Artwork classification based on image features

V tej diplomski nalogi želimo preizkusiti metodo, ki nam omogoči, da s pomočjo računalniške analize sliko pripišemo določenemu slikarju. Testiramo dva načina. Pri prvem pristopu želimo identificirati slikarja glede na način, s katerim preslika človeške obrazne poteze iz fotografije na naslikan portret. Zanima nas, ali so razlike v obraznih razmerjih na fotografiji in sliki statistično pomembne. Pri drugi metodi vsako sliko opišemo z vektorjem značilnic. Značilnice obsegajo barvo, teksturo in dimenzije slike, katerih kombinacija tvori vektor značilnic. Princip testiramo na 3 slikarjih z različnimi stili. Za vsakega od njih imamo množico desetih testnih slik. Zanima nas, ali lahko z gručenjem sliko pravilno pripišemo slikarju samo na podlagi teh vektorjev značilnic.In this thesis we are trying to discover a method that allows us to attribute a painting to a particular artist with the help of image analysis. We are testing two methods. In the first one, we are trying to identify the style of a painter by analysing the way in which he translates a human face from a photograph into a painting. We are testing whether the differences on facial proportions in photographs and paintings are statistically significant. With the other method, we describe every painting with a set of features. The features look at the image color, texture and dimensions to form a feature vector. We test this on 10 pictures for each of the 3 painters with different styles. We are trying to test, whether we can correctly attribute these paintings to a painter just with these feature vectors