Edinburgh Research Explorer

Saliency detection without biasing stimulus estimation

Research output: Contribution to conferencePosterpeer-review

Original languageEnglish
Publication statusPublished - 2014
EventVisual Image Interpretation in Humans and Machines - Macdonald Alveston Manor Hotel, Stratford-Upon-Avon, United Kingdom
Duration: 24 Sep 201425 Sep 2014


WorkshopVisual Image Interpretation in Humans and Machines
CountryUnited Kingdom


In surround modulation in V1 the response of a neuron to a centre grating is modulated by the presence of a surround grating. Generally, the modulation is thought to be suppressive and surround orientation dependent, with the strongest effect when the surround is aligned with the preferred orientation of a neuron. This simple model allows for a wide range of perceptual effects, including saliency detection.
I will first show that while this model of centre-surround interactions does allow for saliency detection, it inherently biases the estimation of the stimulus, even when using a decoder with full knowledge of the effects of the surround modulation.
Next, I will draw inspiration from a range of experiments which have shown that a majority of neurons in V1 experience most suppression when the surround is aligned with the centre, rather than with the preferred orientation. This simple change prevents the bias from arising, while also potentially improving saliency detection.


Visual Image Interpretation in Humans and Machines


Stratford-Upon-Avon, United Kingdom

Event: Workshop

ID: 18017839