Face retrieval system and method based on deep learning

Abstract

The invention relates to a face retrieval system and method based on deep learning. The method includes the steps: retrieval model training: training a model by a caffe deep learning framework and optimizing model parameters in a gradient descent mode to obtain a retrieval model; face feature extraction: feeding pictures of all people in a training set into a neural network for a feed-forward process and storing output of a first fully connected layer and a second fully connected layer; face registration: sufficiently extracting feature vectors and Hash codes of registrants under different illumination and different postures; face retrieval: extracting Hash codes and feature vectors of face images to be retrieved, comparing the Hash codes and the feature vectors with the pictures in the training set. The most similar face images are obtained by combining coarse retrieval of the Hash codes and fine retrieval of the feature vectors, and corresponding similarity values are outputted.

Description

technical field [0001] The invention relates to the fields of machine vision and face retrieval, and in particular to a face retrieval system and method based on deep learning. Background technique [0002] Since Krizhevsky et al. proposed a deep convolutional neural network Alexnet based on deep learning theory, the field of image recognition has entered a new era. The deep convolutional neural network extracts the features of different levels of images from shallow to deep through convolution operations, and uses learning algorithms to enable the network to automatically adjust the parameters of the convolution kernel for learning. It has achieved remarkable results in image classification and recognition. [0003] The current face retrieval system mainly consists of three steps: preprocessing of face images, feature extraction and feature matching. The preprocessing process needs to detect faces. The technology of this part is relatively mature and will not be repeated h...

AI Technical Summary

Problems solved by technology

However, under the influence of different lighting, character poses, and expressions, the face information changes greatly in the mode; some characters wear decorations on their faces, such as beards, glasses, etc., causing face information to be distorted; There are often various shooting angles, and the images obtained by the same face under different angle shooting conditions are very different

These handcrafted features cannot well characterize the images in the above cases

In addition, the feature dimensions extracted above are often high, which can easily cause dimension disasters, which will greatly reduce the speed of face retrieval and cannot return query results in real time

Images