ICT4SS - ICT FOR SMART SOCIETIES
Deep neural networks for reconstruction of compressed sensing data
keywords DEEP LEARNING, VIDEO ANALYSIS
Reference persons ENRICO MAGLI
Thesis type RESEARCH
Description Compressed sensing (CS) has recently emerged as a technique providing single and compressed data representations that may be employed to design novel computational imaging systems. However, image reconstruction algorithms for CS have always suffered from high complexity and relatively low reconstruction quality. Since these algorithms are nonlinear, deep neural networks seems to be an ideal candidate to learn a reconstruction algorithm from training examples. The thesis has the objective to train reconstruction algorithms based on deep learning for the CS problem, and to compare their performance with existing algorithms. The activity will consider training of the reconstruction algorithm both in conjunction with the learning of optimized sensing parameters, and stand-alone reconstruction for a family of sensing matrices.
See also the following tutorial paper on CS: http://ieeexplore.ieee.org/document/4472240/
Required skills Candidate students should have a background in ICT or mathematics. Knowledge of neural networks (including TensorFlow environment and Python programming) is not a prerequisite, although it would help in the initial stage of the activity.
Deadline 20/11/2018 PROPONI LA TUA CANDIDATURA