Geometric Image Parsing in Man-Made Environments

Abstract

We present a new optimization based parsing framework for the geometric analysis of a single image coming from a man-made environment. This framework models the scene as a composition of geometric primitives spanning different layers from low level (edges) through mid-level (lines segments, lines and vanishing points) to high level (the zenith and the horizon). The inference in such a model thus jointly and simultaneously estimates (a) the grouping of edges into the line segments, (b) the grouping of line segments into the straight lines, (c) the grouping of lines into parallel families, and (d) the positioning of the horizon and the zenith in the image. Such a unified treatment means that the uncertainty information propagates between the layers of the model. This is in contrast to most previous approaches to the same problem, which either ignore the middle levels (line segments or lines) all together, or use the bottom-up step-by-step pipeline.For the evaluation, we consider a publicly available York Urban dataset of "Manhattan" scenes, and also introduce a new, harder dataset of 103 urban outdoor images containing many non-Manhattan scenes. The comparative evaluation for the horizon estimation task demonstrate higher accuracy and robustness attained by our method when compared to the current state-of-the-art approaches.

A MAP-Estimation Framework for Blind Deblurring Using High-Level Edge Priors

“…[8] on multi-view stereo reconstruction, where significant improvements are shown through the use of a "Manhattan world" assumption, and Barinova et al . [35] on horizon estimation with also excellent results.…”

“…Here we wish to be able to successfully take into account as many of these regularities as possible, imposing at the same time assumptions that are as general as possible. To that end, we are going to rely on a geometric image parsing prior similar to the one used in recent work [35]. Such a prior is generic enough and has already been shown to successfully apply to a wide variety of cases.…”

“…The individual terms appearing in (8) are defined similarly to [35], and essentially encode all the assumptions that we mentioned above regarding the regularities of the depicted scenes. We next briefly describe each of these terms (and we refer to [35] for a detailed explanation): -E pixel (p i |s, x): this term encodes how well an edge pixel p i of image x is explained by one of the edge segments included in set s…”

“…We next briefly describe each of these terms (and we refer to [35] for a detailed explanation): -E pixel (p i |s, x): this term encodes how well an edge pixel p i of image x is explained by one of the edge segments included in set s…”

“…Given that the weights w pq (e) depend only on the variables s (see (6)), and not on {l, h, z}, the resulting objective function is of similar form to the energy function used in [35], and can therefore be optimized using the same method, which is highly efficient.…”