The recognition of two-dimensional images is studied using neuronetwork and conventional classifiers invariant with respect to rotation, scaling, and shift of the image. Zernike and pseudo-Zemike moments as well as direct images are used as classification attributes. The learning of the neuronetwork classifiers employs various algorithms. The effect of the noise level, sampling of the image rotation angle, and the number of neurons in the hidden layer on the recognition quality is investigated. The results of the numerical experiment are used for a comparative analysis of the characteristics of image classifiers based on various principles.
展开▼
