A face recognition method based on distinguishability feature fusion

Abstract

The invention discloses a face recognition method based on distinguishability feature fusion, comprising the following steps: A, cutting a global image and at least two local images in each training sample image; B, carrying out model training on each intercepted image by adopting a multi-loss function to obtain a corresponding model, wherein the multi-loss function is obtained by combining the a-softmax loss function with the center loss function. C, carrying out fusion and dimension reduction of each model obtain by training by using a ternary loss triplet loss function, and obtaining the final depth feature of the training sample image. The technical proposal disclosed in the application can solve the problem of data fusion, face posture and model fusion in the process of face recognition by using CNN, and achieve better face recognition effect.

Description

technical field [0001] The present application relates to the technical field of face recognition, in particular to a face recognition method based on fusion of distinguishable features. Background technique [0002] Existing face recognition through deep learning methods has achieved a series of great breakthroughs. Several existing technologies are briefly introduced below: [0003] One existing method is to map the difference between face image pairs into distances. The training criterion is that the similarity distance between image pairs of the same category should be small, while the similarity distance between image pairs of different categories should be larger. big. [0004] Another existing method is to pass a non-linear transformation such that there is a distinguishable boundary in the middle of the distance between the same image pair and different image pairs. This method requires input image pairs as input. [0005] In another existing method, an angular sp...

AI Technical Summary

Problems solved by technology

It is difficult to adapt to such a scene of large pose changes by simply training a model

[0010] The existing multi-model fusion method directly connects the depth features of multiple models as the final feature of the image. Not only the feature dimension is high, but there are also problems such as redundancy between features, which affects the classification effect instead.

Some methods connect the features and then perform PCA (Principal Component Analysis) processing. This type of method has a certain effect, but it is more troublesome to implement, and the features processed by PCA may face the problem of poor distinguishability. question

[0011] In order to achieve the stability of the result, the existing face recognition method will combine the original image and the symmetrical image when making the final category judgment. Some methods are to directly connect the features of the two images, and some methods are to extract the features On average, these methods can achieve a certain effect improvement, but the accuracy of category judgment still needs to be improved

Images