MUMBAI, India, Oct. 5 -- Intellectual Property India has published a patent application (202611100955 A) filed by Indian Institute Of Information Technology Allahabad on August 20, 2026, for An Image-Processing System For Spatial-Grid Distributional Evaluation And Selective Refinement In Diffusion-Based Free-Form Image Inpainting.
Inventors include Devendra Izardar; Jagriti Singh; and Prof. Shekhar Verma.
The application for the patent was published on October 02, 2026, under issue no. 40/2026.
Abstract: The present disclosure proposes an image-processing system (100) and method for diffusion-based free-form image inpainting. A context modelling engine (110) fits a multi-component Gaussian Mixture Model to known background pixels and initializes a masked region using sampled pixel values. A conditional denoising engine (112) forward-noises the resulting context-aware seed image and performs reverse denoising with timestep-matched re-insertion of the known background pixels. At an intermediate checkpoint, a spatial-grid evaluation module (114) converts an intermediate predicted clean image into a single-channel representation, partitions the representation into spatial grid cells, and calculates a distributional cross-entropy score for masked pixels in each applicable cell relative to a reference background histogram. A worst-cell gating controller (116) uses the maximum cell score to trigger one or more refinement attempts and select an intermediate state for completion of reverse denoising, thereby generating a final inpainted image without updating parameters of the pre-trained diffusion model.
Disclaimer: Curated by HT Syndication.
