|
| Module Availability |
| Spring Semester |
|
|
| Assessment Pattern |
|
Unit(s) of Assessment
|
Weighting Towards Module Mark(%)
|
|
Examination
|
80%
|
|
Practical Assignment
|
20%
|
|
|
|
| Module Overview |
The module delivers a grounding in Computer Vision, for students with a background in linear algebra similar to that provided in ee2.cvg (graphics) at Level 2 undergraduate, or eem.ipv (image processing and vision) at MSc level. Content is presented as an application-focussed tour of Computer Vision from the low-level (image processing), through to high level model fitting and object recognition.
The module first introduces low-level image processing, discussing how edges may be detected, and how regions of interest may be identified using simple colour classifiers. Mid-level scene representation is then discussed in the context of global shape descriptors and local feature descriptors. These descriptors are combined with knowledge of machine learning; simple classifiers to explore supervised classification problems (shape and object recognition) and applications of unsupervised clustering (e.g. codebook based image retrieval). The latter is explored more deeply through coursework assignments. Dynamical models are then presented in the content of object tracking, with examples of classical and contemporary tracking algorithms. High-level scene description is briefly explored using statistical models of shape. Finally, models of camera geometry and image formation are presented, and their applications to 3D reconstruction are explored. Taught material is reinforced through lab-based exercises in Matlab. |
|
|
| Prerequisites/Co-requisites |
ee2.cvg or similar background in geometric interpretation of Linear Algebra (e.g. homogeneous coordinates and matrices for point transformation e.g. rotation, translation, scaling).
ee3.ood or similar programming background in C
There is a self-test available to check the requisite maths (and a pre-study guide if needed) at:
http://info.ee.surrey.ac.uk/Personal/J.Collomosse/CVPR |
|
|
| Module Aims |
| The module teaches the mathematical principles and concepts of computer vision alongside its practical applications. The module aims to provide a first course in computer vision, encompassing: image formation and low-level image processing; mid-level scene representation; model-based description and tracking. |
|
|
| Learning Outcomes |
On completion, students should be able to:-
· Identify and implement appropriate solutions to low, mid and high level Computer Vision problems.
· Represent problems as a mathematical models and apply appropriate machine learning and optimization techniques to solve those problems.
· Understand the effect of digital image processing operations in terms of the spatial and frequency domain.
Understand and apply statistical representations of static and dynamic objects to enable their detection, classification and/or tracking.
|
|
|
| Module Content |
COMPUTER VISION (30h Lectures)
[1] Introduction to Computer Vision and its Applications.
[2-3] Image Processing: Convolution and Linear filters. Edge detection. Image Interpolation.
[4-5] Pattern classification: Supervised clustering; K-NN; Thresholding and decision boundaries; Eigenmodels and Mahalanobis distance; PCA; Applications (segmentation and visualisation).
[6-9] Recognition and Classification: Feature Descriptors (SIFT, HoG). Concept of a feature space. Unsupervised clustering (K-Means). Visual codebooks. Overview of Support Vector Machines (SVMs). Applications to object classification and video analysis.
[10-12] Shape Description: Moments, Fourier descriptors, Shape Contexts.
[13-16] Tracking: Templates and cross-correlation. Blob trackers. Kalman filter. Bayes Law. Particle filters.
[17-21] Model based vision: Active contours; PDMs; ASMs. Radon transform. RANSAC.
[22-23] Geometry of Cameras: Camera models (orthographic, affine, projective), camera calibration methods.
[24-30] Multiview Geometry: Epipolar Geometry; Fundamental Matrix estimation; Homography and Mosaicing; Bundle Adjustment; Stereo rectification and Depth Estimation; Visual Hull. |
|
|
| Methods of Teaching/Learning |
Lectures 3hrs/week for 10 weeks
Laboratories 1hr/week for 6 weeks:
Weeks 2-3: Image processing exercises using Matlab
Weeks 3-4: Shape and pattern classification exercises using Matlab
Weeks 5-6: Tracking exercises using Matlab
Weeks 3-7: Computer Vision coursework (OpenCV) |
|
|
| Selected Texts/Journals |
|
Szeliski, R.
|
Computer Vision: Algorithms and Applications
|
1848829345
(released Oct ’10, full draft online)
|
Springer
|
£54
|
B
|
|
Forsythe,D. and
Ponce
, J.
|
Computer Vision: A Modern Approach
|
0131911937
|
Prentice-Hall
|
£34
|
B
|
|
Nixon, M. and Aguardo, A.
|
Feature extraction and image processing
|
0750650788
|
Elsevier
|
£40
|
B
|
|
Trucco, E. and Verri, A.
|
Introductory Techniques for 3D Computer Vision
|
0132611082
|
Prentice-Hall
|
£30
|
B
|
|
|
|
| Last Updated |
| 12 August 2010 |
|
