US 12,154,310 B2
Technologies for automatically determining and displaying salient portions of images
Anshul Garg, Bangalore (IN); Vikramaditya Khemka, Belmont, MA (US); and Ajay Joshi, Bedford, MA (US)
Assigned to CIMPRESS SCHWEIZ GMBH, Winterthur (CH)
Filed by Anshul Garg, Bangalore (IN); Vikramaditya Khemka, Belmont, MA (US); and Ajay Joshi, Bedford, MA (US)
Filed on May 28, 2021, as Appl. No. 17/333,508.
Prior Publication US 2022/0383032 A1, Dec. 1, 2022
Int. Cl. G06V 10/46 (2022.01); G06F 18/214 (2023.01); G06T 11/60 (2006.01)
CPC G06V 10/462 (2022.01) [G06F 18/214 (2023.01); G06T 11/60 (2013.01)] 18 Claims
OG exemplary drawing
 
1. A computer-implemented method for automatically determining and displaying salient portions of images, the computer-implemented method comprising:
accessing, by a processor, a digital image comprising pixel data of at least one feature of interest;
analyzing, by the processor applying a saliency detection learning model, the digital image to generate at least one saliency map corresponding to the at least one feature of interest, the at least one saliency map selected from one or more saliency maps of the saliency detection learning model, wherein the saliency detection learning model is trained with pixel data of a plurality of training images depicting respective features of interest and is configured to output the one or more saliency maps corresponding to the respective features of interest;
determining, by the processor, one or more salient portions of the digital image at least by:
determining a salient portion image threshold based on the at least one saliency map to estimate one or more edges present in the at least one saliency map,
generating a threshold image based on the one or more edges,
detecting contours of the one or more salient portions based on the threshold image,
fitting a smallest convex polygon for each of the one or more salient portions that contains all points included as part of the contours,
determining a minimal bounding fit rectangle based on extreme values of each smallest convex polygon, and
combining each minimal bounding fit rectangle to generate a union box; and
displaying, in a user interface, a portion of the digital image that includes at least the one or more salient portions.