Back to search

Democritus University of Thrace (DUTH)

Ανάπτυξη και υλοποίηση νέων τεχνικών επεξεργασίας έγχρωμων εικόνων

Abstract

dc:description

Non-linear image processing is a continually growing field of research. Unlike linear filters, non-linear filters have very good performance in the presence of high frequency and impulse noise, avoiding at the same time extensive blurring of the image, since they preserve details and edges. Thus, they provide solutions in many practical applications, where linear filters are inappropriate (e.g. digital image analysis, digital TV applications, speech processing and coding, etc.). Morphological filters (e.g. opening and closing) and rank order filters (e.g. min, max and median) are of the most well-known non-linear filter classes. Morphological filters are useful in various applications, such as biomedical analysis, electronic inspection analysis, satellite scene recognition for the inspection of the climate and miscellaneous machine vision applications (feature extraction, object recognition, skeletonisation, segmentation, thinning, thickening, etc.). Soft mathematical morphology is a relatively recent extension of mathematical morphology. Its aim is to improve the behavior of morphological filters in the presence of impulse noise. The median filter is used efficiently in impulse noise attenuation, while it preserves at the same time the edges in the image under process. Color image processing is one of the most recent and interesting areas in the wider field of digital image processing. A color image provides much more information than a binary or a gray-scale image. Consequently, the use of color is desirable in digital image processing. In the last decade technological progress has propelled the development of non-linear techniques for color image processing, solving various problems, such as the prohibitively great cost of the hardware required, etc. Most of the non-linear scalar techniques for gray-scale image processing are not directly applicable to color images, due to the vector nature of color. Scalar techniques that are applied component-wise (independently to each vector component), without taking into consideration the dependency between vector components, are improper for color image processing. They may induce loss or corruption of the information in the image under process. Thus, recently proposed techniques (vector techniques), which take into consideration the vector nature of color, are preferable. Till today it has not been presented in the literature a complete and theoretically founded vector technique for morphological color image processing, with experimental results that verify its accuracy. Furthermore, the large computational complexity is the basic disadvantage, which prevents the wide spread and exploitation of various vector median filters proposed for color impulse noise attenuation. In this thesis, a new framework for vector morphological color image processing, compatible with the well-known framework of gray-scale morphology, is founded. By means of a new ordering of vectors in the HSV color space, primary and secondary vector morphological operations for color images are defined. Compatibility with their gray-scale counterparts, as well as their basic properties are studied. In addition, a number of illustrative experimental results certify that the proposed framework can be used in various classic morphological image processing applications, such as color image edge detection, color image skeletonisation, color image object recognition, etc. Moreover, a new approach to soft vector morphology for color image processing is presented. Its basic characteristic is its compatibility with the framework of soft gray-scale morphology. Primary and secondary soft vector morphological operations for color images are defined, and their basic properties are studied. Furthermore, various experimental results provided assure that this approach, analogously to soft gray-scale morphology, improves the performance of vector morphological transforms in the presence of color impulse noise. A new vector median filter applicable to color images is also introduced. Experimental results quoted indicate that this filter is advantageous, regarding color impulse noise elimination, when compared to other vector median filters. In addition, it is shown that the new filter considerably outperforms other existing vector median filters, from the point of view of computational complexity. As we have already mentioned, mathematical morphology is very useful in a variety of applications. In many of these applications time is critical. The vector nature of color increases the computational complexity of vector morphological transforms. Consequently, it increases the time required for the morphological color image processing, as well. It is known that hardware implementation is always faster than software implementation. Therefore, is of significant importance to develop hardware structures for the implementation of vector morphological operations, so as to accelerate morphological color image processing. From this point of view a new FPGA hardware structure, capable of computing primary standard vector morphological operations, is presented. It is a pipelined array with typical system clock frequency 40MHz, which processes two-dimensional 24-bit resolution color images using a 3x3 pixel size color structuring element. Its operation is based on a min/max finding algorithm. The increase of the structuring element’s size does not increase considerably the chip area required, reduces, however, the speed (the rate of result extraction) of the proposed hardware structure. Finally, a new FPGA hardware structure for the implementation of the primary soft vector morphological operations and the primary standard vector morphological operations, as well, is described. It is an FPGA parallel pipelined array, with typical system clock frequency 40MHz, as well, which processes two-dimensional 24-bit resolution color images using a 3x2 pixel size color structuring element. Its operation is based on a quick version of mergesort algorithm that is faster than classic mergesort. Compared to the previously mentioned hardware structure is faster (it has greater rate of result extraction). The increase of the structuring element’s size does not affect at all the speed (the rate of result extraction), increases, however, the chip area for the proposed hardware structure.

Degree

thesis:*
Grantor dc:publisher
Democritus University of Thrace (DUTH)
Year dc:date
2002

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Vardavoulia, Maria
  • Βαρδαβούλια, Μαρία

Subjects

dc:subject × 14

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/17527

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Vardavoulia, Maria; Βαρδαβούλια, Μαρία. Ανάπτυξη και υλοποίηση νέων τεχνικών επεξεργασίας έγχρωμων εικόνων. Democritus University of Thrace (DUTH), 2002. http://hdl.handle.net/10442/hedi/17527