Nehodí se? Vůbec nevadí! U nás můžete do 30 dní vrátit
S dárkovým poukazem nešlápnete vedle. Obdarovaný si za dárkový poukaz může vybrat cokoliv z naší nabídky.
30 dní na vrácení zboží
The problem of extracting an object from an image,known as "image matting," has been deeplyinvestigated in the last two decades. Its extensionto address video, i.e., "video matting," has gainedless attention. This paper studies the difficultiesthat should be addressed when extending image mattingmethods to work on video. In particular we addressthe following problems: (i) While the required userintervention is reduced over the years, it stillremains unpractical to manually provide user marks onevery frame; (ii) the temporal continuity of a videosequence suggests that it may be beneficial to solvethe matting equations for the entire videosimultaneously. This, however, is impractical due tomemory constraints; (iii) because of the huge size ofinput data, interactive results update is nottrivially plausible. To cope with these difficulties,we exploit video continuity. Temporal continuity isused to propagate user provided hints, while spatialcontinuity is used to reduce the number of unknownsand the number of constraints. Finally, we apply anincremental matrix factorization to allow real-timeinsertion of new user-provided constraints.