Purchase this article with an account.
Heiko Schütt, Lars Rothkegel, Hans Trukenbrod, Ralf Engbert, Felix Wichmann; Testing an Early Vision Model on Natural Image Stimuli. Journal of Vision 2017;17(10):783. doi: 10.1167/17.10.783.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Early visual processing has been studied extensively over the last decades. From these studies a relatively standard model emerged of the first steps in visual processing. However, most implementations of the standard model cannot take arbitrary images as input, but only the typical grating stimuli used in many of the early vision experiments. Previously we presented an image based early vision model implementing our knowledge about early visual processing including oriented spatial frequency channels, divisive normalization and optimal decoding (Schütt & Wichmann, VSS, 2016). The model explains the classical psychophysical data reasonably well, matching the performance of the non-image based models for contrast detection, contrast discrimination and oblique masking data. Here we report tests of our model using natural images, exploiting the benefits of image based models of visual processing. First, we assessed the performance of the model against human observer thresholds for detecting noise Gabors masked by patches of natural scenes (Alam et. al., JoV, 2014). Our model predicts the thresholds for this masking experiment well, although it slightly overestimates the sensitivity of observers. Second, we investigated the channel activities for natural scene patches fixated by observers in a free viewing eye movement experiment. Before normalization channel activities follow typically observed biases of natural scenes, including the decline in energy over spatial frequency and the stronger activity along the cardinal axes. After divisive inhibition, the distribution activity is no longer skewed towards low spatial frequencies, while the preference for cardinal axes is preserved. Finally, we observe that the channels are extremely sparsely activated: each natural image patch activates few channels and each channel is activated by few stimuli. Thus our model is able to generalize from simple grating stimuli to natural image stimuli, and it reproduces normative desiderata stemming from the efficient coding hypothesis and natural image statistics.
Meeting abstract presented at VSS 2017
This PDF is available to Subscribers Only