Use this URL to cite or link to this record in EThOS:
Title: Novel techniques for object recognition
Author: Giannarou, Stamatia
ISNI:       0000 0001 3496 413X
Awarding Body: Imperial College London (University of London)
Current Institution: Imperial College London
Date of Award: 2007
Availability of Full Text:
Full text unavailable from EThOS.
Please contact the current institution’s library for further details.
This thesis is concerned with the design of a real time object recognition system. The ultimate goal is to create a cognitive vision system which is robust across environmental changes (causing partial occlusions and cluttered backgrounds). invariant to image transformations (translation, rotation and scaling) and insensitive to intra-category appearance variations (permit recognition of perceptually similar objects that are not mathematically identical). . Initially. a new framework which allows for the guantitative combination of a pre-selected set of edge detectors based on the correspondence between their outcomes is proposed as an essential preprocessing stage of an object detector that operates on edge maps. This is inspired from the problem that despite the enormous amount of literature on edge detection techniques, there is no single one that performs well in every possible image context. Two approaches are proposed for this purpose; the so called Receiver Operating Characteristics (ROC) analysis and Kappa Statistics. For efficient object detection, a novel technigue for the automatic identification of real world objects in complex scenes using Shape Context analysis and multi-stage clustering is introduced. The identification problem requires the comparison of assemblies of image regions with a previously stored view of a known prototype. Shape context representation and matching are employed for recovering point correspondences between the image and the prototype. A multistage type of clustering of suspicious image locations is applied in a novel fashion to enable the identification of regions of interest on the complex scene. based on a set of density and figural continuity metrics. Finally. a novel shape signature matching approach for the automatic identification of real world objects in complex scenes is also proposed. The identification process is applied on isolated objects and requires the segmentation of the image into separate objects. followed by the extraction of representative shape features and the similarity estimation of pairs of objects. In order to enable an efficient object representation, a novel boundary-based shape descriptor is introduced, formed by a set of one dimensional signals called shape signatures. Two different approaches are proposed for the shape signature comparison. In the first approach, the cross-correlation metric is used in a novel fashion to gauge the degree of similarity between objects. In the second approach, the correspondence between the signatures' extreme points is established and is used as a basis for the similarity evaluation
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available