US 11,910,080 B2
Image pickup apparatus for inferring noise and learning device
Makiko Saito, Tokyo (JP)
Assigned to CANON KABUSHIKI KAISHA, Tokyo (JP)
Filed by CANON KABUSHIKI KAISHA, Tokyo (JP)
Filed on Jan. 21, 2021, as Appl. No. 17/154,365.
Claims priority of application No. 2020-015074 (JP), filed on Jan. 31, 2020.
Prior Publication US 2021/0243395 A1, Aug. 5, 2021
Int. Cl. H04N 23/617 (2023.01); G06N 20/00 (2019.01); H04N 25/68 (2023.01); H04N 25/70 (2023.01); G06N 3/08 (2023.01); H04N 23/60 (2023.01); G06N 5/01 (2023.01); G06N 20/10 (2019.01)
CPC H04N 23/617 (2023.01) [G06N 3/08 (2013.01); G06N 20/00 (2019.01); H04N 23/64 (2023.01); H04N 25/68 (2023.01); H04N 25/70 (2023.01); G06N 5/01 (2023.01); G06N 20/10 (2019.01)] 15 Claims
OG exemplary drawing
 
1. An image pickup apparatus comprising:
a solid-state image pickup device including known pixels that emit pixel signals having noise superimposed thereon; and
at least one processor or circuit configured to function as:
a learning unit configured to:
(i) input, to a model-to-be-learned, image data (i.a) that is acquired from the solid-state image pickup device and (i.b) that has not been corrected based on first correction information identifying the pixel signals, emitted from the known pixels, having the noise superimposed thereon,
(ii) detect an error in an output of the model-to-be-learned by using the first correction information identifying the pixel signals, emitted from the known pixels, having the noise superimposed thereon as supervised data for correcting the output of the model-to-be learned, and
(iii) update the model-to-be-learned based on the error in the output of the model-to-be learned detected using the first correction information identifying the pixel signals, emitted from the known pixels, having the noise superimposed thereon to thereby generate a learned model; and
an inference unit configured to infer a pixel signal, emitted from a pixel of the solid-state image pickup device different from the known pixels, on which noise is superimposed by inputting image data, to the learned model, corrected based on the first correction information.