Edge and Boundary interpretation Consistent line drawing labeling via backtracking Presented by Guy Shtub
Perceiving 3D from 2D ► How can humans and machines reconstruct the 3D nature of a scene from a 2D image representing them? ► Additional knowledge is needed
Assumptions ► Trihedral (formed by 3 planes meeting at a point) world. ► General viewpoint ► Edges represent surface and depth continuities
Huffman and Clowes catalog 1971 ► Define edge type symbols:
► Define 16 possible edge intersections
Constraint Satisfaction Problems (CSPs) ► Consist of a set of variables ► Each variable must be assigned a value from a possible domain ► Backtracking attempts to find a solution be trying all combinations in order to find a solution ► Line labeling is a CSP
Naïve Backtracking – 4 queens Example
Forward Checking (FC) - 4 Queens Example
Problem Representation ► Edge intersections are variables - N ► 16 catalog possibilities are the domain for each variable - D ► Variable Conflicts Matrix NxN (adjacent intersections) ► For each pair of adjacent intersections a Constraints matrix DxD ► Both Algorithms implemented in Java
Algorithm Performance Evaluation ► Constraint Checks (CCs) – machine and implementation independent
Results
Results – performance
Conclusions ► Written Program able to provide consistent line labeling ► As expected FC performs better the Naïve Backtracking ► Future work: comparison to line labeling via relaxation labeling performance ► Questions?
References ► M. B. Clowes, "On seeing things," Artificial Intelligence, Vol. 2, No. 1, pp , ► D.A. Huffman, “Impossible objects as nonsense sentences,” Machine Imlligencc 6, pp, ,1971. ► Depth and Shape Inference (II), Introduction to Computational and Biological Vision, Computer Science Department, BGU, Ohad Ben-Shahar ► P. Prosser: "Hybrid algorithms for the constraint satisfaction problem", Computational Intelligence, Vol. 9, pp , 1993.