Abstract: Usually, Sparse Coding can be used to extract image characteristics for analyzing the style of visual art works, while researchers always don't do deep data mining on the trained basis function. The basic functions that could reflect the interior characteristics of a stylish image were trained based on Sparse Coding on art works. The spatial and higher-order characteristic statistics were figured out. Van Gogh's art works of different periods were analyzed through normalized mutual information computing using trained basis function's Gabor transform power, in order to find the diversity of style. The simulation results show that data mining on basis function can digitalize the intuitive feeling for basis function, and can distinguish the art styles of different works to a certain extent, and finally can provide reference for the criticism of art works.

Key words: visual art style, Sparse Coding, basis function, Van Gogh, normalized mutual information, characteristic analysis