A crossover-based optimization algorithm for multilevel image segmentation

Image segmentation plays a pivotal role in numerous image processing applications, with Kapur- and Otsu-based methods frequently applied for bi-level thresholding. However, optimizing multi-level thresholds can be computationally intensive. To address this challenge, we introduce AEFA-CS, a crossover-enhanced optimizer that augments the search capabilities of the artificial electric field algorithm. AEFA-CS incorporates two crossovers into the original algorithm and introduces an adjustment mechanism to balance their contributions. Furthermore, a novel velocity update procedure is proposed to improve the convergence rate. To gauge its effectiveness, we initially applied AEFA-CS to solve a variety of benchmark problems from the CEC 2014 test suite at 30 dimensions, conducting comprehensive analyses of the outcomes using different perspectives including statistical and time complexity analyses. Subsequently, we employ AEFA-CS to tackle a real-world task involving the Berkeley segmentation dataset (BSDS300) with Kapur’s entropy function. Our experiments affirm the superior performance of AEFA-CS, as it consistently outperforms alternative optimization methods across various threshold levels, underscoring its potent optimization capabilities.