Outline

1
Outline

• Announcements
• Theoretical approaches to computer vision
• Classical theories of vision
• Visual perception as information processing

2
Announcements

• Class web page http//www.cs.fsu.edu/liux/courses
  /research/index.html
• Lecture notes and papers can be obtained from
  http//www.cs.fsu.edu/liux/courses/research/calen
  dar.html
• Possible programming assignments
• You can certainly do your own project
• Implement a method from the literature
• Implement your own novel ideas on a problem as
  you are going to discuss in this class
• You can also do a project on top of my programs
• I will make my programs available to you and you
  can make changes

3
Theoretical Approaches to Vision

• Classical theories of vision
• Visual perception as information processing

4
Visual Perception as an Inverse Problem

• Retinal images are generated by the light
  reflected from the 3-D world
• The image formation is determined by the laws of
  optics
• The area of image rendering is called computer
  graphics
• Vision as an inverse problem
• Get from optical images of scenes back to
  knowledge of the objects that gave rise to them

5
Problems with Inversion

• Image formation is a well-defined function
• Each point in the environment maps into a unique
  point in the image
• Inverse process is not well-defined
• Vision goes from 2-D to 3-D
• Each point in the image could map into an
  infinite number of points in the environment
• It is underconstrained

6
Vision as a Heuristic Process

• Visual system makes a lot of assumptions about
  the nature of the environment and conditions
  under which it is viewed
• These assumptions constrain the inverse problem
  enough to make it solvable most of the time
• The resulting solution will be veridical if the
  assumptions are true
• Vision is a heuristic process in which inferences
  are made about the most likely environmental
  condition that could have produced a given image

7
Computer Vision

• The study of how computers can be programmed to
  extract useful information about the environment
  from the optical images
• Computer metaphor
• Minds are like programs that run on machines
  called brains
• Minds are the software of biological
  computation and the brains are the hardware
• Strong AI vs. weak AI
• Artificial intelligence

8
Three Levels of Information Processing

• Marr proposed this meta-theory, a theory about
  theories on vision
• Three levels of any information processing
  systems
• Computational level
• Algorithmic level
• Implementation level

9
Computational Level

• Computational level
• The most abstract description level
• Informational constraints available for mapping
  input information to output information
• It specifies what computation needs to be
  performed and on what information it should be
  based

10
Algorithmic Level

• Algorithmic level specifies how a computation is
  executed in terms of information processing
  operations
• There are in principle many different algorithms
  to accomplish the computational-level mapping of
  input to output
• Two fundamental components
• One must decide a representation for input and
  output information
• One must construct a set of processes

11
Implementation Level

• This level specifies how an algorithm actually is
  embodied as a physical process within a physical
  system
• The same algorithm can be implemented using many
  physically different devices
• Devices include brains as biological devices as
  well as different kinds of computers

12
Representations

• A representation refers to a state of the visual
  system that stands for an environmental property,
  object, or event
• Represented world
• External world outside the information processing
  system
• Representing world
• The internal representation within the
  information processing system

13
Processes

• Processes are the active components in an
  information processing system that transform or
  operate on information by changing the
  representation into the next
• Dynamic aspect of the system that causes
  informational transformations to occur
• Implicit vs. explicit
• One of the most important aspects of processes is
  to make information that was implicit in the
  input representation explicit in the output

14
Processing as Inference

• Helmholtz proposed that vision arrives at the
  interpretation that is the most likely state of
  affairs in the external world that could have
  caused the retinal stimulation
• This proposal is called the likelihood principle

15
Perception as Bayesian Inference

• Images I are observations
• Scene properties S are not known
• p(S) specifies the prior knowledge about the
  scene
• The knowledge you have without looking at the
  image
• Bayes rule

16
Top-down vs. Bottom-up Processes

• Bottom-up processing
• Data driven processing
• Take a lower-level representation as input and
  create or modify a higher-level representation
• Top-down processing
• Expectation-driven processing
• Processes that take a higher-level representation
  as input and produce or modify a lower-level
  representation

17
Four Stages of Visual Processing

• Retinal image
• Image-based stage
• Surface-based stage
• Object-based stage
• Category-based stage