Scholarlink

<p>The domain of image processing is experiencing significant advancements, primarily aimed at refining and enhancing images obtained from various sources, such as medical imaging and satellite imagery, where image segmentation plays a crucial role. This process involves partitioning pixels into distinct segments based on their intensity levels, guided by predefined thresholds. Multilevel thresholding has proven effective in refining segmentation results; however, increasing the number of thresholds can exponentially raise computational complexity, potentially compromising the accuracy of the results. Therefore, identifying the optimal number of thresholds is crucial for effective multilevel thresholding. Traditional methods for this optimization are often costly and susceptible to inaccuracies. While metaheuristic algorithms can enhance outcomes, they may introduce inconsistencies and require significant computational resources. Likewise, improper parameter tuning of metaheuristic algorithms can further degrade performance, complicating real-time applications and leading to suboptimal results. Hence, this paper proposes a novel algorithm for multilevel image segmentation that combines heuristic approaches with cosine similarity measures. The algorithm determines optimal thresholds by encoding random samples from the image histogram and evaluating them using cosine similarity to identify the most suitable candidate solutions. To assess the efficacy of the proposed algorithm, tests were conducted using a dataset of fifteen standard grayscale benchmark images. The algorithm was evaluated across thresholds of 2, 3, 4, and 5, employing performance metrics such as Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), and Friedman ranking tests. Furthermore, the proposed algorithm was compared against eight metaheuristic algorithms utilizing Otsu, Kapoor, and fuzzy entropy methods as their respective objective functions. The comparative analysis indicates that the proposed algorithm outperforms existing methods in terms of accuracy, convergence speed, robustness to noise, and overall efficiency.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Fusion of heuristics and cosine similarity measures: introducing HCSTA for image segmentation via multilevel thresholding

  • 期刊论文

  • 发表日期：

    2025年1月17日

查看全文
  • Shivankur Thapliyal,
  • Narender Kumar

摘要

The domain of image processing is experiencing significant advancements, primarily aimed at refining and enhancing images obtained from various sources, such as medical imaging and satellite imagery, where image segmentation plays a crucial role. This process involves partitioning pixels into distinct segments based on their intensity levels, guided by predefined thresholds. Multilevel thresholding has proven effective in refining segmentation results; however, increasing the number of thresholds can exponentially raise computational complexity, potentially compromising the accuracy of the results. Therefore, identifying the optimal number of thresholds is crucial for effective multilevel thresholding. Traditional methods for this optimization are often costly and susceptible to inaccuracies. While metaheuristic algorithms can enhance outcomes, they may introduce inconsistencies and require significant computational resources. Likewise, improper parameter tuning of metaheuristic algorithms can further degrade performance, complicating real-time applications and leading to suboptimal results. Hence, this paper proposes a novel algorithm for multilevel image segmentation that combines heuristic approaches with cosine similarity measures. The algorithm determines optimal thresholds by encoding random samples from the image histogram and evaluating them using cosine similarity to identify the most suitable candidate solutions. To assess the efficacy of the proposed algorithm, tests were conducted using a dataset of fifteen standard grayscale benchmark images. The algorithm was evaluated across thresholds of 2, 3, 4, and 5, employing performance metrics such as Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), and Friedman ranking tests. Furthermore, the proposed algorithm was compared against eight metaheuristic algorithms utilizing Otsu, Kapoor, and fuzzy entropy methods as their respective objective functions. The comparative analysis indicates that the proposed algorithm outperforms existing methods in terms of accuracy, convergence speed, robustness to noise, and overall efficiency.