一种胶囊内镜视频的遮挡处理方法、电子设备及存储介质
By generating occlusion masks using optical flow features and video segmentation networks, and then recovering occluded pixels using encoding/decoding and recurrent neural networks, the occlusion problem in capsule endoscopy videos is solved, thereby improving the accuracy of image cleaning and AI diagnosis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HYGEA MEDICAL TECH CO LTD
- Filing Date
- 2025-08-12
- Publication Date
- 2026-07-17
AI Technical Summary
In capsule endoscopy videos, artifacts caused by occlusion due to foam, liquid and mucus adhering to the lens surface, fluctuations in lighting conditions, and excessively high endoscope movement speed frequently occur, affecting the quality of doctors' image reading and the accuracy of AI-assisted diagnosis.
By extracting optical flow features between image frames, a pre-trained optical flow network and video segmentation network are used to generate an occlusion mask. The occluded pixels are then restored by combining an encoding/decoding network and a recurrent neural network, thereby utilizing redundant information from multiple frames and gradually restoring the occluded area.
It effectively removes non-rigid, variable, and unstructured obstructed areas in capsule endoscopy videos, improving image cleaning results, enhancing the quality of doctors' image interpretation, and increasing the accuracy of AI-assisted diagnosis.
Smart Images
Figure CN121235950B_ABST