National Technical University of Athens (NTUA)
Εικονική ανακατασκευή και ταυτοποίηση αντικειμένων στο χώρο με τεχνικές υπολογιστικής όρασης και τριδιάστατα γραφικά
Abstract
dc:descriptionThe present dissertation was funded and supported by the research program PENED concerning the development of innovative digital signal processing and image analysis methods, utilizing basic principles from the fields of computer vision and virtual reality. The main focus of this work is the digital reconstruction and restoration of ancient buildings discovered by archaeological excavations in the prehistoric settlement of Akrotiri on the island of Thira. This dissertation resulted from the co-operation with the parallel efforts of the archaeological and architectural scientific community, in order to develop analog reconstruction, utilizing design and manually metering methods. The main orientation of the present dissertation is the research and analysis of the stereo matching problem in images under arbitrary lighting or shading conditions . The reliable matching of characteristics of consecutive stereo images, the precise modeling, the calculation of picture‘s optical characteristics and the 3D reconstruction of the scene, using the recovered camera model information, are the main scientific directions of this dissertation. Classical stereo matching methods are used based on the correlation of pixel intensity. A new correlation measure is suggested, based on the local representation of the images information and utilizing the monogenic signal. The monogenic signal demodulates the local energetic, structural and geometric information of an image patch. Natural images photographed under arbitrary lighting conditions, with complex structures of archaeological interest, were used for the comparative experimental study of 3D digital reconstruction. Comparative results have been addressed between the resulted 3D reconstruction, utilizing the proposed correlation measure and the classical methods of intensity correlation. The experimental set used was a series of stereo images from Erecheio and Erodion. The proposed method of 3D registration and reconstruction leads to more reliable results concerning both the computation of the camera optic model and the depth representation of the 3D graphical model. It is also utilized a Bayesian based energy model of the stereo matching problem. Towards the depth recovery information from the experimental stereo set of images, It is suggested a new similarity function. This proposed similarity function constitutes the data energy function of the energy model and represent the requirement of stereoscopic consistency in the context of a stereo matching problem. The minimization of this energy function corresponds to a successful matching pair of pixels. This similarity function is based on the monogenic demodulation the local geometrical, structural and energetic information of an image. The comparative evaluation of the results, confirms the improvement of RMS error deviation between the computed disparity maps and the corresponding ideal (ground truth) in the case of using the proposed data energy function. In order to assess more reliable evaluation measures we utilized he Structural Similarity Index (Structural SIMilarity index, SSIM index) based on the human perception ability for structural quality of images. Finally we examine the problem multiview stereo matching problem using also the proposed data energy function to satisfy the constraint of stereoscopic consistency in the context of an energy minimization problem.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2010
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Alifragis, Matthaios
- Αληφραγκής, Ματθαίος
Subjects
dc:subject × 20- Όραση υπολογιστών
- Στερεοσκοπική αντιστοίχιση
- Τεχνικές ομοιότητας φάσης
- Μαρκοβιανά τυχαία πεδία
- Μονογενές σήμα
- Αποκοπή γράφων
- Στατιστικά μοντέλα απόφασης
- Ψηφιακή επεξεργασία εικόνων
- Computer vision
- Stereo matching
- Phase congruency methods
- Markov random fields
- Monogenic signal
- Graph cuts
- Statistical decision models
- Digital signal processing
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/23285