bild
Skolan för
elektroteknik
och datavetenskap

Image Based Recognition and Classification: Course Aims

Intended Learning Outcomes

Concretely upon completion of this course a student should be able to:

ILOs

Theory

  • Be able to describe the supervised and unsupervised learning classifiers encountered in the course (Boosting algorithm, SVMS, clustering, perceptron classifier etc.) at a high level, mathematically and/or algorithmically in these terms
    • What criterion the classifier optimizes subject to which constraints
    • How the optimization is performed during training
    • What training data data is required
    • How a test example is classified
    • What is the final form of the classifier and the decision boundary it creates in the feature space
    • Computational issues
    • Overall limitations and strengths of the method
  • Derive Bayes' rule from the definitions of conditional probability and be able to apply it to problems where a decision has to be made from noisy uncertain observations.
  • Give a clear description of EM for the purpose of clustering.
  • Define the Kernel Trick and how it can be easily incorporated into the SVM classifier framework.

Analysis

  • Discuss the limitations associated with machine learning techniques with respect to generalization and complexity of learning.
  • Define how one can describe the qualitative quality of a classifier given test data.
  • Propose plausible solutions to classification tasks based on images, from the feature extraction process to the classification method.
  • Define VC-dimension and how it is used to describe the representational power of a classifier and how it is related to generalization.

Image representations

  • Describe different image feature representations which are commonly used in computer vision and why they are popular.

Implementation

To implement, using Matlab, common machine learning classification and dimensionality reduction algorithms and apply them to images in particular the
  • Boosting algorithm, perceptron learning, PCA, link into SVM software packages, LDA, nearest neighbour classifier,..
Copyright © Sidansvarig: Josephine Sullivan <sullivan@nada.kth.se>
Uppdaterad 2012-02-17