Quantum-Hybrid Stereo Matching with Nonlinear Regularization and Spatial Pyramids

Cameron Braunstein, Eddy Ilg, Vladislav Golyanik

Abstract:

Quantum visual computing is advancing rapidly. This023
paper presents a new formulation for stereo matching with024
nonlinear regularizers and spatial pyramids on quantum025
annealers as a maximum a posteriori inference problem that026
minimizes the energy of a Markov Random Field. Our ap-027
proach is hybrid (i.e., quantum-classical) and is compatible028
with modern D-Wave quantum annealers, i.e., it includes a029
quadratic unconstrained binary optimization (QUBO) ob-030
jective. Previous quantum annealing techniques for stereo031
matching are limited to using linear regularizers, and thus,032
they do not exploit the fundamental advantages of the quan-033
tum computing paradigm in solving combinatorial opti-034
mization problems. In contrast, our method utilizes the035
full potential of quantum annealing for stereo matching, as036
nonlinear regularizers create optimization problems which037
are N P-hard. On the Middlebury benchmark, we achieve038
an improved root mean squared accuracy over the previ-039
ous state of the art in quantum stereo matching of 2% and040
22.5% when using different solvers.