Notes:

Started with 50 van Hateren images used by David Warland (vanHateren_rnd.txt)

Took log of images, subtract mean(log), they look about right displayed from [-1 1]

Inspected power spectra, from this identified some suspicious images, some appear blurred due to camera jitter, etc.: 936, 1266, 1300, 1939, 2286, 2537, 2659, 2821, 3300, 3649, 3797, 3988, 4068.  
Deleted them from David's list to create vanHateren_bao.xls, now down to 35 images.

Whitened/lowpass and down sampled to 512x512 using whiten_images.m  - this creates array IMAGES.

Images 8, 21, and 30 of this array seem to be outliers in terms of variance.  Curiously these are all images of trees.  Possibly eliminate these?

Power spectrum of whitened images looks good, images look good, read to go.