Large-scale highly-efficient face recognition method based on Hamming distance

Abstract

The invention discloses a large-scale highly-efficient face recognition method based on the Hamming distance. The method comprises following steps of constructing a convolution neural network; using a sample picture to carry out training on the convolution neural network to generate a model; using a test engineering code to extract a high-dimensional feature and a low-dimensional feature; storing the high-dimensional feature and converting the low-dimensional feature into a binary hash code; putting a to-be-detected picture subjected to the preprocessing into the trained convolution neural network, and extracting and storing a high-dimensional feature and converting a low-dimensional feature into a binary hash code; and successively using the Hamming distance and the cosine distance to carry out retrieval in a coarse-to-fine manner, so as to finally obtain a recognition result. The beneficial effects are that dimensionality reduction compression can be performed in the convolution neural network and the Hamming distance and the cosine distance are used to carry out retrieval recognition twice in a coarse-to-fine manner, so the traditional PCA algorithm is omitted, recognition processes are simplified, calculation time is shortened and retrieval efficiency is improved.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to a large-scale and efficient face recognition method based on Hamming distance. Background technique [0002] At present, face recognition methods are becoming more and more mature, focusing on face positioning and effective feature extraction to achieve a very high recognition accuracy, but the commonly used similarity algorithms basically have a common problem, that is, when the face database to be recognized reaches When a large amount is required, the time cost is quite high, which is a major defect in practical applications. Therefore, if an efficient and high-precision face recognition method can be developed in the case of a large-scale face database, it will have practical application in the field of medical images, public security investigation of criminals, urban residential area security, and end-to-end access control. Apps are of great help. [0003] Using ...

AI Technical Summary

Problems solved by technology

[0003] Using traditional deep learning for face recognition, many algorithms will perform feature dimensionality reduction after extracting high-dimensional features, because high-dimensional features contain rich image information, although they can achieve ultra-high recognition when used for recognition in theory Accuracy, but to find the similarity between high-dimensional floating-point number vectors, the amount of calculation is very large, so it is necessary to use the PCA algorithm to reduce the dimensionality of high-dimensional features, and then perform content-based image retrieval. The performance of this method Will be much better than the traditional face recognition algorithm

However, the PCA algorithm cannot be implemented in the network. It is necessary to use matlab or other tools to reduce the dimensionality of the high-level features extracted from the network through the PCA algorithm. The overall recognition process is complicated, and the damage rate of low-dimensional features is high.

Images