Human face image recognition
 

FASlib library

Wide interest in human face recognition automatic systems has designated for authentication software and hardware development. Leading firms, manufactures of access control systems, offer special automotive complexes, connected with automatic turnstile, which ensure human face recording, recognition, and access to secured territory. But this complex is very expensive. Staff register software modification is the alternative way.

We have developed FASlib system. It is the function library which authenticates personnel by their face image and assigned to any access control system modification. TV picture with human face and authentication vector from access control database pass to software. If degree of conformity is higher than user set level, personnel access is granted. Sick face color and expression influence on degree of conformity. In this cases additional control is needed.

DLL library FASlib is build-in to special user software, controlling staff access. One or several personnel photographs of a man, which must be authenticated, is transformed by initial library function SynthExtract into authenticated vector. Recognition function AuthCompare is called for image from TV-camera and authentication vector comparison. Obtained degree of conformity must call corresponding user software reaction.

FASlib has six functions: two main - SynthExtract and AuthCompare - and four auxiliary, which provide user software and library interaction. In the basis of library algorithms is modern image processing methods PCA, LDA and KLT, which is most perspective in modern biometric systems.
Information confidence and selected security level provided by authentication vectors, which is stored by user himself and access decision which is made by user himself to.

FASlib application example

// 1. State vector (SV) forming
Graphics::TBitmap *bmp;
FASPVector pv;	// vector of parameters forming
pv.Width=bmp->Width;
pv.Height=bmp->Height;
void *FS=SynthOpen(&pv);	// object creation, used for SV synthesis
FASExtrSV sv;	// SV synthesis parameters
HBITMAP hbmp=bmp->Handle;
sv.NBitmap=1;	// sample images number
sv.bmp=&hbmp;	// sample images pointers array
SynthExtract(FS,&sv);	// synthesis execution
SynthClose(FS);
// 2. Authentication by counted SV
void* FA=AuthOpen(&pv);	// object creation for image authentication
FASImageAuth iv;	// image authentication parameters
iv.bmp=bmp->Handle;	// authenticated image
iv.SVLength=sv.SVLength;	// SV length
iv.SVector=sv.SVector;	// SV pointer
// 3. Authentication execution
AuthCompare(FA,&iv);
if(iv.Similarity > ErrorLevel);	// degree of conformity check
OpenTheDoor();	// access granted
else
AccessDenied();	// access denied
AuthClose(FA);
GlobalFree(sv.SVector);

References

1. Аракчеев П.В. Применение нейросетевых алгоритмов для распознавания контрастных телевизионных изображений // Международный форум информатизации: Доклады международной конференции «Информационные средства и технологии» – М.: Изд.-во «Станкин», 2001. – Т. 1.
2. Buryi E., Jafar A. Invariant Features Extraction Forming for Contrast Planer Objects Recognition // 8–th Int. Conf. ACS’2001, Oct. 17 – 19, Mielno, Poland. – Mielno, 2001. Part. 2. – Szczecin: Informia, 2001.