MATLAB code for predicting the ON and OFF pathway responses to visual input
Main function is onoff.m:
Predict the ON and OFF retinal ganglion cell responses to an image. Separate predictions are generated for foveal and peripheral cells in the P and M pathways
Example call: [on, off, rgcs] = onoff('file','checker.png','arcminperpixel',1)
INPUT: If called with no arguments, you will be prompted to select an image file and the image viewing resolution will be assumed to be 1 arcminute per pixel.
If the 'file' argument is supplied, it will instead try to load that
file from the current working directory
If the 'arcminperpixel' argument is supplied, it will use that value
instead of the default for the visual arcminutes subtended by a single
image pixel
Note that image element values are assumed to be linear with world light intensity. This means that loading an image file directly from a camera will likely produce spurious predictions due to nonlinearities introduced in the image encoding. Loading a bitmap that has been designed for display on a linearized monitor via a LUT, however, should produce meaningful predictions. File types can be anything that MATLAB's imread can handle, as well as .mat files containing a single variable that is a matrix containing the image. Images of matrices containing 3 channels will be assumed to be RGB and a luminance conversion will be applied.
For convenience, the auxilary load_images function can handle images that come from the Van Hateren Natural Image Dataset (http://bethgelab.org/datasets/vanhateren/)
or from the McGill Calibrated Colour Image Database (http://tabby.vision.mcgill.ca)
...so long as the file ending is ".iml" for Van Hateren or the file/path contains the phrase "mcg" for McGill (and the McGill rgb2linear function is also in the path)
Otherwise, various example images are provided in example_images directory for testing
OUTPUT:
on, off: structures that contain matrices of the ON and OFF
response magnitudes at each input image pixel. Matrixes are cropped
relative to the input image to remove boundary artifacts from
convolution
rgcs: structure with info about the rgc model used to produce each
matrix in the on and off structures
Also produces plots illustrating the RGC models used, images of the RGC response magnitudes for P and M pathways, and a bar plot of the summed responses for ON and OFF cells in each population (4 populations total)
Emily Cooper, 2015
Please cite accompanying paper: Cooper, E.A. & Norcia, A.M. Predicting Cortical Dark/bright Asymmetries from Natural Image Statistics and Early Visual Transforms