Advanced Multimedia Processing Lab -- Group Member -- Your Name

Group Member

Congcong Li
PhD. Student

Personal Homepage: http://amp.ece.cmu.edu/people/congcong/
Email: congcong@andrew.cmu.edu

Office: Porter Hall B43
Lab: Porter Hall B6
Phone: 412-268-7115
Fax: 412-268-3890

Mailing Address:
Department of ECE, Carnegie Mellon University,
5000 Forbes Avenue, Pittsburgh,
PA 15213-3890

Research Focus:

Computer Vision

Human Perception

Pattern Recognition

Picture Quality Assessment

We study high level features and utilize machine learning techniques for picture quality assessment. Pictures could include photos, graph arts, human paintings, etc. We focus on human paintings at the current stage. We first assume certain perceptual factors that would distinguish between high quality picutres (like famous artists' paintings) and low quality pictures (like amateurish doodles). To measure the perceptual differences we propose, we extract semantic features from pictures, including harmonious degree, color contrast, color prevalence and so on. These features are then used for classification. Through experiments we could further validate the assumptions proposed before and decide which kind of features are helpful for the quality assessment.

Transfer Picture into Painting of a Specified Artist's Style

We propose learning-based techniques to adjust an input picture into a specified artist's style given a painting of that artist. The adjustment contain two parts: color adjustment and texture transfer. In the color adjustment part, we use works of similar styles to train in advance a Gaussian Mixture Model based on high-level semantic color features. Then we adjust the color of the input image so that its color statistics mathces those in the training set. In the texture transfer part, we first produce correspondence maps for both the input picture and the reference painting. The correspondience map is a spatial map of some corresponding quantity over the two images, which could be image intensity, blurred image intensity and so on. By relating the two correspondence maps, texture of the input image could be replaced patch-by-patch by the reference image. This course could be carried through multiple interations on multiple scales. Combining the color adjustment and texture transfer, we expect to output a painting that meets both the conditions: 1) retains as much information of the original picture as possible; 2) matches the style of the reference painting.

Conference Papers:

C. Li, G. Su, Y. Shang, Y. Li, Y. Xiang. "Face Recognition Based on Pose-Variant Image Synthesis and Multi-level Multi-feature Fusion." Proceedings of Third International Workshop on Analysis and Modeling of Faces and Gestures, pp. 261-275. In conjunction with ICCV2007.

C. Li, G. Su, Y. Shang, Y. Li. "Frontal Face Synthesis Based on Multiple Pose-Variant Images for Face Recognition." Proceedings of IEEE International Conferences on Biometrics (ICB2007), pp. 948-957.

C. Li, G. Su, K. Meng, J. Zhou. "Technology Evaluations on TH-FACE Recognition System." Proceedings of International Conference on Biometrics 2006 (ICB2006). LNCS 3832, pp. 589-597.

K. Meng, G. Su, C. Li, B. Fu, J. Zhou. "A High Performance Face Recognition System Based on a Huge Face Database." Proceedings of the Fourth International Conference on Machine Learning and Cybernetics (ICMLC 2005), pp5159-5164.