Figure 8
From: Mice use robust and common strategies to discriminate natural scenes
Analyses using upper portions of images correlate best with behavior. (a) Analysis workflow for analyzing subregions (60 × 60 pixels) of images. SSIM was computed for the cropped subregions of a target and distractor image pair. The correct rate was plotted versus these subregion SSIM indices, a psychometric curve was fit, and the RMSE of the fit to the data was computed. These RMSE values were color coded and plotted at the center point of the subregion. This process was iterated for all XY positions. (b) SSIM-based curve fitting indicated that subregions in the upper portion of the images yielded SSIM indices that correlated well with behavior and thus resulted in low RMSE fits. (c) Similar analysis was performed for the Gabor filter-based analysis. Again, the upper portion of the image yielded the best fits, and the 0° and 30° Gabor filters performed better than other angles.
