Loading...

Recovery of Binary Images from Samples of the PSF-Blurred Images

Zamani, Hojatollah | 2021

876 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 53879 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Amini, Arash; Marvasti, Farrokh; Fakharzadeh, Mohammad
  7. Abstract:
  8. One of the main challenges in image processing is the reconstruction of images from their samples in the discrete domain to obtain high quality images in the continuous domain. Visual signals in the continuous domain are usually represented by discrete or digital images. This reconstruction process may not be possible in general, and we need to set some specific constraints for this purpose, such as delimiting the signal bandwidth. In this dissertation, we consider the reconstruction of binary images (black and white). These images have infinite bandwidth due to the existence of edges, and thus, the conditions of the Nyquist-Shannon sampling theorem do not hold. Novel methods have been proposed for sampling and reconstructing infinite-bandwidth signals. Considering the system and reconstruction method, these signals are classified in two categories: (i) compressed sensing, and (ii) signals with a finite rate of innovation. In this research, the considered binary images represent shapes that can be described by a finite many of parameters. For instance, the image of blood cells consists of some ellipses that each can be described by a few parameters. In physical structures, however, the image samples are derived after the impact of the point spread function (PSF). We consider the effect of PSF in this research as well. Note that the PSF structure depends of the imaging system under consideration. In this dissertation, we model the effect of PSF and intend to eliminate it. Considering the system structure, when the PSF is limited support, these images are classified as signals with a finite rate of innovation. However, for systems with unlimited support PSF, such as millimeter-wave imaging system, we need to use approximations in the modelling, and thus, exact reconstruction is not possible. In this dissertation, we develop novel methods for reconstructing images with ellipsoid shapes based on the theory of finite rate of innovation for finite PSF imaging systems. Furthermore, we investigate the millimeter-wave imaging system which has a different and complex PSF structure. For image recovery in this system, we benefit from the signal sparsity and propose methods to improve the overall performance. Due to the physical structure of the system, we investigated the phase deviation and proposed novel techniques to counter its effects. Finally, to improve the reconstruction process, we propose a deep learning (DL)-based method which outperforms its previous counterparts
  9. Keywords:
  10. Sampling ; Deep Learning ; Point Spread Function ; Imaging System ; Innovation Finite Rate ; Binary Images ; Band-Limited Signal

 Digital Object List

 Bookmark

...see more