Democritus University of Thrace (DUTH)
Ανάπτυξη και υλοποίηση ηλεκτρονικού συστήματος όρασης βασισμένου στις αρχές λειτουργίας του ανθρώπινου οπτικού συστήματος
Abstract
dc:descriptionBoth biological and artificial visual systems have to face the same problems οn their basis. They both have to compensate for the limited dynamic range of their sensors, compared to the immense range of intensities that are present in the natural scenes. They both have to extract and encode edges, colors, motion, texture and depth. They have to process a huge amount of visual information in real?time and carry out multiple feature classifications. All the above problems are active research fields of machine vision. Yet, they have already been solved by nature, through the process of evolution. Having already a visual system that functions almost ideally, analyzing its characteristics can shed light to many of the contemporary scientific problems. The primary objective of this dissertation is the development of new techniques in the fields of machine vision and image processing, which have been inspired by the characteristics of the Human Visual System (HVS), and generally, the visual system of the primates. The first problem, which is investigated in this dissertation, is the extraction of salient contours. The edges of an image can be a valuable tool in its analysis and processing. Nevertheless, their vast number can pose a problem to an automatic system. Moreover, biological visual systems can easily distinguish between the most important contours in a scene and create its primal sketch. In this context, a new method is presented for the extraction of salient contours, based on the non?classical receptive field of the simple cells of the primary visual cortex. As part of this method, a new set of binary kernels was designed for the implementation of the simple orientation cells. This set improves by a factor of 20 the usual approach, which is based in the convolution of the image with Gabor filters. Additionally, the proposed implementation of the oriented cells favors parallelism and is more compatible with the discrete lattice of positions and orientations of the hypercolumns, found in the primary visual cortex. A new shape descriptor, based on the proposed kernel set is also introduced. This shape descriptor exhibits scale and rotation invariance and can handle distortions in the contours of the shapes. Moreover, a new type of neural network is introduced having as inputs the results of the proposed oriented kernels, for every image region. With every iteration the network increases the values of the kernels which encode salient contours. The network incorporates the Gestalt laws of good continuity and proximity, by affiliating the orientation kernels according to their coexponentiality. This approach was found to exhibit better results than the usual co?circular approach found in the literature. A new affinity function, based in the similarity between the two lobes of the oriented neurons, is also introduced. This means that neurons with the same excitation degree in both their left and right part increase more their output compared to the other neurons. As a result, better continuity in the extracted contours is achieved. The proposed neural network has also the ability to enhance a certain degree of curvature. This can be used for filtering the extracted salient contours, in order to output only one type of contour (e.g. circles, straight lines etch.). Finally, the proposed method exhibits satisfactory results in a great variety of real and synthetic images, as well as, the fastest execution times reported until now in the literature. The present dissertation also addresses the problem of image quality degradation, caused by the limited dynamic range of the image sensor. In these cases, the dynamic range of the sensor is lower than the dynamic range of the scene, causing either under?exposed or overexposed regions to the image and, thus, reducing the visible information. In this context, a new method is proposed for the enhancement of images with under or over?exposed regions. A new contrast modification function is introduced, derived by the shunting inhibition characteristics of the center?surround cells of the HVS. ...................................
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2008
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Vonikakis, Vasileios
- Βονικάκης, Βασίλειος
Subjects
dc:subject × 10- Βιολογικά οπτικά συστήματα
- Ηλεκτρονικά οπτικά συστήματα
- Εξέχοντα περιγράμματα
- Βελτιστοποίηση αποχρώσεων
- Δυαδικοποίηση εγγράφων
- Υλοποίηση σε FPGA
- 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/20986