A Hybrid Quantum-Classical Segment-Based Stereo Matching Algorithm

Our contribution introduces a hybrid quantum-classical Stereo Matching algorithm that demonstrates the potential of using Quantum Annealing in Computer Vision in the future once quantum processors have enough qubits to support practical, real-world Computer Vision applications. The classical component of our approach involves dividing the input image into homogeneous color segments and using a local Stereo Matching technique to estimate their respective initial disparity planes. In the quantum component, we assign a label to each segment by the estimated disparity planes. Such a labeling problem is classically intractable. The outcomes of our experiments on the D-Wave quantum computer indicate that our method produces results that compare well with the ground truth. Nonetheless, the precision of our approach is greatly influenced by the quality of the initial image segmentation, which is a common challenge for all classical Stereo Matching methods that rely on segmentation-based techniques.