These are examples of an image restoration technique applicable even when as much as 85% of the pixels are corrupted. or unknown, as in a sparse image. The examples are of some general interest, but were placed on the Web to aid image engineering researchers. The technical details of the algorithm used is briefly outlined in the paper, Harvey A. Cohen, "Image Restoration via N-Nearest Neighbour Classification", Proc. 1996 IEEE International Conference on Image Processing, Lausanne, Switzerland, 16-19 Sept 1996, Vol 1, pp 1005-1008. This is available as Paper 96:10 in my Paper Download list.
Back to Image Engineering Index
Clown 85% corrupted with black impulse noise
Clown 85% corrupted with black impulse restored
Clown Image uncorrupted
Clown 85% corrupted with white impulse noise restored
The restoration algorithm used requires the determination of the distance of
each corrupt pixel from the nearest good pixels, using
Rosenfeld-Pfaltz distance operator.
The pretty random wallpapers
produced during this process are part of the story
on the page on Distance Transformation Magic, Auras, and Gobblegook.
Lena 85% corrupted FD - BK as per paper
Lena 85% corrupted FD as per paper
Lena 85% corrupted white noise
Lena 85% corrupted Restored
Lena Image Original
Harvey Cohen's Home Page
Back to Image Engineering Index