Wavelets andNeuralNetworks BasedFace

Thispaperintroduces a neural network-based face recognition system. Thte Discrete Wavelet Transform wasused for feature extraction anddimensional reduction methlod as a preprocessing step totheORLdatabase before training thle network A newalgorithm isintroduced forperformance improvement. The algorithm trains twoneural networks, eachwithadifferent input vector, thefirst isthehorizontal imageelements andthesecond is thevertical image elements. Therecognition rate ishighly improved. I. INTRODUCTION Intoday's networked world, theneedtomaintain the security ofinformation orphysical property isbecoming both increasingly important and increasingly difficult. A fundamental flaw intheconventional access control systems is that thesystems donotgrant access by"whoweare", butby "what wehave", suchasIDcards, keys, passwords, orPIN numbers. Noneofthese means arereally defining us.Recently, technology became available toallow verification of"true" individual identity. This technology isbased inafield called "biometrics". Biometric access controls aresuchas fingerprints orface recognition systems, orsomeaspects ofthe person's behavior, like his/her handwriting style orkeystroke patterns. Facerecognition (1,2) isoneofthefewbiometric methods that possess themerits ofboth high accuracy andlow intrusiveness (It hastheaccuracy ofaphysiological approach without being intrusive). Access control byface recognition hasthefollowing advantages incomparison withother biometrics systems: There arenorequirements forexpensive or specialized equipment; asystem maybebuilt using asimple video camera andapersonal computer. Thesystem ispassive; there isnoneed totouch something byfingers orpalm, noneed tosayanywordorlcan eyetoadetcctor. A/ny person just may walkorstaybefore thecamera, andthesystem performs recognition. II.FaceRecognition