2015 | OriginalPaper | Buchkapitel
Maximizing Flows with Message-Passing: Computing Spatially Continuous Min-Cuts
verfasst von : Egil Bae, Xue-Cheng Tai, Jing Yuan
Erschienen in: Energy Minimization Methods in Computer Vision and Pattern Recognition
Verlag: Springer International Publishing
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
In this work, we study the problems of computing spatially continuous cuts, which has many important applications of image processing and computer vision. We focus on the convex relaxed formulations and investigate the corresponding flow-maximization based dual formulations. We propose a series of novel continuous max-flow models based on evaluating different constraints of flow excess, where the classical pre-flow and pseudo-flow models over graphs are re-discovered in the continuous setting and re-interpreted in a new variational manner. We propose a new generalized proximal method, which is based on a specific entropic distance function, to compute the maximum flow. This leads to new algorithms exploring flow-maximization and message-passing simultaneously. We show the proposed algorithms are superior to state of art methods in terms of efficiency.