The paper describes your initial image processing approach as:
- we down-sampled the images to a fixed resolution of 256x256.
Given a rectangular image, we first rescaled the image such that the
shorter side was of length 256, and then cropped out the
central 256x256 patch from the resulting image.
Have you considered using a graphcut (such as gimp's resynthesize) to reduce the image instead of cropping the internal region. I.e. graphcut original to a square - then scale.
http://www.logarithmic.net/pfh/resynthesizer
On most random imagery this approach might give improved results.
Of course I have no data as to the relative squareness of the original images and how much image information was "lost" to cropping.
The paper describes your initial image processing approach as:
Have you considered using a graphcut (such as gimp's resynthesize) to reduce the image instead of cropping the internal region. I.e. graphcut original to a square - then scale. http://www.logarithmic.net/pfh/resynthesizer
On most random imagery this approach might give improved results. Of course I have no data as to the relative squareness of the original images and how much image information was "lost" to cropping.