Aristotle University Of Thessaloniki (AUTH)
Χρήση προηγμένων μεθόδων επεξεργασίας σήματος και εικόνας στην περιοχή της πεπτικής ενδοσκοπίας με χρήση ασύρματης ενδοσκοπικής κάψουλας
Abstract
dc:descriptionThe aim of the present thesis was the development of advanced image processing techniques for the analysis of Wireless Capsule Endoscopy images in order to recognize ulcer regions and Crohn’s Disease (CD)-based lesions. The proposed methodologies lie in four main pillars, according to the mathematical tools used and the information aiming to elicit. Initially, the discrimination of lesion/normal images was based on texture features extracted from grayscale images by using Differential Lacunarity Analysis (DLA) and Bidimensional Ensemble Empirical Mode Decomposition (BEEMD) for denoising. The feature vector (FV) obtained by the BEEMD-DLA combination exhibited encouraging results and led to the utilization of texture-color features by applying BEEMD-DLA on the chromatic components of RGB space. Afterwards, a detailed study of the optimal implementation of DLA was carried out and an improved FV was proposed that contains information about the global and the local behavior of the differential lacunarity curve. Last but not least, the role of three color spaces (RGB, HSV, CIE Lab) in the performance of the recognition system was investigated. The second pillar of approaches tried to unlock the full potential of BEEMD analysis, by analyzing the distribution of lesion textural patterns in the image intrinsic components extracted by BEEMD. In this context, an adaptive reconstruction procedure was proposed that, not only removes the noisy components, but also performs image reconstruction based on the most informative intrinsic components with properties that relate to CD lesions/ulcers. The third pillar of approaches sought further exploitation of BEEMD products by studying in detail the interactions between them and the way they vary according to the type of tissue depicted. Therefore, the measure of second and higher order intrinsic correlation was introduced and a hybrid feature extraction system was presented, that was based on both, the texture features of denoised images and the higher-order intrinsic correlation data. The encouraging results of intrinsic correlation measure lead to the development of an upgraded version of the above scheme and the introduction of an improved hybrid FV that is constructed by the second/higher-order inherent texture correlation information along with the textural features of adaptively reconstructed images. At last, the approaches of the fourth pillar took advantage of Curvelet Transform (CT), a multi-scale analysis tool that decomposes an image into components of various scales/directions that bear information on the curved structures of the image. So, at first, a systematic study of the efficiency of CT products to represent the patterns of normal/abnormal mucosa was carried out and a new FV, based on the statistics of DL curve, was proposed. Thereafter, an innovative processed named Hybrid Adaptive Filtering (HAF) was developed that isolates, through a genetic algorithm, the intrinsic components of the images that mostly relate with the textural characteristics of mucosa. The robustness of the proposed system with regard to the presence of noise and the parameters of genetic algorithm and DLA was evaluated and it was evidenced that the HAF-based approach outperforms others in the literature and all the methodologies proposed in the thesis.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2015
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Charisis, Vasileios
- Χαρίσης, Βασίλειος
Subjects
dc:subject × 14- Επεξεργασία εικόνας
- Γαστρικό έλκος
- Αναγνώριση προτύπων
- Νόσος του Crohn
- Ασύρματη ενδοσκόπηση με κάψουλα
- Image processing
- Gastric ulcer
- Pattern recognition
- Crohn's disease
- Wireless capsule endoscopy
- 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/36695