Purchase this article with an account.
H.Steven Scholte, Sennay Ghebreab; Improving computational models of early visual cortex using single image ERP data. Journal of Vision 2014;14(10):886. doi: https://doi.org/10.1167/14.10.886.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
In the past we have generated models of the early visual system (Scholte et al., 2009, Ghebreab et al, 2011,) that are capable of generating summary statistics of images related to contrast energy and spatial coherency. These parameters group visual images into a structures space that goes from, on one dimension, have a high likelihood of having features like man-made figure-ground objects to having a high likelihood of having features like natural and fragmented. The other dimension organises the images in terms of energy. These summary statistics can explain between single-image ERP's well, particularly around 120 ms, but also extending later in time. Relevantly, the summary statistics derived from the model explained more variance in the ERP than Fourier, or third and fourth order statistics (Groem et al., 2012, 2013). Here we extend this approach by formally using two separate datasets of ERP responses towards natural images. One for fine-tuning parameters via cross-validation and one for final validation. Using this we now not only model on/off cells but also simple cell responses. Adding the parameters obtained from simple cell responses improves the explained variance of our model thought-out the range of modeled information, but in particularly, and as expected after the peak explained variance of the on/off cells, 140 ms. We will also present data in which we use these parameters to see to what degree we can 'explain' away differences between ERP responses from, for instance animal vs non-animal, images.
Meeting abstract presented at VSS 2014
This PDF is available to Subscribers Only