Interactive image segmentation algorithm based on adaptive kernel learning
To address the issue that most existing interactive image segmentation methods suffer from limited segmentation performance due to their susceptibility to noise interference and non-convex structure impacts in the original feature space, an adaptive kernel learning-based interactive image segmentati...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
Editorial Department of Journal on Communications
2025-07-01
|
| Series: | Tongxin xuebao |
| Subjects: | |
| Online Access: | http://www.joconline.com.cn/zh/article/doi/10.11959/j.issn.1000-436x.2025123/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | To address the issue that most existing interactive image segmentation methods suffer from limited segmentation performance due to their susceptibility to noise interference and non-convex structure impacts in the original feature space, an adaptive kernel learning-based interactive image segmentation algorithm was proposed. Firstly, an energy function was constructed by integrating spatial distance information from user annotations on the results of SLIC superpixel segmentation with the pixel neighborhood topological relationships. Then, a kernel mapping mechanism was introduced to embed raw data into a high-dimensional feature space, enhancing linear separability. Subsequently, leveraging the smoothness and positive definiteness properties of RBF kernel functions, an optimized objective function was designed. Kernel parameter <inline-formula><alternatives><math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M2"><mi>σ</mi></math><graphic specific-use="big" xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="alternativeImage/D6DC28B5-C316-4bc2-9E5B-DF29FB27CFDB-M002.jpg"><?fx-imagestate width="1.69333339" height="2.28600001"?></graphic><graphic specific-use="small" xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="alternativeImage/D6DC28B5-C316-4bc2-9E5B-DF29FB27CFDB-M002c.jpg"><?fx-imagestate width="1.69333339" height="2.28600001"?></graphic></alternatives></inline-formula> was dynamically adjusted through iterative optimization strategies. Finally, systematic experiments were conducted on BSDS500 and MSRC datasets using standard evaluation metrics, including intersection over union, variation of information, boundary display error, and rand index. Experimental results demonstrate that the proposed algorithm overperforms existing approaches across all evaluation metrics, achieving outstanding performance, and it validates the effectiveness and universality of the algorithm in handling complex scenarios. |
|---|---|
| ISSN: | 1000-436X |