This item is only available for download by members of the University of Illinois community. Students, faculty, and staff at the U of I may log in with your NetID and password to view the item. If you are trying to access an Illinois-restricted dissertation or thesis, you can request a copy through your library's Inter-Library Loan office or purchase a copy directly from ProQuest.
Permalink
https://hdl.handle.net/2142/22674
Description
Title
Motion modeling and video processing
Author(s)
Rajagopalan, Rajesh
Issue Date
1996
Doctoral Committee Chair(s)
Orchard, Michael T.
Department of Study
Electrical and Computer Engineering
Discipline
Electrical and Computer Engineering
Degree Granting Institution
University of Illinois at Urbana-Champaign
Degree Name
Ph.D.
Degree Level
Dissertation
Keyword(s)
Engineering, Electronics and Electrical
Language
eng
Abstract
Knowledge of motion fields is crucial to several applications such as video coding, image scene analysis and noise reduction. Estimation of this field is frequently done using constraints such as smoothness deduced from physical considerations of the process generating the video. Smoothness of motion is a qualitative statement regarding local relationships of elements of this field. In this work, our primary focus is on quantitatively modeling the relationships between elements of the motion field at spatial neighborhoods of pixels and in filtering of motion. These are accomplished by generalizing popular techniques in statistical signal processing--autoregressive (AR) models and moving average (MA) filtering. First, we show an equivalence between estimates from AR models (output of MA filtering) to the solution of a weighted least squares problem. This least squares problem is then generalized to enable modeling (filtering) of motion fields. Our AR model for motion is significantly different from previous approaches in that instead of computing motion at a pixel as a linear combination of motion at a spatial neighborhood of pixels, we compute the motion at a pixel using the observable data (i.e., pixel intensities) directly. An extension of this temporal AR model to a joint spatiotemporal model is also presented. Applications to interframe estimation reveal that interframe prediction accuracy is improved over previous methods by as much as 37%.
A temporal MA filtering formulation is proposed and applied to preprocessing video prior to coding. Preprocessing results indicate that coding gains using MPEG1 of 20% may be obtained while maintaining the same level of visual quality of decoded pre-processed video as compared to the decoded original sequence. Extending the temporal filtering to a joint spatiotemporal filtering, we propose algorithms for noise reduction. At low and moderate signal to noise ratios, our algorithms perform reasonable well, but worse than the best results in the literature, while at high SNR, they perform better. It is believed that with improved estimation of parameters used in the algorithms, performance may be improved.
Aside from the above main focus, we also present investigations into two other issues: (1) Efficient motion estimation algorithms for overlapped block motion compensation--we present algorithms that can trade off computational complexity for prediction accuracy in an efficient manner and (2) Supports to be used in linear predictive models--algorithms are presented which compute supports yielding up to 37% improvements in prediction over nearest neighbor based supports.
Use this login method if you
don't
have an
@illinois.edu
email address.
(Oops, I do have one)
IDEALS migrated to a new platform on June 23, 2022. If you created
your account prior to this date, you will have to reset your password
using the forgot-password link below.
