|M.Sc Student||Kenig Tal|
|Subject||Blind Deconvolution in Wide-Field Fluorescence Microscopy|
|Department||Department of Electrical Engineering||Supervisor||Professor Emeritus Arie Feuer|
|Full Thesis text|
In each case where an image is taken, there is an underlying object we wish to observe. However, the acquired image is never an ideal representation of this object of interest, as it is often corrupted by blur, noise and other degradations which occur during the acquisition process. Some of the most classical tasks in image processing are deblurring and denoising, which aim at inverting the effect of blur and noise, respectively. One specific field where the application of such methods is especially beneficial is wide-field fluorescence microscopy, as the images acquired by these devices are heavily blurred, in addition to the statistical noise they contain. Also, the reconstruction of those images is a particularly challenging task, since the degradation induced by the microscope is affected by the imaged sample itself, and is therefore practically very difficult to determine a-priori.
The objective of this research is to recover three-dimensional images acquired by wide-field fluorescence microscopes. The recovery procedure includes deblurring the image while suppressing the statistical noise it may contain, without having explicit knowledge of the degradation model. The approach developed in this thesis combines classical, extensively studied deblurring methodologies with state of the art techniques from the field of machine learning. The contributions of this thesis are summarized as follows:
1. An algorithm for the calibration of detector sensitivity effects is proposed. This algorithm is shown to significantly reduce image artifacts originating from varying sensitivity to light of different detector elements. Unlike conventional calibration methods, this algorithm does not require the acquisition of any calibration images.
2. A method for noise reduction embedded into the deblurring process, which was previously proposed in the context of non-blind astronomical image deblurring is generalized and used for blind image deblurring in microscopy.
3. A novel prior term is proposed for the regularization of blind deblurring methods. The proposed methodology is inspired by recent research in the computer vision community and introduces machine learning methods into the blind deblurring process. The proposed technique has sound mathematical foundations, is generic to many inverse problems and is shown to yield excellent results for wide-field fluorescence microscopy imagery.