The Effect of Image Quality for Visual Art Analysis

In the study of computational aesthetics, high-resolution paintings always are used to analyze painting style, but actually the paintings we obtain mostly are low-resolution. In order to analyze the effect of image quality for visual art analysis, the contrast experiments are carried out between high and low resolution paintings in this paper. The sparse coding is used to train basis functions, different features are extracted in frequency domain and Gabor domain from the basis function. Then the normalized mutual information（NMI） is figured out to analyze the difference between high and low resolution painting features in analyzing the painting style. At last, the features with better performance are used to classify the painting style. The results show that, to a certain extent, features extracted from low-resolution paintings still have the ability to characterize the painting style, among which the Gabor energy has the best effect in painting style analysis. That is to say features extracted from low-resolution paintings can be used in painting style analysis.