My main research interests are in image processing, human vision, and multiview computer vision. Specifically, my work concerns the application of properties of human binocular vision to compression and processing of stereo imagery. Click on a topic below for more information.
Discriminability Limits in Spatio-Temporal Stereo Block Matching
We analyze spatio-temporal block matching for disparity estimation. First, we determine conditions under which spatio-temporal stereo matching is a valid measure of disparity based on object motion and camera parameters. Then, we derive expressions for the probability of a false match among spatially similar candidates for a target patch in one view as a function of the distribution of motion, image features, and noise. These results show how efficient motion can be at disambiguating stereo correspondences.
Video Super-Resolution for Mixed Resolution Stereo
In this project, we recover the discarded high-frequency content from mixed resolution stereoscopic coding. We project the high-frequency information from the full resolution view onto the low-resolution view by matching patches using a spatio-temporal Markov network. Our approach provides spatial, spectral, and temporal consistency.
Tally – The Web-Based Subjective Testing Tool
We introduce Tally, a modern tool to help automate data collection for subjective video quality experiments. Its web-based design allows for ubiquitous data access, project sharing and collaboration, and compatibility with mobile devices. Tally can be used for both 2D and 3D videos, any type of display or resolution, and testing multiple subjects simultaneously.
The goal of this project is to synthesize virtual images at any point along the baseline of an input rectified stereo pair.
3D Artifact Analysis
We study various artifacts that appear in 3D video, both the manifestation of 2D artifacts in 3D as well as artifacts that are exclusively exhibited in stereo.