Assessment model for perceived visual complexity of painting images.

In the construction of the knowledge system, visual perception is the primary means of acquiring knowledge. Thus, it is very essential to solve the problems related to visual perception. Visual complexity, as a basic aspect of visual perception, is extremely important for human being to understand and perceive the visual stimuli. This leads to an interesting question: what factors affect visual complexity of images and how to evaluate the visual complexity objectively. In order to address this issue, we take digital painting images as the visual stimuli. We firstly conduct an experiment to collect the subjective complexity labels of painting images and then identify the factors that affect visual complexity perception. Three main factors that affect human visual complexity perception are identified, namely, distribution of compositions, colors, and contents. Secondly, we study theoretical and empirical concepts from psychology and art theory to design 29 global, local, and salient region features which represent the above three factors. Moreover, we provide two ways to estimate the visual complexity of painting images. One is to evaluate the visual complexity level of painting images by classifying the complexity level into three levels (low, middle and high complexity). Another one is to predict a complexity value for painting images by a regression model. The experimental results indicate that the proposed classification method (by Random Forest classifiers) can predict the visual complexity perception of paintings with an accuracy of 86.78%. By the comparisons, the proposed method outperforms other measurements of image complexity with a higher correlation coefficient between subjective complexity and objective measures of complexity. Furthermore, we apply the regression model of visual complexity to predict the other features of painting images. The results show that the regression model has a good ability of measuring aesthetic quality, beauty, and liking of color of the painting images involved in JenAesthetics dataset.