| P. Kapsalas |
| Statistical Evaluation of Image Segmentation Algorithms’ Responses Employed toDetect Corrosion Damage on Stonework |
| Master Thesis, Dep. Of Electronics & Computer Engineering, Tech. Univ. of Crete |
|
ABSTRACT
|
| Corrosion damage of industrial materials and artwork objects form an aspect of high importance nowadays and gathers the interest of many researchers from different scientific fields. The main aim of these research efforts is to extract reliable information on the extent and the types of degradation and thus to propose techniques for effective reconstruction. A challenging issue in the field of corrosion damage estimation is the development of non-destructive to the material evaluation methodologies. The current work introduces a novel approach of deterioration damage analysis based on computer vision techniques for non-destructive quantitative and qualitative evaluation of degradation effects on stonework. Thus, we have developed various segmentation approaches each of which handling in a different way the background in-homogeneities. The detection schemes, implemented in this work, aim at approaching accurately the topology of corrosion patterns while preserving their shape and size features. Thus, methods of adaptive thresholding, based on features of the local background, are initially employed while other techniques that involve Region Growing segmentation or fusion of detection results are also tested. The corrosion damage effects derived by the segmentation procedure are subsequently quantified by the means of several statistical metrics. In this thesis we are also focused towards the performance evaluation and the potential of segmentation processes in correctly detecting and localizing decay effects. A semi-automated framework for validating the algorithms’ performance is thus developed. The framework implementation includes image dataset depicting representative decay effects, ground truth overlays, and source code for extracting ground truth matrixes and performance curves. This framework guarantees reliable and objective estimation of segmentation algorithms’ performance while it allows informed experimental feedback for the design of improved segmentation schemes. Further to exploiting the robust points of each segmentation approach, this work also studies the corrosion mechanisms by investigating the way that degradation state is reflected onto the size of the segmented decay areas and their relative intensities over the background. At the final stage of this work we perform shape analysis on the segmented decay patterns. The analysis scheme is mainly based on boundary information and aims at investigating the way that cleaning state/and or exposure conditions are reflected on the segments’ shape features. Furthermore, through studying the decay patterns’ shape and in particular the existence of holes/and or nested regions within the body of the segmented areas we can track the occurrence of specific degradation mechanisms. Shape features considered in combination with size and intensity characteristics of degraded areas may aid the classification of corrosion damage. Our detection methodologies and performance analysis framework is tested on a variety of images capturing from micro- to macro-scale characteristics of corrosion damage. Thus, the current work involves an examination of the limitations and the potential of various monitoring modalities to determine corrosion damage. The experts inspect the entire detection procedure and performance evaluation and the derived results proved to be in accordance with their own judgments and with previous chemical studies on the same surfaces.
|
| 06 June , 2006 |
| P. Kapsalas, "Statistical Evaluation of Image Segmentation Algorithms’ Responses Employed toDetect Corrosion Damage on Stonework", Master Thesis, Dep. Of Electronics & Computer Engineering, Tech. Univ. of Crete |
[  PDF] [
BibTex] [
Print] [
 Back] |
