AG 1, AG 2, AG 3, AG 4, AG 5, SWS, RG1, MMCI   Info

Public Audience

English

120 Minutes

Saarbrücken

Efficient and accurate interactive image or video segmentation have significant importance in many multimedia applications. For mobile applications, efficiency is more crucial. Moreover, due to small screens of the mobile devices, error tolerance is also a crucial factor. Hence, classical interactive image and video segmentation methods should be revisited for mobile applications. For the interactive image segmentation problem, a novel interaction methodology -coloring- is proposed and will be explained briefly. Most of the attention will be given to the computational efficiency perspective. A novel spatially dynamic graph-cut method is proposed for efficiency increase. Theoretical and experimental analysis of the proposed method will also be discussed. On the other hand, interactive video segmentation problem is formulated as Markov Random Field (MRF) energy propagation along the temporal domain. MRF energy propagation problem is mainly addressed from the perspective of efficiency. For this purpose, a novel and efficient method is proposed to propagate MRF energies throughout the frames via bilateral filters without using any global texture, color or shape model or optical flow computation. Avoiding global models like GMM and expensive computation of optical flow increase the computational efficiency significantly. Furthermore; in order to exploit spatio-temporal information, a bilateral filter with spatio-temporal geodesic distance weights is also utilized. Recently presented bi- exponential filter is utilized to increase the efficiency of the geodesic bilateral filters, whereas a novel technique is also developed to solve dynamic graph-cuts for varying, non-lattice graphs in general linear filtering scenario. Proposed method and computational complexity analysis will be discussed in detail. Computational efficiency and segmentation quality of the proposed method is also experimented. Indeed, for some challenging examples, time efficiency and segmentation quality will be compared with the existing algorithms in the literature.

IMPRS Office Team

0681 93251800

--email hidden

passcode not visible

logged in users only