Photometric Ambient Occlusion

Abstract

We present a method for computing ambient occlusion (AO) for a stack of images of a scene from a fixed viewpoint. Ambient occlusion, a concept common in computer graphics, characterizes the local visibility at a point: it approximates how much light can reach that point from different directions without getting blocked by other geometry. While AO has received surprisingly little attention in vision, we show that it can be approximated using simple, per-pixel statistics over image stacks, based on a simplified image formation model. We use our derived AO measure to compute reflectance and illumination for objects without relying on additional smoothness priors, and demonstrate state-of-the art performance on the MIT Intrinsic Images benchmark. We also demonstrate our method on several synthetic and real scenes, including 3D printed objects with known ground truth geometry.

Intrinsic Decomposition of Image Sequences from Local Temporal Variations

“…On this dataset, only the method by Hauagge et al [12] achieves better LMSE results. Note however, that the MIT dataset only contains isolated objects captured in laboratory conditions, with grayscale shading and no interreflections.…”

“…Weiss et al [34] use a prior derived from the image statistics of natural scenes, while Matshushita et al [26] explicitly model temporal and spatial constraints. Hauagge et al [12] propose a simplified physical model for modeling the local visibility, with a moving directional light source and constant ambient lighting. Sunkavalli et al [31,32] target outdoor scenes and model the lighting as a mixture of two light sources: directional sunlight and ambient skylight.…”

“…Examples of mean runtimes of existing methods as reported in the literature are: [10]: >600s, [29]: >300s, [33]: >200s, [2]: >200s, [16]: 40s, [36]: 3s, [24]:1-3s, Retinex [11]: 1s. On our PC equipped with a Core i5 2.6 Ghz CPU, [12] runs in 257s and ours in 94s on average.…”

“…Recent methods have successfully tackled this problem by incorporating priors on local shading [22,33], enforcing sparse reflectance [10,5], propagating user-specified constraints [7], leveraging reconstructed scene geometry [19,1], or inferring constraints from multiple images with varying lighting [34,12]. Like these latter approaches, we leverage multiple observations of the scene to constrain the decomposition.…”

Shadow Detection and Sun Direction in Photo Collections

“…Prior work has also sought to estimate illumination properties from either shadows cast on a planar surface [22,20,13,21], or from shading information [19]. In a closely related method, Hauagge et al [10] use an intrinsic image technique that considers ambient occlusion [9] to determine shading and illumination outdoors. They then compare the illumination estimate against renderings with a physically-based sun/sky model to determine sun direction.…”