Democritus University of Thrace (DUTH)
Ανάπτυξη νέων τεχνικών επεξεργασίας και εξαγωγής χαρακτηριστικών σε πολυτονικές εικόνες
Abstract
dc:descriptionThis PhD dissertation is constituted from two major parts. The first part is related with a modification of the geometrical moments so that they can be used as image features contributing to higher recognition rates, while the second part with the introduction of a new kind of non linear filters. Both parts have been recognized by the international scientific community with a significant number of publications and citations. More specifically. Improvement of the performance of the geometrical moments, as recognition features, and introduction of a new similarity measure.At this part a modification of the geometrical moments is proposed in order to improve their performance as recognition features. This is achieved by adding the variance σ of the pixel values into the function that calculates the geometrical moments. The performance of the proposed features and that of a newly introduced weighted Euclidian measure are tested on an OCR application.It is turning out that the proposed features and distance measure contribute to a 5,4% improvement in comparison to the conventional geometrical moments when the data are non noisy. The improvement further increased reaching 9,5% when the patterns to be recognized are noisy. . Introduction of a new kind of non linear filters.In this part the theory of Coordinate Logic Filters (CLF) and their applications in image processing are introduced. The proposed filters are based on the execution of logic operations among the binary representations of the pixel values (bitwise). The main idea of CLF is based on three major steps1.binary decomposition of the image2.processing of each binary level using simple Boolean functions 3.composition of the processed levels to a new word value CLF have a wide range of applications in image processing. The main features of CLF are 1.simplicity 2.speed 3.inherent parallelism4.grate performanceThe above features make CLF very appropriate for H/W implementation. According to their definition CLF present similar functionality with stack filters and consequently with the morphological filters (MF), presenting however distinct results due to their definition and the resulting properties. CLF are non increasing functions but satisfy the properties of duality, idempotence and extensivity. It is also proved that CLF can act exactly as their corresponding morphological ones when the image grey levels are quantized and mapped to a specific set of decimal numbers having unary binary representation. This way, special forms of morphological filters can be implemented using CLF, exploiting thus the inherent circuitry simplicity and parallelism of CLF. CLF are effective to many image processing tasks such as edge detection, feature extraction, noise cleaning, coding, zooming, contour following, image cryptography, cellular automata and fractal modeling. In this dissertation, the methodology for designing of CLF for the execution of the above image processing tasks is presented. Moreover the advantage in speed and simplicity, of the CLF against other classical non linear filters, is shown using FPGA implementations. Finally, the features and filters introduced in this dissertation are joined with well known image processing and pattern recognition methodologies to form a paradigm of an image retrieval platform.
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2013
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Τσιρικόλιας, Κωνσταντίνος
- Tsirikolias, Konstantinos
Subjects
dc:subject × 19- Επεξεργασία εικόνας
- Ροπές
- Στατιστικά χαρακτηριστικά
- Μορφολογικά φίλτρα
- Φίλτρα συντεταγμένης λογικής
- Χαρακτηριστικά αναγνώρισης
- Αναγνώριση χαρακτήρων
- Image processing
- Moments
- Statistical fractures
- Morphological features
- Coordinate logic filters
- Feature extraction
- OCR
- Bitwise operators
- 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/30498