| 摘要 |
Method for recovering a relative depth map from a single image or a sequence of still images, comprising: -decomposing the image into sets Uλ, each formed by pixels whose intensity is in a predefined interval; decomposing sets Uλ into connected components Ci λ, a subset of adjacent pixels according to a connectivity rule; the probability of each pixel x ∉ Ci λ to belong to it and being a function of: the distance d(x, Ci λ) from the pixel x to said connected component Ci λ; the curvature &kgr;(χ, Ci λ) of the level line of the distance function d(x, Ci λ) from the pixel x to said connected component Ci λ; for each pixel x of the image, compute a local depth feature Z(x) indicative of the probability of the pixel x belonging simultaneously to different connected components and which provides a relative depth order estimation at pixel x. |
