US 12,230,058 B2
Systems, methods, and storage media for creating image data embeddings to be used for image recognition
Alireza Sepas-Moghaddam, Kingston (CA); Ali Etemad, Kanata (CA); and Mojtaba Kolahdouzi, Kingston (CA)
Assigned to IRDETO B.V., LS Hoofddorp (NL)
Filed by IRDETO B.V., Hoofddorp (NL)
Filed on Mar. 9, 2022, as Appl. No. 17/690,175.
Prior Publication US 2022/0292877 A1, Sep. 15, 2022
Int. Cl. G06V 40/16 (2022.01); G06V 10/82 (2022.01)
CPC G06V 40/172 (2022.01) [G06V 10/82 (2022.01)] 25 Claims
OG exemplary drawing
 
1. A method for generating an image recognition signal that can be used by a learning network for image recognition, the method comprising:
receiving image data representing the image of an object;
detecting N landmarks on the image using a deep regression algorithm, wherein N is an integer;
constructing a weighted, fully connected, graph kn from the landmarks by assigning initial weights for the landmarks randomly;
determining an optimized tree structure based on the initial weights;
generating a sequence by traversing the tree structure;
forming a series of embeddings based on the sequence, the embeddings representing the object image; and
inputting the embeddings into a neural network to thereby generate an image recognition signal based on the embeddings.