A Face Image Reconstruction Method Based on Identity Information

Abstract

The invention provides a face image reconstruction method based on identity information, comprising: Step 1, creating a deep network model; Step 2, reading a pre-trained face recognition model; Step 3, down-sampling the training data and Clipping processing; step 4, input the training data into the deep network model to obtain the reconstructed image; step 5, input the reconstructed image into the face recognition model to obtain the feature vector; step 6, input the original image after clipping In the face recognition model, get the feature vector; step 7, calculate Quadruplet-loss; step 8, calculate MSE-Loss; step 9, update the parameters of the deep network model through back propagation; step 10, through the image before and after reconstruction , and the corresponding label information, update the weight parameters of the face recognition model; step 11, repeat steps 3 to 10, until the loss function value in step 9 converges.

Description

technical field

[0001] The invention relates to a face image reconstruction method based on identity information. Background technique

[0002] In recent years, the rise of deep learning has provided good solutions to many problems in the field of computer vision, enabling more and more computer vision research results to play a role in real life. Face recognition is one of the hottest directions. one. In the wave of deep learning in the past five years, the accuracy of face recognition algorithms has been improved unprecedentedly. From the experimental results, it has surpassed the ability of human beings in face recognition tasks. However, the environment considered in the experiment is often relatively simple, but there are many unexpected interferences in practical applications. For face recognition tasks, the most common is that the image quality is low due to the interference of various external factors, which will seriously affect the accuracy of the face recognitio...

Images