Scientists from the Institute for Experimental Medical Research at the University of Oslo are developing algorithms and software for removing noise from images. The technology employs neural networks, and unlike many existing de-noisers, trains on the noisy image itself. The research has already shown very positive results with computer tomography (CT) and microscopy images. Additionally, the approaches can be applied to non-medical fields, where we aim to test our system on seismic and surveillance images and video.
Inven2 seeks partners for co-development and can offer exclusive access to the software and know-how. We are interested in validating the technology applicability together with a user partner.
The technology is a neural network-based method for improving image quality via novel denoising approaches.
Our prototype extends upon the capabilities of current neural network systems by utilizing a unique paradigm that minimizes artifact generation, whilst maintaining the benefits of self-supervision. An comparison with a current state-of-the-art algorithmic method known as PURE-LET showed considerable and consistent performance improvements with an average SNR increase of 0.6 dB.