Human face feature recognition method combined with image quality analysis and metric learning

Abstract

The invention discloses a human face feature recognition method combined with image quality analysis and metric learning. The method comprises the following steps: processing a human face image, and then adding the processed human face image into a convolution neural network to train so as to acquire the primary feature; obtaining a quality quantized value of the human face image according to the norm of the primary feature; adding a full-connection parameter layer to perform dimension-reduction processing on the primary feature to obtain the final feature; taking the quality quantized value of the human face image as an additional weight information to obtain the weighted mean value according to the final feature, and adding the weighted mean value into the training to obtain a loss function of the metric learning; supervising the training procedure through the loss function to obtain the final parameter of the convolution neural network; taking the final parameter as the final human face feature extraction module and performing the human face feature recognition on the human face image. The quantized value of the image quality can be acquired by using few calculations, and a unified quantization standard is formulated according to the norm value of the human face feature; the network presentation skill is improved, and the adaptability of the network to the low-quality image is enhanced.

Description

technical field [0001] The invention relates to a face recognition method, in particular to a face feature recognition method combining image quality analysis and metric learning. Background technique [0002] Most of today's face recognition systems preprocess the face image data, and use convolutional neural network training to obtain the weight of the network; calculate the face feature vector according to the weight of the trained network, and obtain the face feature vector by processing the feature vector. The result of face recognition. Due to the large difference in image quality between the face data participating in the training, some low-quality face pictures, such as low resolution, blur, too strong or too dark light, large face deflection, etc., will affect the learning of parameters effect, thereby reducing the performance of the recognition system. The existing technologies for facial feature recognition combined with image quality are as follows: [0003] a...

AI Technical Summary

Problems solved by technology

[0004] Disadvantages: Manual labeling consumes a lot of manpower and material resources, and the quantitative standards for image quality are not uniform

[0006] Disadvantages: Need to train the network separately to deal with the quality of the network, more parameters are needed, and the recognition speed is slow

[0008] Disadvantages: low utilization rate of data, poor recognition effect on low-quality pictures such as low resolution

Images