Low-resolution Face Images To Identify The Key Technologies

Face recognition is now a hot research field of Computer vision, and it is also animportant issues of pattern recognition. Face recognition has been widely used foranti‐surveillance department, photo search, access areas such as access and identityfor its ease of use, highly user acceptance, intuitive prominent, fast and difficult tocounterfeit and other advantages. Because of the limitation of data storage space,network transmission speed and image acquisition equipment factors, most of the faceimages which used to identify are low‐resolution. However, face recognition algorithmsare generally based on ordinary resolution face images, so it's a great significance forlow‐resolution face recognition.Practice has showed that local binary pattern (LBP) algorithm to extract thefeatures of low‐resolution images can achieve good recognition results. Therefore, thispaper first analysis the local binary pattern, and then presents a multi‐resolution andmulti‐scale LBP feature extraction algorithm. For each face image, feature is extractedfrom the different resolutions and different scales by LBP to get the local and globalLBP features.Though the multi‐resolution and multi‐scale LBP feature extraction algorithmhas a better presentation of the local and global features of low‐resolution face images,but it also bring a large amount of feature data and high data redundancy problems.Therefore, it is need to find a suitable dimensionality reduction method to solve thisproblem. The traditional method of data reduction although has achieved the purposeof reducing data dimensionality, but it brings the problem of lossing a large number ofuseful information.This article also introduces a new method of feature reduction: Local MarginAlignment (LMA). LAM can not only reduce the dimension of high dimensional data,and the high‐dimensional data is mapped to a lower dimensional subspace, but alsoretain the category structure. This article uses LMA to reduce the dimension of LBP features, to make the calculation reduced greatly, while retaining the class structure,which makes the feature space after dimension reduction is more separability.Experiments showed that a combination of multi‐resolution and multi‐scale LBPfeature extraction and dimensionality reduction algorithm LMA has got a significanteffect on low‐resolution facial image recognition.