Recognition of binary images of the objects suffected to affine transformations

Application of pattern recognition and, in particular, a system of automatic processing of images of objects, allows for identification and prediction of environment condition and the object of study. To reduce the amount of information processed, detectionit is convenient to conduct recognition of binary images. The most of images to be recognized comprise complex of affine transformations. This reduces the efficiency of identification due to distortion of indications Also, the number of comparisons is increased significantly.
As an affine transformation we consider rotation of an image of side of the object around two axis passing through the center of the object image, with a change of scale to the center.

Research is performed on the basis of the system of three equations obtained by the authors which show the dependence of variation of the moments of inertia of the image when it is resized to the center along Y-axis of M times and N times along the Z- axis with the rotation of the axes of the image around the X-axis by the angle p.

By solving this system of equations on the basic of semigraphical method the values of M, N and p. Then turn identified carried image on the corners with the adjustment of the size of the values of M and N, as well as a direct comparison of the image obtained with the standard. Thus, after the preliminary calculations of various provisions of the bust recognizable image is limited to several provisions. If identifiable image corresponds to the standard, after conversion it will be the same as the standard.

In order to confirm the theoretical results of the computer modeling of the proposed method of detection. Experiments were carried out on four images are simplified silhouettes of leaves of trees: oak, aspen, maple and willow. The simulation results showed that only the appropriate reference image identified by the smallest discrepancy is within a certain allowable threshold differences. For other images, it is much higher.

Thus, using the proposed method of pattern recognition, the identification process can be performed on the model, while retaining sufficient reliability.