Kornelije Juric

• The detection of anomalies is one of the key problems occurring for example in commercial quality control applications. This work explores the potentials of a novel machine learning approach referred to as Reconstruction by Inpainting for Visual Anomaly Detection (RIAD), based on encoder-decoder architecture and image inpainting techniques. The approach is applied to the task of detecting anomalies on banknote images such as stains and scribbles while having to cope with inherent banknote print variations. Using a dataset consisting of 50 Euro banknotes, rigorous experimentation is conducted to evaluate the efficacy of this approach and explore simpler and faster solutions. This study aims to offer practical solutions for automating the assessment of banknote fitness, with potential applications in improving the efficiency of currency processing in ATMs and money-counting machines.