US 11,880,977 B2
Interactive image matting using neural networks
Brian Lynn Price, Pleasant Grove, UT (US); Scott Cohen, Sunnyvale, CA (US); Marco Forte, Bray (IE); and Ning Xu, Milpitas, CA (US)
Assigned to Adobe Inc., San Jose, CA (US)
Filed by Adobe Inc., San Jose, CA (US)
Filed on May 6, 2021, as Appl. No. 17/313,158.
Application 17/313,158 is a continuation of application No. 16/365,213, filed on Mar. 26, 2019, granted, now 11,004,208.
Prior Publication US 2021/0256708 A1, Aug. 19, 2021
Int. Cl. G06T 7/11 (2017.01); G06T 7/136 (2017.01); G06T 7/194 (2017.01); G06T 7/90 (2017.01); G06T 3/40 (2006.01); G06N 3/088 (2023.01); G06N 3/045 (2023.01); G06N 3/02 (2006.01); G06N 20/00 (2019.01)
CPC G06T 7/11 (2017.01) [G06N 3/045 (2023.01); G06N 3/088 (2013.01); G06T 3/40 (2013.01); G06T 7/136 (2017.01); G06T 7/194 (2017.01); G06T 7/90 (2017.01); G06N 3/02 (2013.01); G06N 20/00 (2019.01); G06T 2207/10024 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/20096 (2013.01); G06T 2207/20104 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A method for image matting, the method comprising:
receiving an image comprising a plurality of image pixels, the image having a foreground region and a background region;
transforming a binary mask for the image into a set of image feature vectors using a multilayer encoder convolutional network;
transforming the image feature vectors into a trimap using a decoder deconvolutional network, the trimap designating at least some of the image pixels as at least one of background pixels, foreground pixels, or uncertain status pixels, wherein the decoder deconvolutional network is configured to generate a probability value for a particular one of the uncertain status pixels, the probability value indicating a likelihood that the particular uncertain status pixel belongs to the foreground or the background region of the image; and
generating, by a deep neural network (DNN), an alpha matte based on the trimap and the probability value.