A Face Feature Extraction Method Based on Heterogeneous Tensor Decomposition

Abstract

A face feature extraction method based on heterogeneous tensor decomposition: use any camera array to collect color views of different faces at different angles of view, and obtain the initial view set of faces after image processing such as grayscale transformation and normalization ; Extract the images of the initial view set, and stack them in turn to form a third-order tensor, wherein the third-order dimension of the formed third-order tensor corresponds to the total number of face images; TUCKER decomposes the formed third-order tensor to obtain The nuclear tensor, the first mode factor matrix, the second mode factor matrix and the third mode factor matrix, and update them; judge whether the nuclear tensor is convergent; the third mode factor matrix Z obtained by decomposing the high-dimensional face The data is mapped to the mode factor matrix of the low-dimensional feature subspace, therefore, the third mode factor matrix Z is the final extracted face feature. The invention realizes the automatic extraction of the feature of the human face image, avoids the cumbersome steps of the traditional feature extraction, and improves the feature extraction speed.

Description

technical field [0001] The invention relates to a method for extracting human face features. In particular, it involves a face feature extraction method based on heterogeneous tensor decomposition. Background technique [0002] Data from multiple sources are arranged to form a tensor [1]. In order to dig deep into the intrinsic information of the tensor, it is necessary to decompose the tensor. Tensor factorization is an emerging powerful tool for exploring multidimensional data. The TUCKER and PARAFAC models are the most basic models of tensor decomposition. Tensor decomposition extracts and classifies features by capturing the multiple linear and multi-angle structures of large-scale multi-dimensional datasets. Tensor decomposition is widely used in medicine and neuroscience, social network analysis, computer vision [2], recommendation system and other fields. [0003] Supervised and unsupervised dimensionality reduction and feature extraction based on tensor representa...

AI Technical Summary

Problems solved by technology

Traditional algorithms such as Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), etc. treat multidimensional data as matrices or vectors, and cannot make good use of the structural information inside the data.

Images