Keywords
Visual attention, object recognition, biased competition, top-down, visual search
Abstract
Visual attention models can explain a rich set of physiological data (Reynolds & Heeger, 2009, Neuron), but can rarely link these findings to real-world tasks. Here, we would like to narrow this gap with a novel, physiologically grounded model of visual attention by demonstrating its objects recognition abilities in noisy scenes.
To base the model on physiological data, we used a recently developed microcircuit model of visual attention (Beuth & Hamker, in revision, Vision Res) which explains a large set of attention experiments, e.g. biased competition, modulation of contrast response functions, tuning curves, and surround suppression. Objects are represented by object-view specific neurons, learned via a trace learning approach (Antonelli et al., 2014, IEEE TAMD). A visual cortex model combines the microcircuit with neuroanatomical properties like top-down attentional processing, hierarchical-increasing receptive field sizes, and synaptic transmission delays. The visual cortex model is complemented by a model of the frontal eye field (Zirnsak et al., 2011, Eur J Neurosci).
We evaluated the model on a realistic object recognition task in which a given target has to be localized in a scene (guided visual search task), using 100 different target objects, 1000 scenes, and two backgrounds. The model achieves an accuracy of 92% at black, and of 71% at white-noise backgrounds. We found that two of the underlying, neuronal attention mechanisms are prominently relevant for guided visual search: amplification of neurons preferring the target; and suppression of neurons encoding distractors or background noise.
Start Date
13-5-2015 10:40 AM
End Date
13-5-2015 11:05 AM
Session Number
01
Session Title
Motion, Attention, and Eye Movements
Included in
Artificial Intelligence and Robotics Commons, Cognition and Perception Commons, Computational Neuroscience Commons, Systems Neuroscience Commons
Object Recognition and Visual Search with a Physiologically Grounded Model of Visual Attention
Visual attention models can explain a rich set of physiological data (Reynolds & Heeger, 2009, Neuron), but can rarely link these findings to real-world tasks. Here, we would like to narrow this gap with a novel, physiologically grounded model of visual attention by demonstrating its objects recognition abilities in noisy scenes.
To base the model on physiological data, we used a recently developed microcircuit model of visual attention (Beuth & Hamker, in revision, Vision Res) which explains a large set of attention experiments, e.g. biased competition, modulation of contrast response functions, tuning curves, and surround suppression. Objects are represented by object-view specific neurons, learned via a trace learning approach (Antonelli et al., 2014, IEEE TAMD). A visual cortex model combines the microcircuit with neuroanatomical properties like top-down attentional processing, hierarchical-increasing receptive field sizes, and synaptic transmission delays. The visual cortex model is complemented by a model of the frontal eye field (Zirnsak et al., 2011, Eur J Neurosci).
We evaluated the model on a realistic object recognition task in which a given target has to be localized in a scene (guided visual search task), using 100 different target objects, 1000 scenes, and two backgrounds. The model achieves an accuracy of 92% at black, and of 71% at white-noise backgrounds. We found that two of the underlying, neuronal attention mechanisms are prominently relevant for guided visual search: amplification of neurons preferring the target; and suppression of neurons encoding distractors or background noise.