CPC G06T 11/60 (2013.01) [G06T 5/20 (2013.01); G06V 10/32 (2022.01); G06V 10/82 (2022.01); G06V 40/171 (2022.01); G06T 2207/20004 (2013.01); G06T 2207/30201 (2013.01)] | 10 Claims |
1. A method for composing a background and a face using a deep learning network, the method comprising:
(a) receiving, by an image input unit, an original face image and a converted face image;
(b) performing, by a data preprocessing unit, data preprocessing on the original face image and the converted face image;
(c) extracting, by a feature vector extracting unit, a first original face feature vector and a first converted face feature vector from the original face image and the converted face image, respectively;
(d) generating, by a feature vector mask generating unit, an original face feature vector mask and a converted face feature vector mask from the first original face feature vector and the first converted face feature vector, respectively;
(e) generating, by a mask-based adaptive object normalizing unit, a composite mask from the original face feature vector mask and the converted face feature vector mask, extracting a second original face feature vector and a second converted face feature vector using the composite mask, respectively, and then performing adaptive object normalization; and
(f) composing, by an image generating unit, a converted face of the converted face image with a background of the original face image, the background being a remaining part excluding an original face in the original face image, by performing minimum/maximum normalization to change a data range of the composite mask to a minimum value of 0 and a maximum value of 1.
|