[Beau23HS]

Hierarchical Image Segmentation by Stepwise Optimization: New edition of 1984 Thesis

Author:Beaulieu Jean-Marie

Book:Beaulieu Jean-Marie, Ed.

Quebec (Canada)

2023, p. 145

Publisher:Jean-Marie Beaulieu

ISBN:978-1-7388812-0-8

URL:https://BeaulieuJM.ca/pupli/Beau23HS

Abstract: The survey of image segmentation considers four different approaches: pixel classification, pixel linking and region growing, hierarchical segmentation, and segmentation optimization. A new Hierarchical Stepwise Optimization (HSO) algorithm is proposed, which combines these last two approaches. The algorithm employs a sequence of optimization processes to produce a hierarchical segmentation. Starting with an initial image partition, two segments are then merged at each iteration by using an optimization process to select the segment pair that minimizes a “stepwise criterion.” The algorithm is then employed for piecewise image approximation where the stepwise criterion is derived from the global criterion, the overall approximation error. The stepwise criterion is then related to statistical hypothesis testing, and it is shown how the probability of error can be minimized in a stepwise fashion. It is also shown experimentally how convenient stopping points in the hierarchy can be found from the criterion values. Different criteria are tested on Landsat and SAR imagery.

Hierarchical Image Segmentation by Stepwise Optimization: New edition of 1984 Thesis,

Beaulieu Jean-Marie,

Beaulieu Jean-Marie, Ed., Quebec (Canada), Jean-Marie Beaulieu, 2023.

[Bibtex]

```
@book{Beau23HS,
author = {Beaulieu, Jean-Marie},
title = {Hierarchical Image Segmentation by Stepwise Optimization: New edition of 1984 Thesis},
editor = {Jean-Marie Beaulieu},
url = {https://BeaulieuJM.ca/pupli/Beau23HS},
isbn = {978-1-7388812-0-8},
doi = {},
pages = {145},
publisher = {Jean-Marie Beaulieu},
address = {Quebec (Canada)},
year = {2023},
abstract = {The survey of image segmentation considers four different approaches: pixel classification, pixel linking and region growing, hierarchical segmentation, and segmentation optimization. A new Hierarchical Stepwise Optimization (HSO) algorithm is proposed, which combines these last two approaches. The algorithm employs a sequence of optimization processes to produce a hierarchical segmentation. Starting with an initial image partition, two segments are then merged at each iteration by using an optimization process to select the segment pair that minimizes a “stepwise criterion.” The algorithm is then employed for piecewise image approximation where the stepwise criterion is derived from the global criterion, the overall approximation error. The stepwise criterion is then related to statistical hypothesis testing, and it is shown how the probability of error can be minimized in a stepwise fashion. It is also shown experimentally how convenient stopping points in the hierarchy can be found from the criterion values. Different criteria are tested on Landsat and SAR imagery.},
mypdf = {8},
keywords = {Hierarchical segmentation; Similarity measures; Clustering}
}
```

DOWNLOAD the PDF file

SEE ALSO [Bea1984] 1984 Thesis Original