Information Technology Reference
In-Depth Information
Chapter 23
Stereo Correspondence in Information Retrieval
Huiyu Zhou and Abdul H. Sadka
Abstract.
Stereo correspondence is a very important problem in information re-
trieval. Optimal stereo correspondence algorithms are used to generate optimal dis-
parity maps as well as accurate 3-D shapes from 2-D image inputs. Most established
algorithms utilise local measurements such as image intensity (or colour) and phase,
and then integrate the data from multiple pixels using a smoothness constraint. This
strategy applies fixed or adaptive windows to achieve certain performance. To build
up appropriate stereo correspondences, a global approach must be implemented in
the way that a global energy or cost function is designed by considering template
matching, smoothness constraints and/or penalties for data loss (e.g. occlusion).
This energy function usually works with optimisation methods like dynamic pro-
gramming, simulated annealing and graph cuts to reach the correspondence. In this
topic chapter, some recently developed stereo correspondence algorithms will be
summarised. In particular, maximum likelihood estimation-based, segment-based,
connectivity-based and wide-baseline stereo algorithms using descriptors will be
introduced. Their performance in different image pairs will be demonstrated and
compared. Finally, future research developments of these algorithms will be pointed
out.
1
Introduction
Digital video cameras are widely used in our community, and the quantity of digital
videos has significantly increased up to date. For the reuse and storage purpose,
consumers have to retrieve a video from a large number of multimedia resources. To
