Retina transformation for images

Transforms a 24x24 pixel patch (576-dimensional) to 132-dimensional representation by averaging pixels far away from the center (i.e. fovea). There is code for both directions of the transform.