Joint Distributions based on DFB and Gaussian Mixtures for Evaluation of Style Similarity among Paintings

In this paper, we study the ability of joint statistical information of directional subbands in evaluating style similarity among Chinese ink paintings by employing the Gaussian mixture models with different mixture components. The optimal number of mixture components can be automatically learned from the training features by pruning the mixture models. Two types of Gaussian mixture models are built on two different sets of features: one is based on the high-order statistical moments of directional subbands; the other one is based on the parameters of generalized Gaussian density (GGD) of the marginal distributions of directional subbands. The experimental results show that the accuracy of the model based on the parameters of GGD is better than that of the model based on the high-order statistical moments