K.Yu. Gavrilov, E.V. Kononova

A description of objects by means of attributes invariant to the scale and rotation of an object around its axis is introduced in the paper. Calculation of the attributes is performed in two steps. At first, the object edge as a curve of abrupt brightness jump is formed. Hereat, the initial continuous-tone image represented in shades of grey is transformed into the binary one. Then, on the second step, a set of polar attributes is computed in polar coordinate system for fixed number of angles. The attributes represent length of radius-vector that describes contour of the selected object. Invariant to the scale and object rotation attributes named in the paper as frequency polar attributes are calculated as amplitude spectrum of discrete Fourier transform taken from the set of polar attributes. It is shown that for quite informative object description the number of components of one-dimensional vector of frequency polar attributes isn-t large and is limited by the range 20-30. Properties of attribute vector invariance are retained for digital images on which the object dimensions are much greater than one pixel. Efficiency of object representation by the set of frequency polar attributes is illustrated by an example of solving the problem of search and detection of the specified objects on radar and optical panoramic images.