Image statistics

Take home messages:

- Ask not what a neuron is doing but why it is doing it.
- The primary advantage of theory is to abstract away from the details.
- This can also make it difficult to compare directly to data.
- Nature does not waste resources: every bit counts as does every spike.
- Nature's resources are also limited and this affects representations.
- Images statistics are important for both coding and extracting structure
- The problem visual coding solves:
--- Encode the maximum amount of (relevant) information about the (underlying) image in a way that is robust to noise and makes efficient use of neural resources

Presentation:

Part 1
Part 2
Part 3
Part 4

Readings:

Doi E., Balcan D.C., Lewicki M.S.
A Theoretical Analysis of Robust Coding over Noisy Overcomplete Channels
ANIPS. [pdf]

Karklin Y. and Lewicki M.S.
Learning higher-order structures in natural images
Network: Comput Neural Syst. 2003. [pdf]

Karklin Y. and Lewicki M.S.
A Hierarchical Bayesian Model for Learning Nonlinear Statistical Regularities in Nonstationary Natural Signals
Neural Computation. 2005. [pdf]

Tutorials:

imageTutorial
pyramidTutorial