Figure 6
From: Mice use robust and common strategies to discriminate natural scenes
A simple Gabor filter model predicts mouse performance. (a) For a simple model the responses of V1 neurons, Gabor filters with various orientations and wavelengths were used to convolve the stimulus images. (b) The orientation specific similarity (OSS) between the target image and a distractor image was calculated by convolving a single Gabor filter with each image and then comparing the two results pixel-wise. The correct rate from the behavior data was then compared with these OSS values for each distractor image (compared to the target). (c) Example fits for the average correct rate across all mice against OSS values show that a filter with θ = 30°, λ = 10 (left) provided a good fit (RMSE = 0.055); and a filter with θ = 90°, λ = 10 (right) provided a poor fit (RMSE = 0.13). (d) Gabor filter patches with relatively short wavelengths provided lower RMSE fits to the behavior data (mean and S.E.M. are plotted; based on an average OSS over all orientations), and this suggests that relatively high spatial frequency information in the images is used by the mice for this discrimination. (e) The orientation of the Gabor patch used for calculating OSS influenced the resulting RMSE (p = 9 × 10−11, ANOVA; patches had wavelength = 10; mean and S.E.M. are plotted; blue curve is a sinusoidal fit). (f) An examination of the interaction between wavelength and orientation revealed that RMSE of the OSS-based fits depended more on orientation than wavelength of the Gabor filters.
