US 11,748,451 B2
Machine learning techniques for differentiability scoring of digital images
Arshiya Aggarwal, Delhi (IN); Sanjeev Tagra, Redmond, WA (US); Sachin Soni, Delhi (IN); Ryan Rozich, Austin, TX (US); Prasenjit Mondal, West Bengal (IN); Jonathan Roeder, Round Rock, TX (US); and Ajay Jain, Ghaziabad (IN)
Assigned to Adobe Inc., San Jose, CA (US)
Filed by Adobe Inc., San Jose, CA (US)
Filed on Sep. 15, 2020, as Appl. No. 17/21,279.
Prior Publication US 2022/0083809 A1, Mar. 17, 2022
Int. Cl. G06K 9/62 (2022.01); G06F 18/22 (2023.01); G06N 5/04 (2023.01); G06T 5/50 (2006.01); G06V 10/40 (2022.01); G06F 18/2113 (2023.01)
CPC G06F 18/22 (2023.01) [G06F 18/2113 (2023.01); G06N 5/04 (2013.01); G06T 5/50 (2013.01); G06V 10/40 (2022.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A system for identifying a differentiated image for representing an electronic content item, the system comprising:
an image feature extraction module configured for generating an input feature vector of an input image by, at least, applying a feature extraction neural network to the input image;
a differentiability scoring module configured for:
accessing a reference image having a reference feature vector and associated user interaction metrics,
determining a distance between the input feature vector and the reference feature vector,
generating a weighting factor by calculating a proportion of the associated user interaction metrics of the reference image to a quantity of total user interaction metrics that are associated with a group of multiple reference images, wherein the reference image is included in the group of multiple reference images, and
determining a differentiability score of the input image by applying, to the distance, the weighting factor, wherein the differentiability score indicates an estimated visual difference of the input image with respect to the reference image; and
an image modification module configured for:
determining an image modification that increases the differentiability score of the input image, and
generating a recommended image by applying the image modification to the input image, wherein a modified differentiability score for the recommended image indicates an improved estimated visual difference of the recommended image with respect to the input image.