图像边界填充方法及医疗影像分析方法
By acquiring and stitching extended boundary regions in image processing, fusing global context features, and reconstructing boundary features, the problem of insufficient boundary continuity in image processing is solved, thus improving the quality of boundary region analysis in medical image analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTH CHINA NORMAL UNIV
- Filing Date
- 2026-05-18
- Publication Date
- 2026-07-17
AI Technical Summary
In existing image processing and convolutional neural networks, the boundary padding methods lack semantic awareness, resulting in insufficient boundary continuity and affecting the quality of boundary region analysis of images or feature maps. This is especially true in medical imaging scenarios, where organ edges are harshly transitioned, lesion boundary details are incomplete, or soft tissue textures are discontinuous.
By acquiring the initial extended canvas of the input image, the surrounding extended boundary regions are extracted and stitched together, encoded, and fused with global contextual features. An attention interaction mechanism is used to fuse global contextual information into the extended boundary region features, and boundary features are reconstructed and backfilled into the initial canvas to generate a more continuous and consistent output image.
It improves the quality of boundary representation in images or feature maps, provides more stable feature extraction, enhances the analysis quality of organ boundaries and lesion edges in medical image analysis, and solves the problem of boundary discontinuity.
Smart Images
Figure CN122265330B_ABST