There is being tried the two-layer perceptron for its identification in classifying diversely distorted objects. The two-layer perceptron is modeled, trained and tested within MATLAB. Having increased the passes number up to 280 for training over scaled-turned-shifted images with pixel distortion, the perceptron performance is improved as for pure scaled-turned-shifted images, as well as for pixel-distorted those ones, wherein pixel-distorted images are classified excellently anyway. Eventually, as neocognitron on the same MATLAB and operating system configuration is many times slower, two-layer perceptron is asserted to be capable to substitute neocognitron in classifying diversely distorted objects, needing peculiarly longer training process and specific ratio for types of distortion, what depends on the object type.
Ключові слова: classification of diversely distorted objects, neocognitron, two-layer perceptron, monochrome image, pixel-distortion, training set, classification error percentage
Бібліографічний опис для цитування:
Романюк В. В.
Two-layer perceptron for classifying scaled-turned-shifted objects by 26 classes general totality of monochrome 60-by-80-images via training with pixel-distorted scaled-turned-shifted images / В. В. Романюк // Системи обробки інформації. — 2015. — № 7. — С. 98-107.