Correspondence Analysis for Binocular Stereo (Shirai's Algorithm)
Christian
Graf
Uli Schroeder
YongTao Zou
Here is the output of our program using different d1,d2 and d3 (set bold in the captions):
./ass2shirai 1 source\ BMP/complex results\ JPG/complex_01_05.jpg 1 5 5 0.3./ass2shirai 1 source\ BMP/complex results\ JPG/complex_01_10.jpg 1 10
10 0.3
./ass2shirai 1 source\ BMP/complex results\ JPG/complex_05_10.jpg 5 10
10 0.3
./ass2shirai 1 source\ BMP/complex results\ JPG/complex_10_20.jpg 10
20 20 0.3
./ass2shirai 1 source\ BMP/complex results\ JPG/complex_10_50.jpg 10
50 50 0.3
./ass2shirai 1 source\ BMP/complex results\ JPG/complex_20_50.jpg 20
50 50 0.3
./ass2shirai 1 source\ BMP/complex results\ JPG/complex_30_50.jpg 30
50 50 0.3
Shirai's algorithm is time and resource intensive, even if it calcullated disparities just for edge points. Noise images without a prior smoothing could extent the running time heavily, so a smooting and median could be advantegous.
The implementation of the algorithm depends heavily on the datastructures used. We experienced that fact our own.