报告人：刘海霞（香港科技大学 博士后）

时间：2017年11月9日9:00-9:50

地点：创新研究院803

摘要:  

Art authentication is the identification of genuine paintings by famous artists from the forgeries. In this paper, we introduce a novel curvature-based method to authenticate van Gogh paintings. We use curvature images to capture the shape information in the paintings. For each painting, we convert it from RGB to HSI color space. The features we propose are two simple statistics of the three parts: including (i) the H, S, I color information, (ii) their corresponding first order derivatives in x, y directions, and (iii) the corresponding 2D curvature images. In order to select the appropriate features for art authentication, we use a forward stage-wise feature selection method such that van Gogh paintings are highly concentrated and forgeries are spread around as outliers. Numerical results show that our method gives the 88.61% classification accuracy, which outperforms the state-of-the-art methods for art authentication so far.

Joint work with Prof. Xue-Cheng Tai

报告人简介:

Haixia Liu is a postdoc at Department of Mathematics, The Hong University of Science and Technology. Before that, she received her Ph.D degree from The Chinese University of Hong Kong. Liu had her Bachelor’s degree from Hefei University of Technology and the master’s degree from University of Science and Technology of China. Her research interests mainly focus on designing algorithms to solve problems in data sciences (data analysis, machine learning, stylometry (style analysis)) using numerical linear algebra, optimization (convex or nonconvex), probability and statistics.