Dense Labeling with User Interaction: an Example for Depth-Of-Field Simulation

Ana Cambra, Adolfo Muñoz, Jose Guerrero and Ana Murillo

Abstract

This work presents a method to achieve dense labeling in a very simple and efficient way, faster and with more flexibility than related approaches. We use an initial superpixel graph and reformulate its constraints as a sparse linear system of equations, which is efficiently solved as a linear least squares problem. We demonstrate our approach with an interactive application for depth-of-field simulated effects based on dense depth estimation from a single image. This kind of interactive applications requires to continuously obtain a dense labeling estimation, therefore it is necessary to solve the task with high efficiency. Our experiments show our method obtains comparable results for a dense depth estimation against related approaches, while providing a more generic and powerful representation of the problem. The proposed dense labeling system opens new opportunities to design interactive applications that require dense labeling estimation of any other image property.

Session

Posters 2

Files

Extended Abstract (PDF, 780K)
Paper (PDF, 3M)
Supplemental Materials (ZIP, 19M)

DOI

10.5244/C.30.102
https://dx.doi.org/10.5244/C.30.102

Citation

Bibtex

        @inproceedings{BMVC2016_102,
        	title={Dense Labeling with User Interaction: an Example for Depth-Of-Field Simulation},
            author={Ana Cambra, Adolfo Mu\~{n}oz, Jose Guerrero and Ana Murillo},
        	year={2016},
        	month={September},
        	pages={102.1-102.13},
        	articleno={102},
        	numpages={13},
        	booktitle={Proceedings of the British Machine Vision Conference (BMVC)},
        	publisher={BMVA Press},
        	editor={Richard C. Wilson, Edwin R. Hancock and William A. P. Smith},
        	doi={10.5244/C.30.102},
        	isbn={1-901725-59-6},
        	url={https://dx.doi.org/10.5244/C.30.102}
        }