Interactive specification and acquisition of depth from single images

Abstract

We describe a system for interactively acquiring depth for an image-based representation consisting of a single input image. We use layers of images with depth to represent the scene. Unlike traditional 3D modeling and rendering systems that require precise information that are usually difficult to model and manipulate, our system's emphasis is on ease of use, comprehensiveness, and use of potentially crude depth information. Depth is extracted by the user through intuitive, interactive tools using the powerful notion of selection. Once a set of pixels is selected, the user can assign depth by painting and chiseling, using shape from shading, applying filters, aligning and extracting shape from geometry primitives, or using level set methods. The ground plane tool provides an intuitive depth reference for all other tools and serves as an initial step in depth specification. Our system is based on pixels and selections, and therefore does not impose any restriction on the scene geometry. Our system allows the user to interactively perform high quality editing operations on SGI 02s and Octanes. We demonstrate the application of our system in the architectural design (relighting, sketching, 3D walkthroughs from images), complex photocompositing, and fine art exploration contexts.