Stereoscopic correspondence by applying physical constraints and statistical observations to dissimilarity map

T. Y. Chao*, Sheng-Jyh Wang, Hsueh-Ming Hang

*Corresponding author for this work

Research output: Contribution to journalConference articlepeer-review


To deal with the correspondence problem in stereo imaging, a new approach is presented to find the disparity information on a newly defined dissimilarity map (DSMP). Based on an image formation model of stereo images and some statistical observations, two constraints and four assumptions are adopted. In addition, a few heuristic criteria are developed to define a unique solution. All these constraints, assumptions and criteria are applied to the DSMP to find the correspondence. At first, the Epipolar Constraint, the Valid Pairing Constraint and the Lambertian Surface Assumption are applied to DSMP to locate the Low Dissimilarity Zones (LDZs). Then, the Opaque Assumption and the Minimum Occlusion Assumption are applied to LDZs to obtain the admissible LDZ sets. Finally, the Depth Smoothness Assumption and some other criteria are applied to the admissible LDZ sets to produce the final answer. The focus of this paper is to find the constraints and assumptions in the stereo correspondence problem and then properly convert these constraints and assumptions into executable procedures on the DSMP. In addition to its ability in estimating occlusion accurately, this approach works well even when the commonly used monotonic ordering assumption is violated. The simulation results show that occlusions can be properly handled and the disparity map can be calculated with a fairly high degree of accuracy.

Original languageEnglish
Pages (from-to)78-89
Number of pages12
JournalProceedings of SPIE - The International Society for Optical Engineering
StatePublished - 3 May 2000
EventStereoscopic Displays and Virtual Reality Systems VII - San Jose, CA, USA
Duration: 24 Jan 200027 Jan 2000

Fingerprint Dive into the research topics of 'Stereoscopic correspondence by applying physical constraints and statistical observations to dissimilarity map'. Together they form a unique fingerprint.

Cite this