IDENTIFICAÇÃO DE ASSINATURAS EM QUADROS POR MEIO DE CARACTERÍSTICAS SIFT E CLASSIFICAÇÃO COM KNN

Resumo

DOI: 10.5281/zenodo.10732940

With the emergence, popularization and ease of consumption of generative Artificial Intelligences – AIs capable of generating content that does not yet exist, be it written, visual, among others – it has become necessary to create tools that can combat plagiarism and misuse of intellectual property. An example of this problem is a case that became extremely popular in which an image created by the generative AI Midjourney won a digital art competition (Oliveira, Pedro, 2022), raising several debates about it.

This article aims to analyze the images of signatures, in order to reach a conclusion on the topic: is it possible to distinguish the handwriting of a human and an AI?
To this end, the following techniques were used: SIFT (Scale-Invariant Feature Transform) to extract characteristics that distinguish the two types of signatures. The K-Means algorithm to group the data that will be transcribed into a histogram that will be consumed by the K-Neareast Neighbors (KNN) algorithm that will classify the data obtained. 

Using a base of original and artificially generated images of the painter's handwriting, Claude Monet - from the Impressionist art movement - this research aims to develop and conclude the possibility of differentiating a human handwriting from that generated by an AI algorithm based on the signature present in the work, through Machine Learning algorithm technologies.