Stereo Depth Extraction From Directional Derivative
Chienchung Chang, Sandip Chatterjee · 2005
Previous works on matching gray-scale images for stereopsis assume that the two optical axes of the cameras are parallel to each other. Under this restriction the two epipolar lines are parallel. In this paper, we assume that the two cameras are rotated (and translated) with respect to each other so that the epipolar lines are not parallel any more. Based on the estimates of the first directional derivatives (extracted by fitting surfaces), a neural network implementation is done for stereo matching. The depth information is extracted from the location disparity thereafter.