Nachricht

Am Mittwoch, den 08. Mai 2013, hält Herr Zubair Kamran um 16:30 Uhr im Raum P 1.4.17 einen Vortrag mit dem Titel:

Finding appropriate segmentation for visual attention modeling

Zusammenfassung:

Attention is the cognitive process of selectively concentrating on limited aspects of the environment while ignoring the rest. Attention, both in animals and visual systems, can be referred to as the allocation of processing resources. For humans as well as for robots, limited resources require a selection mechanism which prioritizes the sensory input from “very important” to “not useful right now”. Based on concepts of the human visual system, computational visual attention systems aim to detect regions of interest in images.

The talk includes comparison of different attention models in light of signal detection metrics, including “Receiver Operating Characteristics”, “Sensitivity & Specificity”, “Precision & Recall” and “The F measure”. The second part of the talk includes detailed discussion on evaluation of different popular segmentation methodologies and their relevance for attention oriented segmentation. The segmentation approach Super Pixel is discussed in detail. The basic idea in this approach is to deal with significantly lesser number of pixels (super pixels) rather than dealing with millions of pixels for segmentation at the same time. The image is firstly over segmented into Super pixels based on a variety of features derived from the classical Gestalt cues, including contour, texture, brightness and good continuation. Many existing algorithms in computer vision use the pixel-grid as the underlying representation; however it is not a natural representation of visual scenes. It is rather an "artifact" of a digital imaging process. It would be more natural, and presumably more efficient, to work with perceptually meaningful entities obtained from a low-level grouping process, by firstly over segmenting an image. This sub grouping is achieved by super pixel.