Optimal morphological probing of random shape and texture /
The adequate characterization of random shape and texture is an important issue when it comes to any application based on vision. Our ability to recognize different shapes with ease thanks to our sophisticated visual system and brain makes us undermine the difficulties involved in tackling the prob...
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
1998.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=737660861&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | The adequate characterization of random shape and texture is an important issue when it comes to any application based on vision. Our ability to recognize different shapes with ease thanks to our sophisticated visual system and brain makes us undermine the difficulties involved in tackling the problem from a mathematical point of view. The visual field is often modeled as a multidimensional random process and as such the problem of pattern recognition is best posed in the same setting. Several stochastic models have been implemented to simulate these multidimensional random processes but even for the simplest cases our ability to estimate process parameters from realizations was minimal. This thesis tries to adress the problem of designing probes that optimally extract information about a spatial process whether it is a random shape or texture. In the framework of mathematical morphology this translates in optimizing the shape of the structuring elements to minimize an associated error measure. In the case of random parameterized shapes, the problem is posed as a statistical inverse problem where the related conditional densities are only defined as distributions constrained to the solution manifold. The expected error due to a particular choice of structuring elements involves surface integrations on the corresponding solution manifold. The structuring element that yields an estimate of process parameters with minimal associated error is picked as optimal. The approach offers an estimation method consistent with the constraints of the probing unlike many common estimation techniques like the conditional expectation. In the case of stochastic texture processes, morphological granulometric moments have been successfully used as sensitive texture probes capable of detecting slight changes in the spatial processes under investigation. To adress the problem of structuring element optimization, one needs to establish the true ability of a given set of structuring elements in differentiating a predetermined set of textures from each other. This is achieved by generalizing the granulometric procedure to a multivariate setting where each structuring element can be scaled independently of each other. This offers a platform to extract the true potential of a set of structuring elements independent of their relative sizes in the basis set and making performance comparisons with another set possible. |
|---|---|
| Item Description: | Vita. "Major Subject: Electrical Engineering". |
| Physical Description: | xiii, 204 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilms Inc. |
| Bibliography: | Includes bibliographical references: pages 148-161. |