Face living body recognition model compression and transplantation method based on depth separable convolution

Abstract

The invention relates to a face living body recognition model compression and transplantation method based on depth separable convolution. The method comprises the following steps: S1, obtaining a training data set through a data enhancement mode; S2, training the image by using an improved convolutional neural network, and storing a convolutional neural network model obtained after training; S3, compressing the model based on depth separable convolution to reduce the size of the model and reduce model parameters to about 20% of that of an original model, and the size of the model is more suitable for a mobile terminal; and S4, carrying out semi-precision Float16 quantization on the model weight to further compress the model and accelerate the model reasoning speed, so that the size of the model is compressed to be 50% of that of the step S3, the mobile terminal identification speed is shortened to be 400ms, and the model is transplanted in mobile terminal software. Compression and Float16 semi-precision quantization are performed on the model based on depth separable convolution.

Description

technical field [0001] The present invention relates to the fields of computer vision, deep convolutional neural network and model compression, in particular to a face recognition model compression and transplantation method based on depth separable convolution. Background technique [0002] With the increasing maturity of image processing technology and computer vision algorithm, face recognition technology has developed vigorously, and face anti-counterfeiting technology is also an important research topic. Liveness detection is a method to determine the real physiological characteristics of an object in some authentication scenarios. In face recognition applications, liveness detection can use facial key point positioning and face tracking technologies through combined actions such as blinking, opening mouth, shaking head, and nodding. , to verify whether the user is operating as a real living person. It can effectively resist common attack methods such as photos, face c...

AI Technical Summary

Problems solved by technology

[0004] Some of the above-mentioned mainstream network compression techniques are limited to pruning algorithms such as channel deletion, which will greatly damage the model performance and cause a sharp drop in model accuracy.

Images