Speaker / Predavač : Mr Tibor Lukic

Title/Naslov: Gradient based optimization methods in image processing

Abstract / Abstrakt:

Energy minimization models have a wide area of application in image processing. Applicability and efficiency of such models mostly depend on the properties and fitness of the underlying optimization method. Gradient based iterative methods are suitable choice in many problems: 1) due to low memory and computational requirements, they are capable for solving large-scale problems as it is the case in many image processing models; 2) do not require computation of second or higher order derivatives; 3) running time and quality of the solutions, in most cases, are quite acceptable. Within the frame of this presentation we will especially focus on the Barzilai and Borwein (spectral) step-length rule and its applications in the field. Finally, there will be also presented two newly developed methods based on the spectral gradient approach for image denoising and discrete tomography.