In the thesis, we explored how computer vision can be used to automatically extract depth information from artistic images. Two methods were developed, which use different computer vision principles to attempt to reconstruct three-dimensional space in artworks. The first method is based on the detection of faces and mathematical properties of perspective. The second method utilizes the advanced MiDaS model to generate depth images from artworks. The analysis was carried out on 10,484 images from the WikiArt collection. An analysis of the results was also made using various unsupervised machine learning algorithms. Comparison of both methods showed that the second method, with its many advantages, is better.
|
