Discriniination among Individuals Using 0 rt hogon a1 Transformed Face Images
Yasimori Nagasaka, Atsunori Yoshikawa · 1997
In this study. we investigated the discriniination aniong individnals by coiiiparing the individual face iiiiage which applied an orthogonal transform with the refererice images. Image recognition is affected by taking conditions of the image; like the distance from a caniera to a fxe or the brightness of illumination. Therefore a normalizing niethod which adjusts to the change of distance and the brightness is needed for corr recognition. Adj tist merit of distance was carried oiit by enlargirig/reducing tlie iniage based on the positions of eyes. a nose and a ~iioiitli. The orthogoiial transform is suitable for extracting tlie dial rist,ic parts of an image. Moreover. it is riot, affe by the change of brightness of illuiiiination. In this paper. we applied five kiiids of trarisforiiis (Discrete Fourier, WalshHadarnard. Haw. Discr Cosine and Wavelet) to tlie hiiiiiari face iniage. Arid for each transforiii. two kiiids of pattern niatdiing niethods were examined. One is iisirig the whole transfornied iniage with no change, arid tlie other is using the weighted images. Weiglitiiig is perfor~iied to extract only a sriiall part which includes individual characteristics from the iiiiage. We coiiipared discrimination rates for five kinds of transforms. As a resnlt. it has shown that the weightiiigs was eff'ective to improve the discriniinatioii rate. For all five transforms. the discriniination rate was iniproved 5% - 40% by weighting. In addition only siiiall parts of the transfornied iniage were usefiil for discriiiiination. When Discrete Fourier traiisform was iised. the highest discrimination rate 96.8% was obtained.