220 likes | 373 Views
Energy functions. f(p) {0,1} : Ising model Can solve fast with graph cuts V(,) = T[] : Potts model NP-hard Closely related to Multiway Cut Problem Local minimum via expansion move algorithm. Example:. Left image. Right image. Stereo. Potts model for stereo. Multiway cut problem.
E N D
Energy functions • f(p) {0,1}: Ising model • Can solve fast with graph cuts • V(,) = T[]: Potts model • NP-hard • Closely related to Multiway Cut Problem • Local minimum via expansion move algorithm
Example: Left image Right image Stereo
t-link n-link
Green expansion move Expansion moves
initial solution -expansion -expansion -expansion -expansion -expansion -expansion -expansion Expansion moves in action For each move we choose expansion that gives the largest decrease in the energy: binary energy minimization subproblem
Binary image Binary sub-problem Input labeling Expansion move
Expansion move energy Goal: find the binary image with lowest energy Binary image energy depends on f,
Binary image notation Also depends on f,!
Original (non-binary) data energy: Sum this function over pixels p Binary data energy (given f,)
Original (non-binary) smoothness energy: Sum this function over neighboring pixels p,q Binary smoothness energy
Binary energy minimization • Finding the cheapest expansion move requires minimizing • Can be done efficiently by graph cuts!
Graph cuts solution • This can be done as long as V has a specific form (works for arbitrary D) • Regularity constraint: for f, we need
Regular choices of V • Suppose that V is a metric • Then what?
Potts model Linear model Truncated linear model Metric choices of V
Potts model Linear model Quadratic model Truncated linear model Robust Not robust
truncated linear V linear V Robustness matters!
Potts regularity (the hard way) Case f(p)=f(q)=: √ √ Case f(p)=,f(q): Case f(p),f(q): √