18. The concept of "texture". Spectral approach to the processing of texture images. Algorithm for texture analysis on the energy spectrum.

For the tasks of environmental monitoring, it is necessary to develop a system for recognition of aerospace images. Figure 1 shows a fragment of a satellite image of the area of ​​Novosibirsk

At the design stage, it is necessary to calculate and analyze

Signs of texture images that meet the following requirements:

noise immunity, invariance to scale and rotation, computational

efficiency of analysis algorithms.

Based on the analysis of texture processing methods

image developed an algorithm that meets the above

requirements. The algorithm consists of the following steps:

 calculation of the energy spectrum of the image (Fig. 2: a) 2, b) 2);

 plotting the histogram of the spectrum and determining the threshold;

 binarization of the image of the spectrum;

 contour processing (Fig. 2: a) 3, b) 3);

 contour tracking operation;

 determination of geometric moments of features;

 texture classification

The threshold for the subsequent binarization of the energy spectrum image is automatically selected from the local minima of the histogram. To recognize and classify a texture area by shape

energy spectrum is sufficient to have information about its contour. Having a full

information on the contour, it is easy to determine the area, the coordinates of the center of gravity and

morphometric features, as well as geometric moments of features. In work

The following morphometrics were used to form the feature vector.

signs: the ratio of the minimum and maximum axes of inertia,

convexity coefficient, fill factor and orientation of the maximum axis

inertia with respect to the X axis [7]. Calculation of signs on the contours of objects

allows you to eliminate the redundancy of video information and reduce the influence of interference that

increases noise immunity and speed.