US 12,254,667 B2
Multiple scenario-oriented item retrieval method and system
Xiushan Nie, Jinan (CN); Yang Shi, Jinan (CN); Jie Guo, Jinan (CN); Xingbo Liu, Jinan (CN); and Yilong Yin, Jinan (CN)
Assigned to Shandong Jianzhu University, (CN)
Filed by Shandong Jianzhu University, Jinan (CN)
Filed on Aug. 5, 2022, as Appl. No. 17/881,732.
Claims priority of application No. 202111286401.4 (CN), filed on Nov. 2, 2021.
Prior Publication US 2023/0138302 A1, May 4, 2023
Int. Cl. G06V 10/70 (2022.01); G06V 10/40 (2022.01)
CPC G06V 10/70 (2022.01) [G06V 10/40 (2022.01)] 10 Claims
OG exemplary drawing
 
1. A multiple scenario-oriented item retrieval method, comprising:
extracting, by Hashing learning, image features from an obtained image training set, and training a pre-built item retrieval model;
when an image is in a scenario of hard samples, introducing an adaptively optimized similarity matrix into Hashing learning, optimizing the similarity matrix by an image transfer matrix, and constructing an adaptive similarity matrix objective function with the optimized similarity matrix and an image category label; constructing a quantization loss objective function between the image and a Hash code according to the image transfer matrix; and conducting nonlinear Hashing mapping on the adaptive similarity matrix objective function and the quantization loss objective function to obtain a model objective function;
when the image is in a scenario of zero samples, introducing into Hashing learning an asymmetric similarity matrix configured to constrain the generation of Hash codes, constructing an objective function by taking an image category label as supervisory information in combination with equilibrium and decorrelation constraints of a Hash code, and conducting nonlinear Hashing mapping on the objective function to obtain a model objective function; and
training the item retrieval model based on the model objective function, and obtaining a retrieved result of a to-be-retrieved target item image based on the trained item retrieval model.