Multi-population estimation of distribution algorithm for multilevel thresholding in image segmentation

Optimization algorithms are crucial for solving complex optimization problems in fields such as image processing. Image segmentation is complicated and requires accurately identifying and outlining regions within an image. Metaheuristic algorithms (MAs) are highly versatile solutions to this challen...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Neurocomputing (Amsterdam) Jg. 641; S. 130325
Hauptverfasser: Ramos-Frutos, Jorge, Oliva, Diego, Miguel-Andrés, Israel, Casas-Ordaz, Angel, Ramos-Soto, Oscar, Aranguren, Itzel, Zapotecas-Martínez, Saúl
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier B.V 07.08.2025
Schlagworte:
Footwear industry
Image segmentation
Multi-population-based algorithms
Multilevel thresholding
Footwear industry
Multilevel thresholding
Image segmentation
Multi-population-based algorithms
ISSN:0925-2312
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Optimization algorithms are crucial for solving complex optimization problems in fields such as image processing. Image segmentation is complicated and requires accurately identifying and outlining regions within an image. Metaheuristic algorithms (MAs) are highly versatile solutions to this challenge. This work presents the Multi-population Estimation of Distribution Algorithm (M-EDA), which estimates distributions with multiple populations. The algorithm is initialized with low discrepancy sequences and random numbers. Two sampling methods are proposed: the first uses the normal distribution with 70% of individuals generated, while the second uses the Cauchy distribution with 30% of individuals. The M-EDA is used for multilevel thresholding, achieving good results with the Berkeley Segmentation Dataset 500 (BSDS500). Additionally, a procedure is developed to detect defects in footwear images, where the M-EDA helps to perform the detection. The study shows that the M-EDA is the best algorithm for optimizing the Minimum Cross-Entropy (MCE), winning at the four thresholding levels studied. Six statistics are used to map the algorithms’ results to different image histogram structures (minimum, first quartile, mean, median, third quartile, and maximum). Furthermore, the results are analyzed using Friedman’s non-parametric statistical test, supporting the significant difference between the statistical results obtained. The same study is applied to eleven images from the BSD500, which succeeds in all cases. Finally, the results obtained in footwear images with scratch defects are visually and numerically competitive because they segment the regions where the defects are found. The source code of the M-EDA can be found at https://github.com/JARF1095/M-EDA. •An estimation of distribution algorithm with multiple populations.•Multilevel thresholding image segmentation solved with an EDA.•BSDS500 dataset completely analyzed.•Footwear defect detection using multilevel thresholding.•Use of two symmetric distributions in sampling step.
ISSN:0925-2312
DOI:10.1016/j.neucom.2025.130325