A face image quality assessment method and system based on convolutional neural network

Abstract

The invention discloses a face image quality assessment method and system based on a convolutional neural network. The method steps include: 1) constructing a deep convolutional network as a large network, and a small network with a shallow convolutional layer; 2) using the marked training samples to train the large network and the small network respectively, until the small network outputs The feature vector of the large network is basically the same as the feature vector output by the large network; wherein, in each iteration training, the feature vector output by the large network and the feature vector output by the small network are used as the input of the regression loss function layer of the small network. 3) Input the target face image into the small network trained in step 2), obtain the feature vector of the target face image and input it into the quality assessment network, and use the quality assessment network to calculate the image quality of the target face image. The present invention greatly optimizes the accuracy and real-time performance of image quality evaluation.

Description

technical field [0001] The invention belongs to the technical field of image processing, and is especially applied in the field of face recognition in real-time video monitoring scenes, and relates to a face image processing method and system. Background technique [0002] Face recognition technology is a hot topic in current research and has broad application prospects in a variety of application scenarios, including account opening identity authentication in the financial field, VIP identification in the access control system, key person identification in the security field, etc. Although the recognition performance of face recognition technology has been greatly improved, for face recognition in surveillance video in complex environments, due to the influence of interference factors such as face angle, scale, illumination, occlusion, noise, motion blur, etc. , leading to generally poor quality of the acquired face images, resulting in a large number of misrecognition and ...

AI Technical Summary

Problems solved by technology

This method solves 1) the problem of inconsistent normalization standards for different types of feature fusion in traditional methods, and can obtain a unified effective feature vector representing the quality of face images, 2) the problem of slow feature extraction in deep convolutional networks, through To balance the requirements of speed and accuracy, first construct a neural network of appropriate size, and then use the feature expression obtained by the large network to guide the learning of the small network (such as the logistic regression cost function), so that the small network can obtain the same feature vector as the large network. And the feature extraction speed is equivalent to the calculation time of conventional methods. 3) The problem of low precision of directly constructing the trained shallow neural network is found through experiments: 1. Generally speaking, the larger the model, the better the feature expression ability. 2. The fusion of multiple models Features are more expressive than single-model features, and multiple large models are used to obtain better features in order to improve accuracy, but in actual scenarios, the model is too large to be practical. Therefore, the features learned through multiple large models Guide the small network to learn, and finally get a usable small model, which can achieve the best results in speed and accuracy

Images