The methodology behind the Art Recognition Engine relies upon training artificial neuronal networks in a deep learning setup. We combine statistical methods with explicit feature engineering techniques, by integrating a large number of learning units to increase the performance and accuracy.
The training process rests on the commonly accepted hypothesis that the brushstroke pattern constitute a unique finger print of the artist. The feature which needs to be learned is thus the brushstroke of a particular artist.
Once the learning process has been completed, the algorithm can recognize with very high precision the details of the brushstroke of that artist. When analyzing a new artwork, the brushstroke is compared with the learned one: if the brushstrokes are identical within a very small margin, the artwork is labeled as original, otherwise it is a fake.