowed to
use the training dataset (one face image per subject) to train your method, the
test dataset (total of 1344 images) is used for evaluation purpose. The true
face IDs for the test dataset are saved in ‘testLabel.mat’. When assessing your
method, we will use an independent dataset (hidden from you) that have the
same format to test your methods.
Method:
You will be guided in the lab session to build a simple face recognition program,
which is a baseline method. You will then be asked to implement two alternative
methods which should achieve better recognition accuracy than the baseline
method. Potential methods will be introduced (but not in great detail) in the
lectures.
Matlab code:
You need to implement the algorithm using Matlab (no other programming
language accepted). Example files “Evaluation.m” and “FaceRecognition.m”
for the baseline method will be provided. You must design your main files
following the format in the example files. When assessing your code, we will
run the “Evaluation.m” file. Implement the two face recognition methods as
functions with the format of:
[RecFaceID]=FaceRecognition1(“TrainDirectory”,“TestDirectory”)
[RecFaceID]=FaceRecognition2(“TrainDirectory”, “TestDirectory”).
Save all .m files and the final report into a single folder and compress it into a
single .zip file for submission. The input parameter “TrainDirectory” is the folder
path that contains the training images. “TestDirectory” is the folder path that
contains the test images. The output RecFaceID is a vector that contains the
output recognised face ID that corresponds to the set of test images. Following
this format is important in order for your work to be properly marked.
Report:
You also need to submit a report that describes your work. A template in word
and Latex are provided on Moodle, which is an IEEE conference paper format.
You need to follow the template format in terms of font size and layout (double
column). In the report, you must include the following sections: Abstract,
Introduction, Methodology, Method Evaluation, Conclusion and Reference. You
are expected to include evaluation results of all three methods (i.e. baseline
method and two alternative methods). The length of the report needs to be
minimum of 3 pages but no more than 4 pages (Reference could be in the
5th page). Scientific writing will be introduced in one of the tutorials.
4 Marking Criteria
Matlab code 40%
Result evaluation
20%
The mark is objectively produced that is proportional to the recognition
accuracy.
Computation speed
15%
The mark is objectively produced that is proportional to the
computational speed.
Coding style
5%
Robustness of code and coding style
Report 60%
20% Description of methodology
15% Explanation and presentation of the results obtained.
15% Discussion of the strengths and weaknesses of the chosen approach
and methods
10% Scientific writing and clarity
5 Plagiarism
Copying code or report from other students, from previous students, from any other
source, or soliciting code or report from online sources and submitting it as your own
is plagiarism and will be penalized as such. FAILING TO ATTRIBUTE a source will result
in a mark of zero – and can potentially result in failure of coursework, module or degree.
All submissions are checked using both plagiarism detection software and manually
for signs of cheating. If you have any doubts, then please ask.