US 11,928,150 B1
Image-based item identification
Justin-Josef Angel, San Francisco, CA (US); Milena Sadee, San Francisco, CA (US); Cressida Magdalen Harding, Cambridge (GB); Eric Alan Breitbard, Oakland, CA (US); Animesh Mishra, Cambridge (GB); Vlastimil Pis, Cambridge (GB); and Robert Steven Murdock, Fairfax, CA (US)
Assigned to Amazon Technologies, Inc., Seattle, WA (US)
Filed by Amazon Technologies, Inc., Seattle, WA (US)
Filed on Mar. 18, 2022, as Appl. No. 17/698,869.
Application 17/698,869 is a continuation of application No. 16/746,472, filed on Jan. 17, 2020, granted, now 11,281,713.
Application 16/746,472 is a continuation of application No. 15/670,225, filed on Aug. 7, 2017, granted, now 10,540,390, issued on Jan. 21, 2020.
Int. Cl. G06V 10/94 (2022.01); G06F 16/432 (2019.01); G06F 16/583 (2019.01); G06V 10/40 (2022.01); G06V 10/75 (2022.01)
CPC G06F 16/5838 (2019.01) [G06F 16/434 (2019.01); G06F 16/5854 (2019.01); G06V 10/40 (2022.01); G06V 10/751 (2022.01); G06V 10/95 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A method, comprising:
receiving image data;
sending, prior to identifying an item associated with the image data, the image data to a first image processing system;
receiving, from the first image processing system associated with a first category of items identified by a first parameter sent by the first image processing system, first data representing a first identity of the item associated with the image data, the first identity associated with a first confidence score;
sending, prior to identifying the item, the image data to a second image processing system, the second image processing system differing from the first image processing system;
receiving, from the second image processing system associated with a second category of items identified by a second parameter sent by the second image processing system, second data representing a second identity of the item associated with the image data, the second identity associated with a second confidence score;
determining that the image data is associated with the first category of items;
weighting, based at least in part on the image data being associated with the first category of items, the first confidence score such that a weighted confidence score is generated; and
identifying the item utilizing the first data instead of the second data based at least in part on the weighted confidence score indicating greater confidence than the second confidence score.