.jpg)
METHOD, COMPUTER PROGRAM, SYSTEM AND COMPUTER SYSTEM FOR ELIMINATING NOISE IN AN IMAGE SEQUENCE
WO21229129
Provided are methods, programs and systems for eliminating noise in an image sequence produced by a method for generating images with volumetric data, using an algorithm for tracing light ray paths in participating media. Each image in the sequence comprises pixels, and the path-tracing algorithm produces path samples for each pixel. These methods comprise, for each pixel of a current image in the sequence: temporarily reprojecting the pixel according to a motion vector to obtain an average accumulated colour and a predictive model of the pixel; averaging a colour of the pixel with the average accumulated colour to obtain an updated average accumulated colour; including the updated average accumulated colour in a feature vector; and determining a final colour of the pixel by updating the predictive model according to the updated feature vector and obtaining the final colour from the updated predictive model.
This technique introduces one of the first practical approximations for the noise elimination in real-time, suitable for the volumetric data visualization systems based on 3-D reconstruction algorithms simulating photorrealistic illumination effects (CVR). It allows a significantly reduction in the levels of noise in sequency of images generated by MC rendering techniques (VPT)
This technique and technology improves the quality of image at a very advanced levels, which makes it suitable for 3-D visualization, medical (diagnostic imaging), AR and VR, autonomous driving, musealistic installations...
Currently working on TRL 5 achievement with external institutions, and open to discuss collaboration opportunities.