Recognition of noisy monochrome images using Hopfield neural network
Abstract
Recognition of noisy monochrome images using Hopfield neural network
Incoming article date: 14.10.2024The relevance of the problem of pattern recognition lies not only in the quality of recognition - classification of images, but also in the possibility of their rapid restoration in noisy conditions. Such solutions are useful, for example, for automatic access control systems to a protected area in the case of recognition of license plates or an on-board computer when recognizing license plates in real time. It is shown that a recurrent neural network with the Hopfield architecture copes well with the recognition of simple monochrome images of small size in conditions of their noisiness. The architecture of the Hopfield neural network is given, the peculiarity of which is a small amount of memory, which determines the scope of application of the neural network of this architecture. The algorithm for training the Hopfield neural network is given. Examples and results of recognition of noisy monochrome images are demonstrated using the example of road signs. The results of the experiment on noisy images demonstrate the possibility of image restoration with less than 40% of distorted bits.
Keywords: pattern recognition, recurrent neural network, noisy monochrome image, reference sample, training