RPI ID: 2021-034-401
Innovation Summary: Noise2Sim is a self-supervised image denoising method that leverages similarity-based learning without requiring clean reference images. The algorithm identifies consistent patterns across noisy inputs and uses them to reconstruct cleaner outputs. It is particularly effective in medical and scientific imaging where ground truth is unavailable. The method supports integration with deep learning frameworks.
Challenges / Opportunities: Obtaining clean training data for image denoising is often impractical. This invention bypasses that need by using internal image statistics and similarity cues. It enables robust denoising in low-data environments and supports real-time processing. The approach is adaptable to various imaging modalities.
Key Benefits / Advantages: ✔ Self-supervised learning ✔ No clean reference required ✔ Effective in noisy environments ✔ Compatible with deep learning ✔ Real-time denoising
Applications: • Medical imaging • Scientific visualization • Low-light photography
Keywords: #imagedenoising #selflearning #similaritylearning #medicalimaging #computervision #deeplearning
Intellectual Property: US Application 18/035571 US20230394631A1 filed 05-May-2023