US 12,079,957 B2
Modeling continuous kernels to generate an enhanced digital image from a burst of digital images
Michael Gharbi, San Francisco, CA (US); and Camille Biscarrat, Cambridge, MA (US)
Assigned to Adobe Inc., San Jose, CA (US)
Filed by Adobe Inc., San Jose, CA (US)
Filed on Jan. 24, 2022, as Appl. No. 17/582,266.
Prior Publication US 2023/0237628 A1, Jul. 27, 2023
Int. Cl. G06K 9/00 (2022.01); G06N 3/08 (2023.01); G06T 5/50 (2006.01); G06T 7/33 (2017.01)
CPC G06T 5/50 (2013.01) [G06N 3/08 (2013.01); G06T 7/337 (2017.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/20212 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A non-transitory computer readable medium comprising instructions that, when executed by at least one processor, cause a computing device to generate an enhanced digital image from a plurality of burst digital images by performing operations comprising:
generating, from the plurality of burst digital images, a latent vector representation of a kernel neighborhood of an output pixel of the enhanced digital image;
determining for a plurality of samples of the plurality of burst digital images within the kernel neighborhood of the output pixel, a plurality of continuous displacement vectors;
generating, utilizing a continuous kernel neural network, learned kernel weights for the plurality of samples from the plurality of continuous displacement vectors and the latent vector representation of the kernel neighborhood of the output pixel; and
generating the output pixel for the enhanced digital image from the plurality of burst digital images utilizing the learned kernel weights and the plurality of samples.