MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202641124735 A) filed by Dr. S. Bhuvana; Mrs. P. Anlet Pamila Suhi; Pooja H L; Dr. M. Aarthi Elaveini; Leena R; Selvaparvathi S; V V S Sasank; Dr. G. Renuka; Dr. Vijayasamundiswary. S; and W. Stanley Karunakaran on October 18, 2026, for System And Method For Uncertainty-Guided Image Segmentation And Topology-Consistent Feature Extraction.
Inventors include Dr. S. Bhuvana; Mrs. P. Anlet Pamila Suhi; Pooja H L; Dr. M. Aarthi Elaveini; Leena R; Selvaparvathi S; V V S Sasank; Dr. G. Renuka; Dr. Vijayasamundiswary. S; and W. Stanley Karunakaran.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: [049] A system and method for automated image segmentation and feature extraction are disclosed. An image is divided into spatially registered overlapping tiles, and a trained machine-learning engine produces class-probability maps and pixel-level uncertainty estimates. Overlapping predictions are reconciled using weights determined by tile-boundary distance and local uncertainty. A candidate segmentation mask is assessed against configurable object-topology constraints. Regions exhibiting both topology inconsistency and elevated uncertainty are selectively reprocessed using expanded image context. A revised mask is accepted according to topology and confidence criteria. Calibrated object-level features are extracted from accepted masks and stored with uncertainty indicators and provenance records linking each feature to its contributing mask and processing configuration. The arrangement reduces tile-seam artifacts, limits propagation of segmentation errors into measurements, and improves reproducibility of image-derived features. Accompanied Drawing [FIGS. 1-2].
Disclaimer: Curated by HT Syndication.
