US 12,229,967 B1
Enhanced AI image segmentation
Oran Gilad, Herzliya Pituach (IL); Samuel Chenillo, New York, NY (US); and Oren Steinfeld, Tel Aviv (IL)
Filed by Oran Gilad, Herzliya Pituach (IL); Samuel Chenillo, New York, NY (US); and Oren Steinfeld, Tel Aviv (IL)
Filed on Aug. 29, 2024, as Appl. No. 18/818,601.
Claims priority of provisional application 63/579,583, filed on Aug. 30, 2023.
Int. Cl. G06T 7/194 (2017.01); G06V 10/762 (2022.01)
CPC G06T 7/194 (2017.01) [G06V 10/762 (2022.01); G06T 2207/20084 (2013.01); G06V 2201/07 (2022.01)] 10 Claims
OG exemplary drawing
 
6. A system for producing high quality image segmentation masks, comprising:
a neural network lightly trained to recognize foreground objects in a first sequence of video frames;
a sequence of low-quality segmentation masks produced by using said lightly trained neural network to sequence a second sequence of video frames;
one or more groups of consistently segmented pixels obtained from said sequence of low-quality segmentation masks by:
clustering segmented pixels on each of said segmentation masks into groups;
comparing groups of clustered pixels on adjacent segmentation masks; and,
selecting groups of clustered pixels that form a chain over a number of segmentation masks as said consistently segmented groups of pixels; and,
a set of augmented segmentation masks obtained by assigning a gradient alpha value to pixels neighboring said consistently segmented groups of pixels, said set of augmented segmentation masks being said desired set of high-quality image segmentation masks.