US 11,914,951 B2
Semantically-guided template generation from image content
Vinay Aggarwal, New Delhi (IN); Vishwa Vinay, Bangalore (IN); Rizurekh Saha, Kolkata (IN); Prabhat Mahapatra, Ghaziabad (IN); Niyati Himanshu Chhaya, Hyderabad (IN); Harshit Agrawal, Navi Mumbai (IN); Chloe McConnell, Berkeley, CA (US); Bhanu Prakash Reddy Guda, Ongole (IN); and Balaji Vasan Srinivasan, Bangalore (IN)
Assigned to Adobe Inc., San Jose, CA (US)
Filed by Adobe Inc., San Jose, CA (US)
Filed on Feb. 16, 2023, as Appl. No. 18/170,125.
Application 18/170,125 is a continuation of application No. 17/450,250, filed on Oct. 7, 2021, granted, now 11,610,054.
Prior Publication US 2023/0196008 A1, Jun. 22, 2023
This patent is subject to a terminal disclaimer.
Int. Cl. G06F 40/186 (2020.01); G06F 40/109 (2020.01); G06F 40/30 (2020.01); G06N 20/00 (2019.01); G06F 18/2411 (2023.01)
CPC G06F 40/186 (2020.01) [G06F 18/2411 (2023.01); G06F 40/109 (2020.01); G06F 40/30 (2020.01); G06N 20/00 (2019.01)] 18 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
obtaining a user-selected input image including contents of different content types;
extracting characteristics associated with the user-selected input image using one or more machine learning models, wherein the characteristics comprise: 1) layout information indicating a position and a content type of each of the contents within the user-selected input image; and 2) text attributes indicating at least a font of text of each textual element included in the user-selected input image, wherein the text attributes are identified by processing a plurality of textual elements identified in the layout information concurrently, and wherein extraction of the text attributes comprises cropping out regions of textual elements and superimposing each region of textual elements on a plain background; and
generating a template including editable regions at positions corresponding to the contents of the user-selected input image, the editable regions including editable text regions that are each tagged with the text attributes of the corresponding textual element of the user-selected input image, wherein the template provides a baseline to generate a final image that emulates the characteristics associated with the user-selected input image.