We have developed a new method for surface reconstruction from stereo images. For a given initial seed point, i.e. for a pair of corresponding points in the left and right image, the proposed algorithm grows the surface without directly computing the point correspondences. The method assumes the Lambertian surface reflectance model. Our approach is based on a surface normal calculation from the left and right image gradients. Knowing the surface normal, the algorithm grows the surface in the directions defined by the tangent plane. This is illustrated in Fig. 1, while Fig. 2 shows a zoomed-in view of the initially grown surface superimposed on the two camera views and Fig. 3 shows the reconstructed surface. The algorithm is independent of the camera model, and requires the placement of an initial seed point for each surface to be reconstructed. Experimental results on real brain images show that the surface reconstruction algorithm is robust to the position of the initial seed point.

Figure 1: The left and right view of the exposed brain surface. In (a) and (b) the calibration object is the white block at the left. Note a few spots on the brain surface that have specularities. Images (c) and (d) show estimated surface normals together with the corresponding patches. Projections to the left and right image of the grown surface are shown in images (e) and (f). The images were initially masked to exclude the specular regions. The figure is from [1] and it is used with permission; Copyright © 2000 IEEE; All rights reserved.

Figure 2: A closer look at the corresponding surface patches shows very good match. The two images show the left and right projection of the early stage of surface growing. The central grid point was the initial seed point. The figure is from [1] and it is used with permission; Copyright © 2000 IEEE; All rights reserved.

Figure 3: Reconstructed exposed brain surface. The figure is from [1] and it is used with permission; Copyright © 2000 IEEE; All rights reserved.

References:

[1] Skrinjar, O., Tagare, H., Duncan, J., "Surface Growing from Stereo Images", IEEE Conference on Computer Vision and Pattern Recognition, Hilton Head Island, SC, USA, Volume 2, pp. 571-576, June 2000. LINK