東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back

Switch To: Labeled | MARC Mode | ISBD

Polynomial Models in Nonrigid Motion...

Liu, Wei.

Linked to FindBook

Google Book

Amazon

博客來

Polynomial Models in Nonrigid Motion Estimation and Analysis.

Record Type:	Language materials, printed : Monograph/item
Title/Author:	Polynomial Models in Nonrigid Motion Estimation and Analysis./
Author:	Liu, Wei.
Description:	181 p.
Notes:	Source: Dissertation Abstracts International, Volume: 72-06, Section: B, page: .
Contained By:	Dissertation Abstracts International72-06B.
Subject:	Computer Science. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3452323
ISBN:	9781124581446

Polynomial Models in Nonrigid Motion Estimation and Analysis.
Liu, Wei.

Polynomial Models in Nonrigid Motion Estimation and Analysis. - 181 p.

Source: Dissertation Abstracts International, Volume: 72-06, Section: B, page: .

Thesis (Ph.D.)--Florida Institute of Technology, 2011.

In this thesis, we study the usage of polynomial models in the estimation and analysis of nonrigid motions. We present these models in the context of two different applications, namely, shape registration and flow-field analysis. The common theme throughout this thesis is the use of polynomials as parametric models to present nonrigid motion. Within this common framework, we propose three novel methods for nonrigid motion estimation and analysis.

ISBN: 9781124581446Subjects--Topical Terms:

626642
Computer Science.

Polynomial Models in Nonrigid Motion Estimation and Analysis.
LDR:04621nam 2200325 4500 001 1403724
005 20111111143232.5
008 130515s2011 ||||||||||||||||| ||eng d
020 $a 9781124581446
035 $a (UMI)AAI3452323
035 $a AAI3452323
040 $a UMI $c UMI
100 1 $a Liu, Wei. $3 1260300
245 1 0 $a Polynomial Models in Nonrigid Motion Estimation and Analysis.
300 $a 181 p.
500 $a Source: Dissertation Abstracts International, Volume: 72-06, Section: B, page: .
500 $a Adviser: Eraldo Ribeiro.
502 $a Thesis (Ph.D.)--Florida Institute of Technology, 2011.
520 $a In this thesis, we study the usage of polynomial models in the estimation and analysis of nonrigid motions. We present these models in the context of two different applications, namely, shape registration and flow-field analysis. The common theme throughout this thesis is the use of polynomials as parametric models to present nonrigid motion. Within this common framework, we propose three novel methods for nonrigid motion estimation and analysis.
520 $a First, we study the relationship between image moments and incremental polynomial deformations. Since image moments are integrals of images with polynomial kernel functions, there exists an interesting interaction between image moments and polynomial deformations. Specifically, changes in image moments can be expressed as a system of linear equations. Based on this observation, we propose a new formulation of the registration problem by matching image moments instead of individual pixel values, and using the derived system of linear equations to help design robust registration methods. A comparison between the classic Lucas-Kanade and a recent moment-based method by Domokos and Kato shows that our method converges faster than the Lucas-Kanade method, and is more robust to image noise.
520 $a Next, we introduce a second type of polynomial image deformation model that blends local polynomial representations into a global one based on a partition-of-unity. The proposed model does not rely on a control-point grid, in contrast to existing polynomial-based models such as B-splines, and belongs to a group of so-called meshless models. Meshless models can be easily adapted to challenging shapes and topological changes. We apply our meshless model to register shape contours. In addition, we propose a novel consistency regularizer to penalize undesired fluctuations in the deformation field, and to avoid degenerated results. In comparison with existing regularizers for meshless models, our regularizer significantly simplifies the registration framework.
520 $a To facilitate contour registration, we extend a representation of shape contours based on distance transforms. We propose a variational chamfer-matching energy term that measure the alignment of shape contours without relying on the narrow-band function. By combining our meshless model with the proposed chamfer-matching term, we obtain a novel method for registering shape contours. Experiments show that our method achieve higher registration results, and is able to handle challenging shapes of high curvature and large deformations.
520 $a Finally, we extend our study of polynomial-based deformation models into the complex domain. We propose to model vector-flow fields using complex-valued polynomials. As complex-valued monomials form a basis for holomorphic flows, we decompose a flow field onto this basis. Then, to address modeling limitations of holomorphic flows, we extend our representation to non-holomorphic flows by including monomials of negative indices, based on Laurent's series. Complex-valued polynomials allow us to handle flow-field rotations and scaling algebraically. As a result, we propose a scale- and rotation-invariant descriptor using normalized Laurent coefficients. In addition, we design a multi-scale algorithm for detecting singular patterns of different sizes. We test the sensitivity of our method against different parameter values, and we demonstrate its application in high-level tasks such as classification and clustering of vector-flow patterns.
520 $a We believe polynomial models are powerful tools for representing nonrigid deformations, and there are still a number of directions to be explored regarding to its usage and improvements.
590 $a School code: 0473.
650 4 $a Computer Science. $3 626642
690 $a 0984
710 2 $a Florida Institute of Technology. $3 718970
773 0 $t Dissertation Abstracts International $g 72-06B.
790 1 0 $a Ribeiro, Eraldo, $e advisor
790 $a 0473
791 $a Ph.D.
792 $a 2011
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3452323