Hybrid Artificial Root Foraging Optimizer Based Multilevel Threshold for Image Segmentation

Yang Liu, Junfei Liu, Liwei Tian, Lianbo MaComput. Intell. Neurosci 2016 · OpenReview

TL;DR

This paper introduces HARFO, a novel optimization algorithm inspired by plant root growth, designed for multilevel image segmentation. By combining root-to-root communication and a hierarchical coevolutionary structure, HARFO effectively balances exploration and exploitation, achieving superior segmentation accuracy and efficiency compared to existing methods.

What Problem Does This Paper Address?

Image segmentation is a crucial preprocessing step for image recognition, aiming to partition an image into meaningful regions. Multilevel thresholding is a popular technique for this, but determining the optimal thresholds can be computationally challenging, especially as the number of thresholds increases. Traditional methods like Otsu's criterion become exponentially complex. While metaheuristics inspired by animal foraging have been applied, this paper highlights that plant-based strategies, with their unique adaptive growth mechanisms, have been underutilized. The authors aim to bridge this gap by developing a sophisticated plant-inspired optimization algorithm for this complex problem.

Methods

The core of this paper is the Hybrid Artificial Root Foraging Optimizer (HARFO). It builds upon the Artificial Root Foraging Optimizer (ARFO) by incorporating two key enhancements:

  • Root-to-Root Communication: Individuals (representing potential solutions) exchange information through various network topologies (like Von Neumann and Star). This allows for more efficient information sharing and exploration of the search space.
  • Hierarchical Coevolution Mechanism: The population is structured into multiple levels (two in this case). Level 1 handles the primary optimization with different topologies, while Level 2 uses a global best individual from Level 1 to guide its search. This hierarchical structure, inspired by plant root systems and ecosystems, helps maintain population diversity and prevents premature convergence.

The algorithm simulates plant root growth through:

  • Auxin Concentration: A key factor regulating root growth and information exchange, analogous to fitness.
  • Growth Strategies:
    • Regrowing: Elongating a root tip.
  • Branching: Creating new root tips (solutions).
  • Shrinkage (Dead Roots): Eliminating less fit individuals.
  • Tropisms: Roots are guided towards optimal positions (solutions) influenced by environmental stimuli (nutrient distribution, analogous to fitness).

Datasets

The paper evaluates HARFO on two sets of benchmark functions:

  • Basic Benchmarks and CEC 2005 Test Beds: Functions f₁-f₁₀ are used for initial performance comparison.
  • CEC 2014 Test Bed: Functions f₁₁-f₂₀, which are more complex due to rotation and shifting, are used to assess scalability and robustness.

For the real-world application, the authors use a set of standard tested images commonly employed in image segmentation research, including:

  • Avion.ppm
  • House.ppm
  • Lena.ppm
  • Peppers.ppm
  • Safari04.ppm
  • Hunter.pgm

These images are used to evaluate the performance of HARFO for multilevel threshold image segmentation.

Key Insights and Ideas

  • Plant-Inspired Optimization is Underexplored: The paper argues that the adaptive and sensing abilities of plants offer rich inspiration for novel optimization algorithms, a domain largely dominated by animal-inspired approaches.
  • Hierarchical Coevolution Enhances Robustness: Structuring the optimization process into multiple levels with different communication topologies (Level 1: Von Neumann, Level 2: Star) effectively balances local exploration with global guidance, preventing the algorithm from getting stuck in local optima.
  • Root-to-Root Communication is Key: The ability of "roots" (solutions) to share information through defined topologies significantly improves the exploration capabilities of the algorithm.
  • Auxin as a Fitness Indicator: The concept of auxin concentration effectively mimics how plants respond to nutrient availability, providing a natural mechanism for guiding growth and selecting fitter individuals.
  • HARFO's Adaptability: The algorithm demonstrates strong performance across various benchmark functions and scales well to higher dimensions, suggesting its potential for complex real-world problems.

Key Results

  • Benchmark Performance: HARFO consistently outperformed or matched state-of-the-art algorithms like PSO, ABC, ARFO, and CCGA on a variety of benchmark functions, particularly in higher dimensions. For instance, on 30-dimensional CEC 2014 benchmarks, HARFO showed significant improvement over other algorithms, especially on complex shifted and rotated functions.
  • Image Segmentation Accuracy: When applied to multilevel threshold image segmentation, HARFO achieved competitive or superior objective values (fitness) compared to other methods, including IDPSO, ABC, and CCGA, especially for higher numbers of thresholds (M-1 = 5, 7, 9).
  • Computational Efficiency: HARFO demonstrated better efficiency, consuming less CPU time than its counterparts while achieving comparable or better results in image segmentation. This is crucial for real-time applications.

Limitations and Future Work

  • Parameter Sensitivity: The paper includes sections on parameter sensitivity analysis, suggesting that while HARFO performs well, careful tuning of parameters like population number (M), branching threshold (T_Branch), and death threshold (T_Nmority) is important for optimal performance. The authors identify M=8, T_Branch=10, and T_Nmority=5 as good general settings.
  • Future Work: The authors express interest in further developing the HARFO framework and exploring its application in other complex optimization problems. They also mention focusing on perfecting the optimization framework from the perspective of relevance theory.

Summary updated

Abstract

This paper proposes a new plant-inspired optimization algorithm for multilevel threshold image segmentation, namely, hybrid artificial root foraging optimizer (HARFO), which essentially mimics the iterative root foraging behaviors. In this algorithm the new growth operators of branching, regrowing, and shrinkage are initially designed to optimize continuous space search by combining root-to-root communication and coevolution mechanism. With the auxin-regulated scheme, various root growth operators are guided systematically. With root-to-root communication, individuals exchange information in different efficient topologies, which essentially improve the exploration ability. With coevolution mechanism, the hierarchical spatial population driven by evolutionary pressure of multiple subpopulations is structured, which ensure that the diversity of root population is well maintained. The comparative results on a suit of benchmarks show the superiority of the proposed algorithm. Finally, the proposed HARFO algorithm is applied to handle the complex image segmentation problem based on multilevel threshold. Computational results of this approach on a set of tested images show the outperformance of the proposed algorithm in terms of optimization accuracy computation efficiency.
BibTeX
@inproceedings{liu2016hybrid,
  author = {Yang Liu and Junfei Liu and Liwei Tian and Lianbo Ma},
  title = {{Hybrid Artificial Root Foraging Optimizer Based Multilevel Threshold for Image Segmentation}},
  booktitle = {Comput. Intell. Neurosci},
  year = {2016},
  url = {https://openreview.net/forum?id=fG22IMEummG},
}
Figure