A low-resolution single-sample face recognition method

Abstract

The invention discloses a low-resolution single-sample face recognition method, comprising the following steps: constructing a single-sample training set and a low-resolution test set; constructing a unified local feature extraction model, using the model to extract the difference between the test sample and the training sample Face convolution features with fixed dimensions in local blocks; build a local collaborative representation model, use this model to reconstruct the local block convolution features of a single face sample in the training set, and generate local blocks with multiple intra-category changes Block convolution features; build a fusion discriminant model, use this model to calculate the comprehensive similarity distance between all local blocks of the test sample and the reconstructed training sample, use this distance to calculate the minimum reconstruction error of the sample, and will have the minimum weight The training sample label of the structural error is used as the category label value of the current test sample. The face recognition method based on fusion and discrimination of local collaborative representation features constructed by the present invention can significantly improve the face recognition rate in a low-resolution single-sample scene.

Description

technical field [0001] The invention relates to image processing technology, in particular to a low-resolution single-sample face recognition method. Background technique [0002] For the low-resolution face recognition problem, the current solutions fall into the following two categories: based on super-resolution reconstruction techniques and unified feature space projection. Among them, the super-resolution reconstruction technology is to reconstruct the low-resolution face features to increase the features with discriminative characteristics in the face image. The method based on unified feature space projection is to project face features of different resolutions into a unified feature space, and perform face recognition in the projected feature space. The above two solutions can effectively solve the problem of low face recognition rate in low-resolution scenes. If the above solution is directly applied to the face recognition of a single training sample, then the fa...

AI Technical Summary

Problems solved by technology

If the above two methods are directly used to recognize low-resolution faces, it will be impossible to extract face features with discriminative characteristics from the image because the image resolution is too low, resulting in a severely low face recognition rate, which cannot meet the actual requirements. Application requirements

[0004] From the above analysis, we can see that none of the existing solutions can effectively solve the problem of low resolution of test samples and only one training sample for each class.

Images