ECE 495: Cameras, Images, and Statistical Inverse Problems

Professor Stanley H. Chan, Purdue University, Spring 2022

Announcement

  • 10/01/2021 Course website launched.

Course Information

Lecture: TBD
Room: TBD

Instructor: Professor Stanley H. Chan
Room: MSEE 338
Email: stanleychan AT purdue DOT edu
Office Hour: TBD

Lecture Note

Part 1: From Sensors to Images to Statistics

  • Lecture 01: Understanding your cameras

    • Lecture 01-1: Cameras in the 21st century Lecture Note (PDF, 6MB)

    • Lecture 01-2: CCD, CMOS, and single-photon image sensors

    • Lecture 01-3: Noise, SNR, and dynamic range

  • Lecture 02: Gaussian random variables

    • Lecture 02-1: Single-variate Gaussian random variables

    • Lecture 02-2: Multi-variate Gaussian random variables

    • Lecture 02-3: Why is Gaussian everywhere?

  • Lecutre 03: Poisson random variables

    • Lecture 03-1: Basics of Poisson random variables

    • Lecture 03-2: The physics of photon arrivals

    • Lecture 03-3: Poisson statistics for single-photon image sensors

Part 2: The Three Estimation Tools: ML, MAP, MMSE

  • Lecture 04: Optimization tools

    • Lecutre 04-1: The least squares problem

    • Lecture 04-2: Objective function, constraint, and gradients

  • Lecture 05: Maximum-likelihood estimation (ML)

    • Lecture 05-1: Maximum-likelihood estimation: principles

    • Lecture 05-2: Applications of ML estimation in image recovery

  • Lecture 06: Maximum-a-posteriori estimation (MAP)

    • Lecture 06-1: Maximum-a-posteriori estimation: principles

    • Lecture 06-2: Priors for MAP estimation

  • Lecture 07: Minimum mean square estimation (MMSE)

    • Lecture 07-1: Understanding the mean squared error

    • Lecture 07-2: Deriving the MMSE solution

Part 3: Noise Removal Techniques

  • Lecture 08: Iterative algorithms

    • Lecutre 08-1: Total variation and dictionary

    • Lecture 08-2: A gentle introduction to ADMM

    • Lecture 08-3: Plug-and-play priors

    • Lecture 08-4: Handling the Poisson likelihood

  • Lecture 09: Patch reoccurence

    • Lecture 09-1: Bilateral filter and recursive filters

    • Lecture 09-2: Non-local means and 3D block matching (BM3D)

    • Lecture 09-3: Symmetric smoothing filters and the Sinkhorn symmetrization

  • Lecture 10: Statistical techniques for Poisson noise removals

    • Lecture 10-1: Variance stabilizing transform

    • Lecture 10-2: Shrinkage and Stein's unbiased risk estimator

  • Lecture 11 Limit of denoising

    • Lecture 11-1: Is denoising dead?

    • Lecture 11-2: MMSE lower and upper bound

Part 4: Advanced Topics

  • Lecture 12: Learning-based noise removal techniques

    • Lecture 12-1: Network designs: UNet, REDNet, DnCNN

    • Lecture 12-2: Unrolling the ADMM algorithm

    • Lecture 12-3: Deep Wiener filter

    • Lecture 12-4: Knowledge distillation

    • Lecture 12-5: One size fits all

Homework

TBD

Quiz

TBD