Scholarlink

<p>The phenomenon of intensity inhomogeneity frequently occurs, which poses significant difficulty for image segmentation. Although existing solutions have made some contributions to addressing this issue, they fail to achieve a trade-off between intensity inhomogeneity and segmentation efficiency. In this work, we propose a novel active contour model based on pre-bias field estimation and K-means++ clustering (PBFK) to mitigate the above-mentioned problem, improving the overall segmentation effect. The clustering centers are calculated by K-means++ within sliding windows, which describe the energy information of different regions throughout the entire image. A novel global piecewise function is introduced to estimate the bias field before the gradient flow equation is constructed, i.e., pre-bias field estimation. By synergistically combining the clustering centers and the pre-bias field, fine-grained image features with suppressed interference are obtained. PBFK achieves 1 – 3% accuracy improvement and 1.01 – 1.39<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\times \)</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> speedup over previous baselines on BSDS and COCO.</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

PBFK: Leveraging pre-bias field and K-means++ clustering based active contour model for robust image segmentation

  • 期刊论文

  • 发表日期：

    2025年11月29日

查看全文
  • Jian Su,
  • Guirong Weng,
  • Fuzheng Zhang

摘要

The phenomenon of intensity inhomogeneity frequently occurs, which poses significant difficulty for image segmentation. Although existing solutions have made some contributions to addressing this issue, they fail to achieve a trade-off between intensity inhomogeneity and segmentation efficiency. In this work, we propose a novel active contour model based on pre-bias field estimation and K-means++ clustering (PBFK) to mitigate the above-mentioned problem, improving the overall segmentation effect. The clustering centers are calculated by K-means++ within sliding windows, which describe the energy information of different regions throughout the entire image. A novel global piecewise function is introduced to estimate the bias field before the gradient flow equation is constructed, i.e., pre-bias field estimation. By synergistically combining the clustering centers and the pre-bias field, fine-grained image features with suppressed interference are obtained. PBFK achieves 1 – 3% accuracy improvement and 1.01 – 1.39 \(\times \) × speedup over previous baselines on BSDS and COCO.