Multi-Level Thresholding Based on Composite Local Contour Shannon Entropy Under Multiscale Multiplication Transform.

Abstract

Image segmentation is a crucial step in image processing and analysis, with multi-level thresholding being one of the important techniques for image segmentation. Existing approaches predominantly rely on metaheuristic optimization algorithms, which frequently encounter local optima stagnation and require extensive parameter tuning, thereby degrading segmentation accuracy and computational efficiency. This paper proposes a Shannon entropy-based multi-level thresholding method that utilizes composite contours. The method selects appropriate multiscale multiplication images by maximizing the Shannon entropy difference and constructs a new Shannon entropy objective function by dynamically combining contour images. Ultimately, it automatically determines multiple thresholds by integrating local contour Shannon entropy. Experimental results on synthetic images and real-world images with complex backgrounds, low contrast, blurred boundaries, and unbalanced sizes demonstrate that the proposed method outperforms six recently proposed multi-level thresholding methods based on the Matthew's correlation coefficient, indicating stronger adaptability and robustness for segmentation without requiring complex parameter tuning.