253 
Mitsuteru SAKAMOTO, Wei LU, Pingtao Wang & Yukio KOSUGI 
(a) Mapped Left Image (b) Depth Map Projected to Left Image 
Fig.15 The CEM Result with Enhanced Consensus Operation 
Fig. 16 The Application Result with Cl Operator 
(a) Mapped Left Image (b) Depth Map Projected to Left Image 
Fig. 17 The CEM Result with Cl Operated Images 
(a) Mapped Left Image (b) Depth Map Projected to Left Image 
Fig. 18 The CEM Result with Edge Support 
In edge detection and matching, two types of different operator, 
SUSAN and Haar’s wavelet are investigated. Experimental 
results show that detection of edge segment with SUSAN 
operator can provide more valid results. In the case of wavelet, 
another process has to be considered instead of binarization. It 
is verified that the proposed edge matching technique works 
satisfactorily. 
We presented an example of stereo matching result in urban 
area with CEM, where the conjugate edge segments are utilized 
as global search information and immovable points in CEM 
process. Experimental results show that the balance of 
parameters between competition phase and consensus phase is 
important. The efficiency of Cl operator was tested and it is 
observed that Cl operator can produce promising results for 
stereo matching of buildings. However more improvements are 
needed. CEM with edge segments constraints showed 
improving around building edges. In the present, CEM process 
of edge points and other areas are not definitely distinguished, 
therefore local disturbances occur at some areas. More 
investigation is necessary about this subject. 
For further improvement, handling of edge points and other 
areas have to be separately processed in CEM with edge 
constraints. It is also important to investigate adjusting 
processing parameters adaptively for object’s size in image. 
In the next stage, we will try to study detecting buildings with 
edges and depth map derived from in this study. 
REFERENCES 
[1] Mitsuteru Sakamoto, Osamu Uchida, Takeshi Doihara, 
Kazuo Oda, Wei Lu and Masayoshi Obata, "Geo-Plotter a 
Softcopy Mapping System for Low Cost Digital Mapping 
Process", IAPRS XXXIII-B4/2, pp.889-892, 2000. 
[2] Mitsuteru Sakamoto, Takeshi Doihara and Yukio Kosugi, "A 
Study on Improvements of Stereo Matching Techniques for 
Urban Area - Automatic Estimation of Camera Parameters and 
Utilization of Edge Segments Technical Report of IEICE, 
Vol.100, No.499, pp.115-120, 2000. 
[3] S.M.Smith and J.M.Brady, "SUSAN -A New Approach to Low 
Level Image Processing", Int. Journal of Computer Vision, 
Vol.23, No.1, pp.45-78, 1997. 
[4] Y. Kosugi, M. Sase, H. Kuwatani et al., “Neural Network 
Mapping for Nonlinear Stereotactic Normalization of Brain MR 
Images”, Journal of Computer Assisted Tomography, Vol.17, 
No.3, pp.445-460, 1993. 
[5] Yukio Kosugi, Munenori Fukunishi, Mitsuteru Sakamoto, Wei 
Lu and Takeshi Doihara, "Detection of Sheer Changes in Aerial 
Photo Images Using An Adaptive Nonlinear Mapping”, to be 
presented in Geoinformatics & DMGIS'2001. 
7. CONCLUSION AND FUTURE WORKS 
We have introduced a new stereo matching approach that 
combines edge matching results and nonlinear matching with 
CEM.