The purpose of this course is a self-contained introduction to modern techniques in image processing. We will understand concepts involved in acquiring, sampling, representing, compressing, and processing of multi-dimensional signals (images and videos). We will discuss the notion of inverse problems and ways to solve them – from traditional prior-based methods to modern learning-based methods. Finally, apart from the standard image acquisition model, we will also look at more exotic computational and medical imaging schemes and understand the challenges and applications involved there within.
At the end of the course, the student will:
- Get comfortable with the standard notions involved in acquiring, sampling, representing, compressing, and processing of images.
- Understand the evolution of ideas in image processing
- Know how to implement these state-of-the-art image processing algorithms in Python.
- Perform a small research project using the studied notions and techniques.
Evaluation: 60% Homework assignments, 40% final project.
Language: The course will be taught in English.
||Lecture 2||Signals & systems, Fourier||Convolutions|
||Lecture 3 (until Poisson summation)||Adjoint, Dirac’s delta, Poisson summation||Fourier transform|
||Lecture 3 (the rest)||Sampling & interpolation||Computerized tomography|
||Lecture 4||Discrete domain signals & systems||Sampling & iterpolation||HW1|
||Lecture 5a||Introduction to random signals||Inverse filtering|
||Lecture 5b||Inverse problems & statistical estimation||ML vs. MAP|
||Lecture 6||Patch-based priors||Bilateral filters, NLM||HW2|
||Lecture 7||Sparsity-based priors||PatchMatch|
||Lecture 8||Structured-based priors||Sparse coding, BM3D|
||Lecture 9||Learning image priors||Dictionary learning|
||Lecture 10||Sampling & sensing||Deep learning (part 1)|
||Lecture 11||Computational imaging||Deep learning (part 2)|