Development the algorithm of automatic searching man's faces defined points by photo, страница 2

The pupil on the image is represented by a circle, accordingly parametrical equation describing it looks like:

where  - pupils center coordinates,  - pupils radius. As for the task of a circle it is necessary to know three parameters, then the appropriate phase space (discrete phase space also is called as the accumulator) will be three-dimensional. Further there is filling the accumulator: the meaning of each cell is considered equal to the amount of black pixels on the image belonging, even one curve (5).

The following step is the search of local maxima of the accumulator laying on distance,  (d – diameter of a pupil equal to 12mm).

2.4.  Anthropometrics points definition using Gabor wavelets

For allocation informative attributes on the image of the person it is possible to use filters on a basis Gabor wavelets [4, 5]. The algorithm of definition basic anthropometrics point (defined points) on the image basis on the assumption of existence of such variants of Gabor filters , which have extreme meanings in required defined point in comparison with meanings calculated in its vicinity.

The general view of the Gabor filter is represented as follows

where                                                 

 - two-dimensional Gaussian with standard deviations  , , revolved on a corner . Each variant of the filter is described in six parameters. For definition of coordinates given defined point the discrete meanings of the filters parameters get out, all possible combinations of these meanings are calculated and the variants of filters are generated. Each of variants of the filter is calculated for all pixels of a vicinity of the defined point. The meanings of the Gabor filters are complex numbers; therefore in each concrete case it is necessary to calculate extremums for the real and imaginary parts separately. Thus, are defined four extremums for each filter in a vicinity of the separately taken anthropometrics point of a concrete photograph? The site of each of extremums of the separate filter concerning a required point is estimated by the metrics in pixels and averaging out under all processed images. The received results are analyzed with the purpose of selection of the filters variants with the minimal deviation founded extremums from the given point.

3.  conclusion

The results of the algorithms researching shows that the Gabor filters allow reliably determine coordinates of the person’s anthropometric points. Is established, that the results of preliminary processing of the images essentially influence accuracy of search; the Derish operator is expedient to apply to definition of coordinates of the majority defined points [6, 7].

4.  REFERENCES

1.  U .Pratt. Digital Image Processing: Moscow: Mir, 1982 (in Russian).

2.   A. Fyodorov. Binarization of Black-White Images: current state and perspectives.

URL: http://iu5.bmstu.ru/~philippovicha/ITS/IST4b/ITS4/Fyodorov.htm.

3.  K. Marinichev, V. Vezhnevets. Parametric Curves Selection Algorithms Based on Huf Transform. URL: http://cgm.graphicon.ru/ .

4.  Benait Duc, Stefan Fisher, Joseph Bigum “Face Authentication with Gabor Information on Deformable Gtaphs” IEEE Transaction on Image Processing, Vol 8, No 4, April 1999.

5.  V. N. Gorlov, A. A. Malafeeva, S. I. Malafeev. Algorithmic Means of Human Identification by Portrait Based on Geometric Transforms. / Design and Technology of Electronic Devices, No 4. -  pp. 45-52, 2005.

6.  A.V. Nefian and M.H. Hayes “Face recognition using an embedded HMM” in Proceeding of the IEEE Conference on Audio and Video-based Biometric Person Authentication, pp. 19-24, 1999.

7.  V. N. Gorlov, S. I. Malafeev, F. A. Krasnov. System of Automatic Video Recognition Based on Neural Network, No 2. - pp. 31-39, 2001