学术文献库

In the traditional art pricing models, the variables that capture the painting's content are often missing. Recent research starts to apply the computer graphic techniques to extract the information from the painting content. Most of the research concentrates on the reading of the color information from the painting images and analyzes how different color compositions can affect the sales prices of paintings. This paper takes a different approach, and tries to abstract away from the interpretation of the content information, while only focus on measuring the quantity of information contained. We extend the concept of Shannon entropy in information theory to the painting's scenario, and suggest using the variances of a painting's composing elements, i.e., line, color, value, shape/form and space, to measure the amount of information in the painting. These measures are calculated at the pixel level based on a picture's digital image. We include them into the traditional hedonic regression model to test their significance based on the auction samples from two famous artists (Picasso and Renoir). We find that all the variance measurements can significantly explain the sales price either at 1% or 5% level. The adjusted R square is also increased by more than ten percent. Our method greatly improves the traditional pricing models, and may also find applications in other areas such as art valuation and authentication.