Ganugapati Seshu Srilakshmi, (1996) Edge detection methods for speckled images. Masters thesis, Memorial University of Newfoundland.
PDF (Migrated (PDF/A Conversion) from original format: (application/pdf))
- Accepted Version
Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.
Images obtained from coherent imaging systems such as laser, sonar, radar, synthetic aperture radar (SAR) and ultrasound are often corrupted by a phenomenon known as image speckle. Speckle is characterized as multiplicatively signal dependent and may be spatially highly correlated noise. It differs from other types of noise such as the additive white Gaussian noise (AWGN) most commonly found in digital images. Observed from a human or computer vision point of view speckle gives a granular patterned appearance to images, thus obscuring underlying image details. -- Gradient methods taking differences between pixel values may give inconsistent estimates regarding true edge pixels and therefore are not suited for use with speckled images. By contrast, taking ratios between the pixel values tends to factor out the multiplicative noise effect present in speckled images and to generate meaningful edge maps for these images. But these methods generate thick and ambiguous edge maps, and may also require gradient information supporting the ratio edge strength values in order to generate better edge maps on speckled images. This thesis investigates methods to improve the performance of existing speckle specific edge detection operators. A ratio edge detector based on maximum strength edge pruning (MSPRoA) which uses both edge strength magnitude and direction is proposed. -- The MSPRoA method is different from previous methods in that it uses the edge orientation information that is implicitly expressed in some other ratio based methods,explicitly, thus enabling the generation of precise and well defined edge maps for speckled images. The MSPRoA method does not require either gradient information or edge thinning operators and hence computational savings are achieved. The use of the MSPRoA at multiple scales in order to extract edge information at both micro and macro levels is also suggested. The MSPRoA and multi-scale MSPRoA methods are tested using both synthetic and real airborne SAR images of varying scene contents and business. Test results which confirm the suitability of the method for use on speckled images are presented. The use of the MSPRoA method is recommended for detecting edges images in which speckle phenomena are manifest.
|Item Type:||Thesis (Masters)|
|Additional Information:||Bibliography: leaves 151-161.|
|Department(s):||Engineering and Applied Science, Faculty of|
|Library of Congress Subject Heading:||Image processing--Digital techniques; Speckle|
Actions (login required)