ie gray
À good
d truth
1 small
moved
shown
where
Herbert Jahn
The well-known image pair Pentagon (figure 5) shows the limits of the present status of the algorithm better because in
that image pair are many occlusions and especially small non-overlapping structures with big disparities which the
algorithm cannot handle satisfactorily up to now. The x- disparity image shown in figure 6 reveals this.
Figure 5. Stereo pair “Pentagon” Figure 6. Disparity image
4 CONCLUSIONS
The few results presented here show that the algorithm works in some image pairs generated with aerial stereo cameras.
In section 2 some remarks concerning necessary future research were already made which shall be supplemented now.
Because of its local gradient computation (5), (7) the algorithm gives only good results when the disparities are small,
i.e. if there is an overlapping of structures in both images. To cope with big disparities the gradient computation must be
extended to a non-local operation. First experiments have given promising results but more investigations are necessary.
Furthermore, the J-function (11) to be minimized has to be generalized. More geometric rather than radiometric
information should be included because some disturbances which occur in only one image of the stereo pair (e.g.
reflections of sun light) can lead to wrong disparities.
REFERENCES
Belhumeur, P. N., 1996. A Bayesian Approach to Binocular Stereopsis, Int. J. of Computer Vision, Vol. 19, pp. 237-
260
Gimel'farb, G., 1999, Stereo Terrain Reconstruction by Dynamic Programming, in: Handbook of Computer Vision and
Applications (eds. B. Jáhne, H. Haussecker, and P. Geisser), Vol. 2, Academic Press, San Diego, pp. 505-530
Himmelblau, D. M., 1972. Applied nonlinear programming, Mc Graw-Hill, New York
Hubel, D. H., 1995. Eye, Brain, and Vision, Scientific American Library, New York
Jahn, H., Reulke, R., 1995. Systemtheoretische Grundlagen optoelektronischer Sensoren, Akademie Verlag, Berlin
Jahn, H., 1998. A neural network for image smoothing and segmentation, Proc. of Joint IAPR Int. Workshops SSPR'98
and SPR'98, pp. 329-338; Lecture Notes in Computer Science 1451, Springer
Jahn, H., 1999, Feature grouping based on graphs and neural networks, Proc. of CAIP'99, Lecture Notes in Computer
Science 1689, pp. 568-577, Springer
Jolion, J.-M., Rosenfeld, A., 1994. A Pyramid Framework for Early Vision, Kluwer Academic Publishers, Dordrecht
Kim, Y.-S., Lee, J.-J., Ha, Y.-H., 1997. Stereo matching algorithm based on modified wavelet decomposition process;
Pattern Recognition, Vol. 30, pp. 929-952
Klette, R., Schlüns, K., Koschan, A., 1998. Computer Vision. Three-Dimensional Data from Images, Springer
Wei, G.-Q., Brauer, W., Hirzinger, G., 1998. Intensity- and Gradient-Based Stereo Matching Using Hierarchical
Gaussian Basis Functions, IEEE PAMI, Vol. 20, pp. 1143-1160
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 443