Abstract
Combining cellular automata (CA) with convolutional neural networks (CNN) has been overtime an important topic of scientific interest in the field of computer vision. This has been due to the high number of common aspects between the two models, such as extracting information and making decision based on pixel neighbourhoods. While there are many recent papers on the subject, they are mainly focusing on the visual aspects of the problem and ignore the use of numerical evaluations for the obtained results. In this articles we present a numerical evaluation of a neural cellular automaton (NCA) with the task of reconstructing images. For this evaluation we employed several image quality metrics and the use of multiple training algorithms for the presented model. As such, we determined that cellular automata could be used as a method of restoring images, starting from a damaged state of the image, with a high accuracy. We also concluded that the algorithm presents certain weaknesses when it comes to images with very few details in them, where almost the entirety of the image is predominated by one colour.
Citare
@Inproceedings{Loghin2023ImageRU,
author = {Mihai-Adrian Loghin and A. Andreica},
booktitle = {Hybrid Artificial Intelligence Systems},
title = {Image Reconstruction Using Cellular Automata and Neural Networks},
year = {2023}
}