Home – Home – Jornals – Avtometriya 2006 number 4

Synthesis and analysis of the maximum likelihood algorithm for detecting a spatially extended object by its image are carried out. Efficiency of the algorithm and the scope of applicability of the found asymptotic expressions for the detection characteristics are determined by means of statistical computer modelling.

Methods for determining the parameters of the algorithm for segmenting object contours in their parametric representation are proposed. Analytical formulas that establish connection between the algorithm parameters, the noise characteristics, and the conditional probability of error consisting in taking the rectilinear segment of a contour line as the curvilinear one are derived. It is shown that for the parametric representation of contour lines the segmenting algorithm efficiency does not depend on the orientation of a contour line on the plane.

Efficiency criteria for numerically strict evaluating the quality of various methods and algorithms for filtering images distorted by pulse noise are proposed and substantiated. Results of computational experiments concerned with investigating statistical dependence of the criteria on the picture variance, the value of coordinate discretization (size), and the intensity level quantization of images while filtering in the range of pulse noise from 0 up to  are presented and discussed.

Results of synthesizing noise-proof signals obtained for the case of equidimensional alphabet and signal are generalized. A problem of synthesizing discrete complex-valued signals with maximum mutual distances is solved for the case of different alphabet and signal dimensions. The solution is based on results obtained in synthesis of simplex signals. It is shown that for a signal, dimension k the simplex signal alphabet volume is and the total number of alphabets is infinite. The obtained results make it possible to estimate potential noise immunity of image recognition systems in the case of arbitrary signal dimensions and alphabet volumes.

A spectrum estimation algorithm based on representing the signals in the modified Sobolev space with the use of the fast Fourier transform is proposed. Expressions for determining the parameter of the modified Sobolev space, which ensures the minimum spectrum estimate error caused by the low sampling frequency and the quantization noise, are derived. The possibilities of increasing the spectrum computation speed by a factor of 2 compared with algorithms in the space L_{2 and also reducing the quantization noise by 5}

Relationships between distances in different spaces, namely, in Euclidean, visual, and also in Klein and Poincare models are found. Typical surfaces in the global structure of the isotropic visual space are shown. Functions of perceived distances in estimating the object size and distance are considered. Problems concerned with the visual space anisotropy are discussed.

Designing a mirror to generate a directivity diagram in the form of a line segment represented by a vector function of one argument is considered. An analytic solution of the mirror design problem to produce a directivity diagram in the form of a line segment for a point source is obtained. The mirror design is reduced to the solution of a conventional differential equation resolved with respect to the derivative. A mirror to produce a directivity diagram in the form of a line segment with an angular size of and a segmented mirror to produce a square-shaped directivity diagram are designed.

A method for numerical calculating the nonuniform transmittance of one of the cavity mirrors to obtain the given output intensity distribution is proposed. The calculation was done for the axially symmetric case of field distribution in the cavity. Transmittances for obtaining a beam with a uniform intensity distribution in a circle are calculated.

Results of investigating the accuracy of reconstruction a 3D surface of a human trunk model by computer optical topography are considered in comparison with the multislice computer tomography.

An adaptive algorithm for preprocessing signals that represent short information phase-shift keyed packages against noise in the presence of Doppler effect is considered. Processing based on phase break allocation of initial signals makes it possible to avoid frequency shift equalization in problems concerned with estimating the time delay of phase-shift keyed signals in the case of multichannel distribution.

A Doppler signal processor representing a complex envelope tracking filter supplemented with a digital error correction system is described. The LDA processor characteristics in measuring the gas flow turbulent characteristics are studied.

Design features and construction principles of silicon multiplexer matrices based on production technologies are considered. Two versions of unified 128 x 128 multiplexers based on 1.2 m CMOS and 3 m CCD/CMOS technologies are presented.