Study On The Effect Of Image Quality For Visual Art Analysis And Painting Style Classification

The development of computational aesthetics makes that using calculation methods to analyze the painting style more and more deeply possible, then we can have a in-depth appreciation of paintings. But, in the study of painting style analysis, researchers often require the used paintings are high-resolution. As everyone knows, it is difficult to obtain high-resolution paintings. Now, we can easily obtain the paintings we needed from the internet, and they can be downloaded for free. But, in order to reduce the amount of image storage, easy to transport, the websites decrease the resolution of paintings in the process of digitalization. Therefore, the comparative study of high and low resolution paintings on one hand can judge the representativeness of features extracted from low-resolution paintings, get that which feature has the best representativeness, and on the other hand, can also provide help for the selection of paintings in the further study.Firstly, sparse coding, information theory, and Weka are summarized and introduced in this paper. Then, three experiments are carried out based on sparse coding.1). Basis functions are trained from high and low resolution paintings, and features are extracted from frequency domain and Gabor domain of the trained basis functions. Then the representativeness of features extracted between high and low resolution paintings are compared.2). The normalized mutual information (NMI) is figured out using the better performance feature to compare the representativeness of features extracted between high and low resolution paintings.3). Detectors trained by high-resolution paintings can be successfully used in style classification. Finally, features extracted by information theory are used to classify the painting style through Weka.Researches show that, basis functions of low-resolution paintings are like those of high-resolution paintings that can represent the artists’brush characteristic. And low-resolution paintings still have the representativeness of artists’paintings style. Compared with the high-resolution paintings, low-resolution paintings have a good sensitivity to the painting style, and can be well used to judge the painting style, at the same time. In the classification of the painting style based on the features extracted by information theory, we separate the same and the different painting style paintings successfully with Weka. The study has a meaningful guiding function for selecting paintings and features in the future study. This also brings a new vision in the study of paintings.