A gan-based face retrieval method

Abstract

The present invention relates to a GAN-based face retrieval method, comprising: selecting blurred pictures of real scenes and ID card photos corresponding to the blurred pictures as training data sets; selecting different numbers of blurred pictures of real scenes as different test data sets; According to the pre-trained face recognition network model, extract the blurred picture features and the ID card photo features corresponding to the blurred pictures; calculate the correct rate of face retrieval in the test data set and use it as a reference benchmark; input the blurred pictures into the generation network G and get the output Results: The cosine distance between the output feature of the fuzzy image and the ID card photo feature corresponding to the fuzzy image constitutes the loss function of the generation network G; the GAN generative confrontation network model is trained with the training data set; the calculation test data set is passed through the GAN The face retrieval accuracy rate after the generative confrontation network model is processed; the GAN generative confrontation network model is optimized according to the reference benchmark.

Description

technical field [0001] The invention relates to the fields of computer vision and image processing, in particular to a GAN-based face retrieval method. Background technique [0002] Face recognition technology is a biometric recognition technology based on human facial feature information. Now this technology has been widely used in government, military, banking, social welfare, e-commerce, security and defense and other fields. In the field of security, for some people captured by surveillance, public security personnel often need to obtain the identity information of these people. At this time, they will use the photos of a person they have taken and the ID photos in the information database to perform face comparisons one by one. Generally speaking, the photo with the highest similarity score in the comparison results will be considered to be the person. ID photo. [0003] In real scenes, the captured photos are often blurred to a certain extent due to the movement of p...

AI Technical Summary

Problems solved by technology

[0003] In real scenes, the captured photos are often blurred to a certain extent due to the movement of people or the camera is out of focus, and the current search results for blurred photos are often unsatisfactory.

Images