This pap er presents a new superv ised learn ing sc heme, which use s hyb rid infor mation i.e . global and
local object inform ation, for accur ate id entifica tion and cl assifica tion at conside rably high speed bo th in
trainin g and tes ting phase. The first contrib ution of this paper is a unique image represe ntation usin g
bidir ectional two -dimension al PCA and Ferns style approa ch to represe nt global and local infor mation,
respec tively, of an object . Second ly, the app lication of extreme learning machi ne s upports reliab le
recogn ition with m inimum error and learning speed approx imatel y tho usands of times faster th an
traditio nal neur al netw orks. The prop osed m ethod is cap able of classi fyin g variou s dataset s in a fra ction
of second compa red to other modern algo rithms th at require at least 2–3 s per image