Face Recognition Algorithm And Experimental Study Based On Hyperspectral Image

Face recognition technology can determine the face position,size,facial features and other information of the face in the input of the face image or video based on human facial features,it can also extract the facial features to complete the identification simultaneously.Face recognition technology is involved in a wide range of fields,including biology,economics,military,pattern recognition and some other fields.It plays a pivotal role as a part of biometric identification.Face recognition technology first appeared in the late 19 th century,with the continuous progress of the science technology,the great needs of various fields and the development of the pattern,the development of face recognition technology attract more and more people’s attention,which make it become a hot area of pattern recognition and computer industry.Hyperspectral technology is a technology about ground observation,rising in 1980 s,began with a research program of the imaging spectrometer.Hyperspectral images are different from common gray images and color images,it acquires gray images on a serious of continuous spectra with a small spectral intervals,revealing the reflection and absorption information of electromagnetic energy.Due to the hyperspectral image contains rich image information,hyperspectral image can still be accurately recognized face with good robustness,when the facial features effected by expression changes,deflection,occlusion and other factors.This paper consists of two parts,the first is the establishment of SDNU-HSFD,because of expensive instruments,the hyperspectral face database which established by other schools or research institutes is not public,others can’t obtain to use in their owe study,so the establishment of a rich hyperspectral database for future research is necessary.The establishment of the database contains design the database contents,image acquisition,calibration,denoising,sub-format preservation.The database includes the faces images of 20 volunteers,which contains normal face,side face,occlusion face and the face with facial expression changes.Each image contains 256 bands,which is important for the study of face recognition in hyperspectral images.Secondly it proposes a hyperspectral face recognition algorithm based on Gabor feature fusion.We extract characteristic band of the hyperspectral image firstly,and then we use Gabor filter to extract the feature of the gray image on each feature band in 5 scales and 8 directions,subsequently reduce dimension of the extracted features to obtain the feature vector.In order to obtain the contribution of each feature vector to the recognition rate,a weighting matrix is obtained by weighting method,and the feature vector of each hyperspectral image is obtained by multiplying with the Gabor feature vector,and then we use the K nearest neighbor classifier and the voting method to get recognition rate.The algorithm is tested on PolyU-HSFD,compared with(2D)2PCA and 3d Gabor wavelets,it can improves the recognition rata,and the recognition rate is stable.The algorithm is also tasted on SDNU-HSFD with the recognition rate about 98%.