A decision-theoretic generalization of on-line learning and an application to boosting
EuroCOLT '95 Proceedings of the Second European Conference on Computational Learning Theory
Distinctive Image Features from Scale-Invariant Keypoints
International Journal of Computer Vision
A no-reference objective image sharpness metric based on the notion of just noticeable blur (JNB)
IEEE Transactions on Image Processing
Feature-Assisted Sparse to Dense Motion Estimation Using Geodesic Distances
IMVIP '09 Proceedings of the 2009 13th International Machine Vision and Image Processing Conference
Motion Estimation for Regions of Reflections through Layer Separation
CVMP '11 Proceedings of the 2011 Conference for Visual Media Production
Reflection detection in image sequences
CVPR '11 Proceedings of the 2011 IEEE Conference on Computer Vision and Pattern Recognition
Separating transparent layers in images and video
Separating transparent layers in images and video
Analysis of Superimposed Oriented Patterns
IEEE Transactions on Image Processing
| Hi-index | 0.00 |
Reflections in image sequences violate the single layer model used by most current image processing techniques. As a result reflections cause many techniques to fail e.g. detection, tracking, motion estimation, etc. Recent work was proposed by Ahmed et al. [5] to detect reflections. Their technique is robust to pathological motion and motion blur. This paper has three main contributions. The first simplifies and fully automates the technique of Ahmed et al. User feedback is common in post-production video manipulation tools. Hence in the second contribution we propose an effective way of integrating few user-assisted masks to improve detection rates. The third contribution of this paper is an application for reflection detection. Here we explore better feature point tracking for the regions detected as reflection. Tracks usually die quickly in such regions due to temporal color inconsistencies. In this paper we show that the lifespan of such tracks can be extended through layer separation. Results show reduction in missed detections and in computational load over Ahmed et al. Results also show the generation of more reliable tracks despite strong layer mixing.