US 12,254,543 B2
System and method for personalized cartoon image generation
Rahul Prasad, Bangalore (IN); Abhishek Sharma, Noida (IN); and Mudit Rastogi, Raebareli (IN)
Assigned to Talent Unlimited Online Services Private Limited, South Delhi (IN)
Filed by TALENT UNLIMITED ONLINE SERVICES PRIVATE LIMITED, Delhi (IN)
Filed on Nov. 11, 2022, as Appl. No. 17/985,449.
Claims priority of application No. 202271038666 (IN), filed on Aug. 5, 2022.
Prior Publication US 2024/0046536 A1, Feb. 8, 2024
Int. Cl. G06T 11/40 (2006.01); G06T 7/10 (2017.01); G06T 11/00 (2006.01)
CPC G06T 11/40 (2013.01) [G06T 7/10 (2017.01); G06T 11/001 (2013.01); G06T 2207/30201 (2013.01)] 15 Claims
OG exemplary drawing
 
1. A method (100) for personalized cartoon image generation comprising the steps of:
a. launching a keyboard interface by clicking on an application icon from a device application launcher by the user (101);
b. capturing a digital picture displaying a photorealistic content, acquired with a plurality of sensors used on a mobile device (102);
c. subjecting the digital picture to face segmentation to produce a segmented picture of the user's face (103);
d. normalizing the segmented picture to obtain a facial normalized image (104), and wherein normalization is performed to scale the segmented picture into a range which is familiar or normal to the senses, such that visual appearance of image is increased for the visualizer;
e. feeding the facial normalized image to a trained AI (Artificial Intelligence) model as an input for face cartoonification (105), and wherein the trained AI model learns cartoon facial characteristics that co-relate to a photorealistic image facial feature;
f. aligning the facial normalized image fed to the trained AI mode as an input using facial landmark extraction library and extracting facial landmark of the photorealistic image (106) by using 68 face landmark points to obtain a plurality of cartoon images, and wherein the extraction of facial landmark includes one to one mapping of media-pipe facial points to dlib facial points;
g. passing aligned facial normalized image to an encoder architecture to extract, optimize, and condense a latent vector;
h. manipulating the condensed latent vector using direction vector of expressions, and wherein the direction vector of expressions includes smile sad;
i. decoding manipulated condensed latent vector using a GAN (Generative Adversarial Network) generator to transfer facial expression feature; and
j. customizing the plurality of cartoon images (107) produced by the trained AI (Artificial Intelligence) model based on the facial landmark extracted, and wherein the plurality of cartoon images generated is customized with a plurality of sticker bodies and a plurality of styles.