Model-Aided Coding of Multi-Viewpoint Image Data

The paper presents a novel coding technique based on approximate geometry for images taken from arbitrary recording positions around a 3-D scene. Such data structures occur in image-based rendering applications, where many hundreds to thousands of images need to be stored and transmitted. After reconstructing and compressing the scene's approximate geometry, the 3-D model is used to infer disparity, occlusions and object silhouette, leading to improved prediction of images recorded from arbitrary angles. The images are hierarchically coded to ensure efficient exploitation of inter-image similarities. The coding scheme is capable of generating novel views by texturizing the available geometry model. The presented algorithm is validated using synthetic as well as natural image data sets, achieving up to 1000:1 compression at acceptable reconstruction quality.

• Type: Conference
• Year: 2000
• Where: ICIP
• Authors: Marcus A. Magnor, Peter Eisert, Bernd Girod
• Keywords: Approximate Geometry, Arbitrary Recording Positions, Available Geometry Model, ICIP 2000, Image Processing