Nachricht

On Tuesday, October 14th. 2008, Mr. Zaheer Aziz will give a presentation at 4:00 pm in P 6.2.03 titled:

Behavior Adaptive and Real-Time Model of Integrated Bottom-Up and Top-Down Visual Attention

Summary:

Visual attention is an important component of natural vision that helps it to optimize the amount of data that reaches the brain for detailed processing. Attention mechanism applies a filtration process in the visual input that selects only relevant and important portions from the viewed scene for high level analysis. Computational models of attention attempt to perform this filtration for the machine vision systems. The work presented in this dissertation proposes a region-based approach for modeling visual attention as an alternative to the other existing paradigms. The proposed model integrates bottom-up and top-down pathways of attention into a single architecture and makes combined use of these pathways under different visual behaviors. This was not done by any computational model of attention before. In order to obtain real-time results on mobile vision systems new faster algorithms are developed for feature extraction and saliency computation. These algorithms compute contrast in five feature channels in context of local neighborhood as well as the global context of the whole view. The innovation in terms of top-down attention is the creation of fine-grain saliency maps apart from the bottom-up maps. The proposed model produced valid results and has shown good performance in comparison to other available attention models hence this research has openned new directions for investigations in this field that can lead to the ultimate target of biologically plausible machine vision.