Iteratively Reweighted Graph Cut for Multi-label MRFs with Non-convex Priors

TL;DR

This paper introduces an iterative algorithm for optimizing multi-label Markov Random Fields with non-convex priors, ensuring energy decrease at each step and leveraging graph cuts for efficient minimization.

Contribution

It presents a novel iterative reweighting approach that approximates non-convex energies with surrogate energies solvable by graph cuts, enabling effective optimization.

Findings

Outperforms state-of-the-art methods on stereo and inpainting tasks.

Guarantees energy decrease at each iteration.

Handles a wide range of non-convex priors effectively.

Abstract

While widely acknowledged as highly effective in computer vision, multi-label MRFs with non-convex priors are difficult to optimize. To tackle this, we introduce an algorithm that iteratively approximates the original energy with an appropriately weighted surrogate energy that is easier to minimize. Our algorithm guarantees that the original energy decreases at each iteration. In particular, we consider the scenario where the global minimizer of the weighted surrogate energy can be obtained by a multi-label graph cut algorithm, and show that our algorithm then lets us handle of large variety of non-convex priors. We demonstrate the benefits of our method over state-of-the-art MRF energy minimization techniques on stereo and inpainting problems.