User-Assisted Disparity Maps

Abstract

In this paper, we present an intuitive and efficient system for correcting the artifacts and noises caused by imperfect disparity estimation. Disparity estimation is a very important and basic process for almost all stereoscopic 3D applications. Although there are many disparity estimation methods, none of them are perfect and suitable for any arbitrary case. Given a defect disparity map estimated from a stereoscopic image pair, our system incorporates some simple user hints and rich information derived from sparse feature matching as guides for disparity correction. We also utilize 3D connectivity prior and planarity assumption to overcome the challenge presented by textureless surfaces and occluded regions. By providing an intuitive user interface with integrated stereo visualization, our system allows novel users to correct their own disparity maps easily.