A diversity enhanced tree-seed algorithm based on double search with genetic and automated learning search strategies for image segmentation

Abstract

Image segmentation represents a critical yet inherently complex problem in the field of image processing, with the objective of extracting significant information from visual data. Traditional methodologies often encounter difficulties in effectively retrieving pertinent information. In contrast, swarm intelligence techniques, which optimize through collaborative interaction and stochastic exploration without dependence on prior knowledge, are more adept at addressing image segmentation challenges. The Tree-Seed Algorithm (TSA), a prominent swarm intelligence optimization technique, has been extensively utilized to tackle intricate optimization issues. Nonetheless, the reliance on a singular seed generation approach may result in inadequate exploration, premature convergence, diminished diversity, and local stagnation. To address these deficiencies, a hybrid variant known as the Tree-Seed-Gene Algorithm (TSGA) is proposed, drawing inspiration from the Genetic Algorithm (GA) and incorporating a double search strategy that integrates genetic and automated learning strategies. The genetic search contains mechanisms such as elite, crossover, and mutation. Furthermore, an opposition-based learning strategy is introduced to bolster population diversity, thereby enhancing exploration capability. The efficacy of the TSGA algorithm is assessed in comparison to both classical and contemporary meta-heuristic algorithms, including their variants, utilizing benchmark functions from the IEEE CEC 2014, 2017, 2020, and 2022. The performance of the TSGA is substantiated through statistical analyses, specifically, the Wilcoxon signed-rank and Friedman tests. The findings indicate that the TSGA algorithm exhibits superior performance in resolving image segmentation issues. In conclusion, the experimental results consistently affirm the TSGA has significant potential for practical applications in the domain of image segmentation.