The document discusses the concept of representations in computer vision, addressing various questions around their necessity, optimality, and the relationship between representation and task performance. It also references multiple disciplines such as statistics, information theory, biology, computer architecture, cognitive science, and philosophy related to the topic. Additionally, related papers in these fields are cited to provide further context on the subject.

1. Alessandro Ortis
Università degli Studi di Catania
Dipartimento di Matematica e Informatica
Image Processing Lab - iplab.dmi.unict.it
Representations in Computer Vision
(intro)

2. CV input and output
CV
Images
Videos
…
Detection
Recognition
Categorization
Navigation
Manipulation
Tracking
…
Generally the output are actions and/or decisions.
A. Ortis - Representations for CV

3. Representation concept regards
several disciplines
Statistics
Information
Theory
Biology
Computer
Architecture
Cognitive
Science
Philosophy
A. Ortis - Representations for CV

4. • Are they necessary?
• Can we do without?
• What representation should we compute and mantain
in memory for a computer vision algorithm?
• How do we measure how ‘’useful’’ it is?
• Does it depend on the task?
• What if the representation is the task itself?
• Is there an optimal representation?
• If so, can it be computed? If not, can it be
approximated?
• What do animals store in memory? How?
• …
About representation
A. Ortis - Representations for CV

5. Related papers
Statistics:
- Sufficient Statistics, G. Lebanon
Information Theory:
- Data Processing Inequality, S. D. Servetto
- Information Theory, Cover and Thomas
- The Information Bottleneck Method, N. Tishby et al.
Biology:
- The Biology of Memory, E. R. Kandel
Computer Architecture:
- The Impact of Memory and Architecture on Computer
Performance, N. H. F. Beebe
Cognitive Science
- What is a Knowledge Representation, R. Davis et al.
Philosophy:
- Mental Representation, SEP
A. Ortis - Representations for CV