Grayscale images and synthetic textures which are difficult to segment

Question

What are some real world images(grayscale) and synthetic textures on which segmentation is difficult to perform? (difficulty can be either computational or algorithmic)

Answer 1

Animals that blend into their environment are good candidates (just ignore two of the colors). There are some in the Berkeley Segmentation Data set found in the link below:

http://www.eecs.berkeley.edu/Research/Projects/CS/vision/bsds/

Technically, I would say two regions that have a similar spatial frequency spectrum (fft2) would be difficult to segment. Some of the animals have similar shapes and patterns but are usually too regular, too straight, or are simply different at different scales (size of the fft2). So the animals would have similar spatial spectra with (small?) differences.

Hope this helps.