CS 175: Project Proposal Guidelines

1
CS 175 Project Proposal Guidelines

• Padhraic Smyth
• Department of Computer Science, UC Irvine
• CS 175, Fall 2007

2
Project Milestones and Deliverables

• Timeline
• Assignment 6 project proposal
• due next Friday, Nov 9th, 12 noon
• (but feel free to submit earlier)
• 2-page progress report
• due Monday Nov 19th, noon
• In-class presentations
• Thursday Dec 6th
• About 4 minutes per student, questions
• Final project reports
• due noon Wednesday December 12th (finals week).

3
Class Grading

• Assignments 1 through 5
• best 4 from 1st 5 assignments are selected
• worth 40 of your grade, 10 for best 4 out of
  5
• Assignments 6 and 7 (project proposal and
  progress report)
• 20 of your grade (10 each)
• Final Project report and in-class demonstration
• worth 40 of your grade

4
Guidelines for Projects

• All projects are individual projects (no group
  projects)
• Discussion of ideas with other students is
  encouraged
• however, no sharing of code with other students
• You can use publicly-available software if you
  wish as part of your project
• MATLAB code made available by researchers on the
  Web
• e.g., other classifiers
• e.g., feature extraction/image-analysis
  algorithms
• You must clearly indicate which (if any) code in
  your project was not written by you and you must
  reference the source.
• You cannot however only use such code in your
  project
• i.e., you need to write at least part of the
  project code yourself

5
Submission of Proposals

• Proposals can be submitted any time next week
• must be submitted no later than noon next Friday
• Upload to Project Proposals in EEE
• Once I approve your project I will let you know
  straight away by email so that you can start
  working on it
• If the project is not approved, I will give you
  feedback, and you must resubmit.
• Detailed instructions will be available on the
  class Web site
• Format of project proposals
• should contain all of the required sections
• please use the supplied proposal template (in
  Word, online)
• should be clear and professional
• will be graded like a regular assignment

6
Recommended Reading (papers on Web page)

• Face Recognition a Literature Survey
• This is a very comprehensive article, but quite
  long, so I don't expect you read all of it.
  Please try to read as much of sections 1, 2, 3
  and 5 as you can. 
• Robust Real-Time Object Detection
• A state-of-the art algorithm for face detection
• Face Recognition Features versus Templates
• a well-written article that describes in detail
  methods and experiments comparing feature-based
  recognition of faces versus template-based
  recognition.
• Read what you can from these papers
• Introductory sections are recommended
• You may find good project ideas and suggestions
  in these papers

7
Optional Reading (on Web page)

• Image Analysis for Face Recognition 
• good survey paper on face recognition.
• Face Recognition HomePage 
• many useful papers here under "Interesting
  Papers", "New Papers", and "Algorithms".
• Neural Network-Based Face Detection
• describes in detail a fairly complex system for
  detecting faces in images using multilayer neural
  networks.
• The FERET Evaluation Methodology for
  Face-Recognition Algorithms
• a paper describing a set of government-sponsored
  tests to evaluate different face recognition
  algorithms and systems.

8
Project Proposals

• Distinct parts to your project proposal
• The specific classification task you will try to
  solve
• The data sets you will use for your experiments.
• The specific image representation (features) you
  will investigate for solving this problem
• The classification (or other) algorithms you plan
  to use.
• An extended task
• A specific plan for how you will test and
  evaluate your algorithms
• Specific milestones and deliverables
• These sections are described in detail in the
  handout for Project Proposals (on the Web page)

9
Part 1 Classification Tasks

• Pose Recognition
• Classify faces into one of 4 classes, i.e., up,
  straight, right, or left. This is a
  generalization of the problem in Assignment 4.
• Sunglasses Recognition
• Classify face images into two classes,
  "sunglasses" or "no sunglasses".
• Expression Recognition
• Classify faces into 4 classes based on what
  expression the person has (angry, happy, sad,
  neutral), e.g., based on straight faces
• Individual Recognition
• Classify the faces according to the name of the
  individual, i.e., into one of 20 classes, given
  any pose or expression for that individual.

10
Easy versus Hard Tasks

• In order of easiest to hardest
• Sunglasses recognition
• Quite easy so you should go beyond the basic
  problem to make this a bit more interesting and
  challenging
• Pose recognition
• Relatively easy (e.g., see assignment 4)
• Individual recognition
• Relatively easy
• Can be made more difficult by cropping the images
• Expression recognition
• Quite difficult

11
Other Possible Classification Tasks

• Combine various options
• Recognize individuals both with sunglasses and
  without
• Recognize individuals using different poses
• Gender recognition
• Classify images of men versus images of women
• Likely to be quite difficult with the default
  images
• Divide individuals randomly into 2 groups
• authorized
• non-authorized
• train only on a subset of images authorized
  individuals
• test on
• (a) other images from authorized individuals
• (b) non-authorized individuals

12
Proposal Part 2 Data Sets

• Standard face image data set we have been using
• 20 different individuals (as in Assignments 4 and
  5)
• 4 basic expressions neutral, happy, angry, sad.
• 4 "poses" straight, up, right, and left.
• For each of the above combinations, there are 2
  images one with sunglasses and one without.
• Total number of images
• for each individual, there are 4 x 4 x 2 32
  different images
• Overall there are roughly 20 x 32 640 images
  available.

13
Image Resolution

• 3 Different Image Resolutions
• Full resolution images are 120 x 128 pixels in
  size
• half-resolution are 60 x 64
• quarter resolution are 30 x 32.
• The smaller images are just averaged versions of
  the larger images.
• The smaller images are easier to download,
  process, and manipulate.
• However, the full images have more detail and are
  recommended
• You are free to use whichever images you wish
• you could compare recognition at multiple
  resolutions if you like
• you could use the smaller images for initial
  development

14
Other Data Sets

• You can use either the standard data set
  (previous slides) or some of these other data
  sets
• Other data sets often have more individuals,
  higher resolution
• Links on the Web page to
• Stirling University face database
• Yale face database
• CMU face image databases
• University of Manchester face images
• Other data sets available under databases at
  Face Recognition Homepage
• at http//www.face-rec.org/

15
Aspects of Other Data Sets

• Formats
• Can use imread to read different image formats
  into MATLAB (e.g., jpg, pgm, gif, etc) may need
  to write a script to do this.
• Color
• Some data sets are in color
• Get 3 color intensities (R,G,B) per pixel rather
  than just 1 grayscale
• More complicated, but better for recognition than
  grayscale
• Resolution
• Resolution is often much higher than our 120 x
  128 default set, especially with newer image data
  sets
• Computationally more intensive, but can give
  better results

16
Part 3 Image Representation

• 2 basic ways to represent images for
  classification
• 1. Pixel Representation
• m x n pixel image gt vector of mn pixel-valued
  features
• i.e., just use the pixels directly as features
• 2. Feature Representation
• extract more abstract features from an image
• e.g., average brightness in a certain region
• e.g., relative location of an eye and nose
  template
• in your project you are required to experiment
  with at least 2 high-level features

17
Examples of Image Features

• Low-level features
• Pixel summaries average, minimum, maximum
  brightness.
• Location and shape information from a thresholded
  version of the image.
• Information from a blurred version of the image
  the location or brightness of the brightest
  "blob" in an 8 x 8 reduced version of the image.
• Higher-level features
• Information from the edge response image, or
  edge map, e.g., fit an ellipse to the edge
  information
• Template-based methods, which use templates (as
  described in class) to estimate where the eyes,
  nose, mouth, ears, etc., are, and then tries to
  measure various features such as their relative
  positions, size, relative brightness, etc.

18
Two Simple Examples of Features

• Example of Edge-based features
• calculate response_image edge(image)
• divide response_image into 16 local subimages
• calculate the average pixel value in each
  subimage
• this gives a vector of 16 local edge features
• Example of Template-based Features
• generate response_image template(image,
  eye_template)
• f1 xmin_location, f2 ymin_location, f3
  min(response_image)
• produces a vector of 3 features
• Feel free to be imaginative and creative in
  designing features!

19
For example..
Feature.m
Input image
Edge features
Template features
Other features
feature vector
20
Part 4 Classification Algorithms

• Minimum-distance classifier
• kNN classifier
• perceptron
• any others that you wish to implement on your own
• e.g., you are allowed to use other
  implementations of classifiers in MATLAB (e.g.,
  available on the Web), such as boosting or
  support vector machines but you must
  acknowledge this in your project writeup.
• You should use at least 2 classifiers in your
  experiments
• kNN and minimum-distance are the easiest to work
  with
• Note data sets, tasks, and features often have
  much more influence on classification accuracy
  than the classifier used

21
Part 5 Extended Tasks

• Effect of resolution on classification
• e.g., compare results at the 3 different
  resolutions
• Effect of pose on classification
• Effect of expression on classification
• Effect of sunglasses on classification
• Add random noise to the image
• evaluate how accuracy degrades as a function of
  noise added
• Simulate obscuring objects
• e.g., add a blank rectangle at random locations
  over the face
• evaluate how this affects accuracy

22
Part 6 Evaluation

• Describe a specific evaluation methodology
• At a minimum, some cross-validation is required
• define which images you will use
• there may be multiple experiments, e.g., full to
  1/4 resolution
• compare pixel-based representation and features
• e.g., pixel-based vs feature set 1
• e.g., feature set 1 vs feature set 2
• robustness
• test how your classifier works on smaller images
• sensitivity
• how sensitive is your classifier to changes in
  number of training images, to the value of k in
  kNN, etc?
• Other options
• Training on data set and testing on another
• Comparing algorithm to human performance

23
Evaluation continued..

• Cross-validation experiment 1
• usual cross-validation methodology
• e.g., v 10 random test subsets
• Cross-validation experiment 2
• do cross-validation using leave-one-out on
  individuals
• e.g., for pose, expression or sunglasses
  recognition
• 19 individuals in the training data
• 1 individual in the test data
• repeat 20 times and average
• a more realistic test of the classifiers
  performance

24
Part 7 Milestones

• Provide a short time-table
• Outline what you plan to accomplish each week
• For example
• Week 1
• Download data, write code for features, try some
  simple kNN classification
• Week 2
• Implement some new features
• Use cross-validation to select/compare different
  features.
• Week 3
• Implement a new classifier
• Perform large-scale cross-validation experiments
• Week 4
• Experiments on extended task, test robustness of
  classifier
• Write final report
• Keep in mind that a progress report will be due
  (worth 10 of your class grade) on Monday Nov
  19th (2 weeks from Monday)

25
Project Degree of Difficulty

• Project needs to have a minimum degree of
  difficulty,
• e.g., repeating Assignment 4 (classification of
  right v. up) will not be eligible for a good
  grade!
• Some problems are easier, some harder
• sunglasses recognition fairly straightforward
• expression classification difficult
• You will be graded on the quality of your work
  relative to the difficulty of the task
• your grade will not depend on the accuracy of the
  classifier as long as a good attempt was made
  (e.g., you could get full points even if your
  classifier is only 60 accurate)
• It is more important to understand why a
  particular technique works (or does not work)
  than it is to get high accuracy

26
Techniques we have not yet discussed..

• It is ok in your project proposal to propose a
  more advanced technique (e.g., eigenimages) that
  we have not discussed in class
• You should have some basic knowledge about the
  technique (find an introductory paper and read
  it)
• You should make sure to include a simpler
  baseline method to compare to
• Topics we will cover in future lectures
• Image segmentation using clustering and region
  growing
• Eigenimages and SVD techniques
• Morphological image analysis
• Image resizing

27
Project Milestones and Deliverables

• Timeline
• Assignment 6 project proposal
• due next Friday, Nov 9th, 12 noon
• (but feel free to submit earlier)
• 2-page progress report
• due Monday Nov 19th, noon
• In-class presentations
• Thursday Dec 6th
• About 4 minutes per student, questions
• Final project reports
• due noon Wednesday December 12th (finals week).

28
Your to do list

• Complete Assignment 5 by Tuesday
• Read Assignment 6 (project proposal instructions)
  carefully by next Tuesday have questions ready
  in class. In Tuesdays lecture we will discuss
  projects again, answer questions, etc.
• Work on your project proposal next week.
• No lecture next Thursday, instead we will have
  office hours from 930 to 11 in Bren Hall 4212.
  Please feel free to come by to discuss your
  project ideas.
• Submit your project proposal to EEE no later than
  noon Friday Nov 9th (next week). I will send you
  an approved/not approved email as soon as the
  project reports are graded.