A study of image registration for 2D and 3D applications
Date of Issue2014
School of Electrical and Electronic Engineering
A fundamental task in many image processing applications is image registration. Image registration can be defined as the process of aligning or matching two images such that the two images can be transformed into one coordinate system. The images can be from different viewpoints, different time and also from different sensors, sensing different regions across the electromagnetic spectrum. Image registration algorithms defined in literature have been very specific to an application and hence most surveys conducted revolve around a particular application. The need of the hour is an image registration technique which is universal, accurate, robust and yet be computationally efficient. This dissertation accomplishes a survey of existing image registration techniques for both two dimensional and three dimensional image processing applications and presents a qualitative and quantitative analysis on a range of test data to determine the generalization aspects of these registration techniques. The dissertation also explores an important class of applications requiring the three dimensional image registration process known as the 3D volume reconstruction. The dissertation proposes a technique to compute the real-world volume of an object from its three dimensional model. This has far ranging applications from healthcare to interactive gaming. For example in a hospital, when the patient is immobile and his/her weight has to be recorded and traditional techniques may be intrusive. Therefore, a non-intrusive approach is highly desirable and this dissertation proposes an approach for the volume generation of an object from its 3D model. The dissertation also documents the important challenges which have to be overcome to obtain a robust and accurate volume estimate of the 3D model.
DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Signal processing