Adaptive dictionary construction method for sparse representation of face recognition

Abstract

The invention discloses an adaptive dictionary construction method for the sparse representation of face recognition. The method comprises a step of obtaining a face image to be tested and a plurality of face image samples, storing the face image samples into a training sample set, and extracting the face image to be tested and the LCP characteristic of each sample in the training sample set, a step of obtaining the average LBP characteristic similarity and average MiC characteristic similarity of all samples in the face image to be tested and the training sample set, a step of judging whether the LBP characteristic and the MiC characteristic belong to valid characteristics, and refusing the recognition of the face image to be tested, otherwise going to a next step, a step of marking the corresponding sample in the training sample set as the neighbor sample of the face image to be tested when the LBP characteristic similarity is not smaller than an LBP characteristic neighbor threshold and the MiC characteristic similarity is not larger than an MiC characteristic neighbor threshold in the valid characteristics, and a step of taking the neighbor sample of the face image to be tested as an atom and constructing an adaptive face recognition redundancy dictionary.

Description

technical field [0001] The invention belongs to the field of face recognition, and in particular relates to an adaptive dictionary construction method for sparse representation face recognition when a classification-based sparse representation algorithm is adopted. Background technique [0002] Face recognition is a technology that uses people's facial images for identity recognition, and has broad application prospects in economic, civil, military, public security and other fields. Sparse representation utilizes an over-complete redundant dictionary to perform an optimal linear representation of the input test data, and the representation coefficients are sparse vectors. Classification-based sparse representation algorithms are a class of algorithms that use sparse representation vectors and various training samples to reconstruct and classify test data class by class. A large number of literatures show that classification-based sparse representation algorithms have a high...

AI Technical Summary

Problems solved by technology

First, increase the number of atoms to increase the sparsity of the representation vector to improve the recognition rate, but the calculation speed of any sparse solution algorithm will decrease as the number of atoms increases

Second, for any input test image, the redundant dictionary is fixed and not adaptive

Images