US 12,236,610 B2
Generating improved alpha mattes for digital images based on pixel classification probabilities across alpha-range classifications
Brian Price, San Jose, CA (US); Yutong Dai, Adelaide (AU); and He Zhang, San Jose, CA (US)
Assigned to Adobe Inc., San Jose, CA (US)
Filed by Adobe Inc., San Jose, CA (US)
Filed on Oct. 13, 2021, as Appl. No. 17/500,736.
Prior Publication US 2023/0112186 A1, Apr. 13, 2023
Int. Cl. G06T 7/194 (2017.01); G06T 5/50 (2006.01); G06T 5/94 (2024.01); G06T 7/11 (2017.01)
CPC G06T 7/194 (2017.01) [G06T 5/50 (2013.01); G06T 5/94 (2024.01); G06T 7/11 (2017.01); G06T 2207/20076 (2013.01); G06T 2207/20084 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A non-transitory computer-readable medium comprising instructions that, when executed by at least one processor, cause a computing device to:
determine an alpha matte for a digital image utilizing an object mask neural network comprising an encoder and a decoder, wherein the decoder comprises an alpha-range classifier function and a number of output channels, by:
generating, utilizing the encoder, a feature map from the digital image;
dividing an alpha-range of alpha values into a number of alpha-range classifications comprising sub-ranges of the alpha-range, wherein the number of alpha-range classifications correspond to the number of output channels;
decoding the feature map utilizing the alpha-range classifier function to determine, for a pixel of the digital image, a plurality of alpha-range classification probabilities for the number of alpha-range classifications; and
generating the alpha matte by utilizing a refinement model to determine an alpha value for the pixel from the plurality of alpha-range classification probabilities for the number of alpha-range classifications.