The coordinates of the three vertical boundaries in patches 1, 2 and 3 can be downloaded into an ASCII file.
Patches 1, 2 and 4 contain aerial textures, 5 is synthetic whereas 3 uses Brodatz. Normally patches 4 and 5 give good results with methods for micro-texture (cooccurrence matrix etc); the others with methods for macro-texture (Gabor filters). Most patches are not easy, but number one is a real challenge!
The file format is 256x256 bytes, 1 byte/pixel, grayscale GIF. You can click on an image in order to download it in GIF format, and you need to use xv or ImageMagick to convert it to raw format that you can easily read in your programs.
You are FREE to use any of these images in your work and publications, provided that
you refer to at least one of the publications given below under References!
If you inform me, I will include your links with results or publications under Ongoing Research in order to facilitate the comparison of results!
patch1.gif - aerial
patch2.gif - aerial
patch3.gif - brodatz
patch4.gif - aerial
patch5.gif - synthetic
Most images have been used in the following publications on feature extraction and segmentation:
Jan Puzicha, who is doing his PhD at Bonn University, is getting very nice results using histogram grouping. For methods and results, see the Bonn Computer Vision Group.
Cristobal, G. and Hormigo, J. (1999) Texture segmentation through eigen-analysis of
the Pseudo-Wigner distribution. Pattern Recognition Letters 20, pp. 337-345.
Good results for patch 3 and reasonable ones for patch 1, but messy boundaries (well, see also my own distinct boundary regions (du Buf, 1992)
Please send comments to Hans du Buf:
Please visit the Vision Laboratory.
Last update: October 1999 (HdB)