期刊
INTERNATIONAL JOURNAL OF IMAGE AND GRAPHICS
卷 23, 期 2, 页码 -出版社
WORLD SCIENTIFIC PUBL CO PTE LTD
DOI: 10.1142/S0219467823500213
关键词
Image processing; multilevel thresholding; optimization; error minimization
This paper proposes a multilevel optimal threshold selection method using opposition equilibrium optimizer. The method minimizes the segmentation error function and is independent of the spatial distribution of gray values, resulting in improved threshold selection. Experimental results showed that the proposed method performs well qualitatively and quantitatively, making it valuable for biomedical image segmentation.
Image segmentation is imperative for image processing applications. Thresholding technique is the easiest way of partitioning an image into different regions. Mostly, entropy-based threshold selection methods are used for multilevel thresholding. However, these methods suffer from their dependencies on spatial distribution of gray values. To solve this issue, a novel segmentation error minimization (SEM)-based method for multilevel optimal threshold selection using opposition equilibrium optimizer (OEO) is suggested. In this contribution, a new segmentation score (SS) (objective function) is derived while minimizing the segmentation error function. Our proposal is explicitly free from gray level spatial distribution of an image. Optimal threshold values are achieved by maximizing the SS (fitness value) using OEO. The key to success is the maximization of score among classes, ensuring the sharpening of the shred boundary between classes, leading to an improved threshold selection method. It is empirically demonstrated how the optimal threshold selection is made. Experimental results are presented using standard test images. Standard measures like PSNR, SSIM and FSIM are used for validation The results are compared with state-of-the-art entropy-based technique. Our method performs well both qualitatively and quantitatively. The suggested technique would be useful for biomedical image segmentation.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据