Unconstrained face recognition method based on weighted block tensor sparse graph mapping

A face recognition, non-constrained technology, applied in the field of face recognition, can solve the problems of difficult, unknown, and complex data distribution in pre-defined neighbor graphs, and achieve the effect of being conducive to accurate recognition and improving accuracy.

CN111783615APending Publication Date: 2020-10-16NANJING INST OF TECH
0 Cites 2 Cited by

Patent Information

Authority / Receiving Office
CN · China
Current Assignee / Owner
Publication Date
2020-10-16

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention discloses an unconstrained face recognition method based on weighted block tensor sparse graph mapping, and relates to the technical field of face recognition methods. Firstly, an original sample image is divided into B blocks, each image block is represented by a second-order tensor, a category label is introduced, and a supervised over-complete block tensor dictionary is constructed; on the basis of solving similar sparse reconstruction coefficients of the block samples through regular constraints, intra-class compactness constraints and weight constraints are added, intra-class differences between the similar block samples are further represented through distance weights, and a more accurate sparse neighbor graph is constructed in a block sample tensor space in a self-adaptive mode; and finally, bilateral low-dimensional projection is performed on the block samples, and a global constraint factor is introduced to obtain a bilateral low-dimensional projection matrix. According to the method, the complexity of the non-constrained face image data is fully considered, the neighbor distribution graph of the high-dimensional tensor data is obtained in a self-adaptive mode through the sparse representation technology, the low-dimensional manifold essential structure of the highly-distorted non-constrained face data is effectively extracted, and the accuracy of non-constrained face recognition is greatly improved.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The present invention relates to the technical field of face recognition methods, in particular to the technical field of unconstrained face recognition methods based on weighted block tensor sparse graph mapping. Background technique

[0002] With the rapid development of mobile Internet, electronic sensing technology, and machine learning theory, real-time collection of face images for identity authentication, video surveillance and human-computer interaction has become an important application of artificial intelligence in actual work and life. Due to the mixed interference of various factors such as illumination, posture, expression, occlusion, age, and resolution, the face data collected in the real environment leads to the diversity of face images and a highly complex nonlinear distribution in high-dimensional space. . Therefore, how to effectively reduce the dimensionality of high-dimensional massive unconstrained face data is particularly impo...

Examples

Embodiment Construction

[0020] The present invention is based on the unconstrained face recognition method of weighted block tensor sparse graph mapping, combines sparse representation, block tensor representation and multi-dimensional projection technology, and proposes a new weighted block tensor sparse graph mapping (Weighted Block Tensor Sparse Graph Embedding, WBTSGE) algorithm. First, the original sample image is divided into B blocks, each image block is represented by a second-order tensor, category labels are introduced, and a block tensor dictionary (Block TensorDictionary, BTD) with super-complete supervision is constructed; secondly, block samples are solved under regular constraints On the basis of the sparse reconstruction coefficient of the same class, the intra-class compactness constraints and weight constraints are added to enhance the neighbor relationship between the same block samples, and the distance weights are used to further characterize the intra-class differences between th...