Thema
Features for Visual Saliency
Typ Master, Forschungspraxis, Bachelor, Ing.prax.
Betreuer Mikhail Startsev
Tel.: +49 (0)89 289-28550
E-Mail: mikhail.startsev@tum.de
Sachgebiet Human and Computer Vision
Beschreibung The term “visual saliency” [1] refers to modeling the mechanism of human attention, which is essential in combination with our visual system, since our eyes can clearly see only a small part of the visual field: humans have to move their eyes in such a way that would allow the projection of the desired stimulus to fall onto the fovea of the eye [2]. By carefully (though very intuitively) choosing the targets for our eyes, we are able to stay aware of the world around us, even while performing complex tasks (ex. driving).

There are two major theories describing why do we look where we look, and therefore how this process can be modeled (i.e. approaches to visual saliency modeling). On the one hand, our sight can be directed by so-called “bottom-up” features of the scene, such as motion or color change in our periphery. This can drive our eyes to examine the observed disturbance, which from the evolutionary point of view it can be regarded as quickly reacting to something potentially dangerous.

On the other hand, prior knowledge of the scene structure allows us to keep track of important objects (such as people or cars, for instance) without having to look at them constantly. To then switch our point of regard in order to look at the object again, we do not need any bottom-up features, we already know where to look. This is called top-down processing, and could be related to social interaction.

When attempting to model the attention mechanism on a computer, we cannot directly copy the visual signal processing chain of the human brain, not least because it is not entirely understood so far. We therefore have to assume some sort of a computational model behind the extraction of features that could be relevant for perception. Usually one of the top-down or bottom-up models is assumed, or a mix thereof. Switching the feature model can influence the performance of an attention predictor dramatically.

A candidate would implement or adapt various feature extraction mechanisms within an existing saliency prediction framework. Systematically evaluating and comparing them would potentially further our understanding of stimulus features that catch our eye.

[1] en.wikipedia.org/wiki/Salience_(neuroscience)
[2] en.wikipedia.org/wiki/Visual_search
Voraussetzung Programming skills are desirable, experience with computer vision is a plus.
Bewerbung If you are interested in this topic, we welcome the applications via the email address above. Please set the email subject to “<Type of application> application for topic 'XYZ'”, ex. “Master’s thesis application for topic 'XYZ'”, while clearly specifying why are you interested in the topic in the text of the message. Also make sure to attach your most recent CV (if you have one) and grade report.