Computational imaging and spectroscopy

• Apply the concepts of Harmonic analysis and Compressed Sensing to imaging
• Design computer vision and computational imaging systems and algorithms
• Analyze and process signals from cameras and optical sensors
• Apply the concepts of inverse problems and linear optimization to imaging
• Recover, analyze and process spectrally resolved data from optical sensors or images
• Recover and analyze scene physics from optical sensors or digital images
• Apply Deep Learning and Machine Learning frameworks to solve computational imaging problems
• Process and analyze hyperspectral images with Machine Learning methods

The course presents a wide range of computational methods, classic and novel, used for analysis of digital images. It covers for instance recovering data or physical information from sparse or dense measurements, reconstructing and restoring images, performing automated image analysis such as facial recognition or AI aided medical diagnosis, and designing advanced imaging systems with applications in e.g. spectroscopy.

The course consists of five parts:
1. Introduction to digital imaging
2. Sparse representations and image restoration
3. Scene analysis and spectral imaging
4. Introduction to deep learning for computational imaging
5. Computational spectroscopy and hyperspectral image analysis

The course focuses first on traditional digital imaging: after presenting the fundamentals of colorimetry, the functional principles of sensing devices used for image acquisition and classic image processing tools, such as kernel filtering, are covered. Then advanced harmonic analysis methods will be introduced to exploit the sparse nature of digital images, with applications in image restoration, time and frequency analysis, and compression. This will be followed by an overview of state of the art image restoration methods, including sparse optimization methods, numerical methods for image inpainting and classic filter based approaches for denoising. Spectral imaging and scene physics analysis will then be presented with the motivation to retrieve scene physics and spectral information from digital imaging. Afterwards, deep learning methods will be introduced as an overview of concepts and architectures for application to image analysis. Finally, we will address the applications of deep learning and sparse optimization to imaging spectroscopy and hyperspectral imaging analysis, with a focus in medical imaging.

Matlab or Python programming

Lectures, exercises/problem solving, project