Previous [ 1] [ 2] [ 3] [ 4] [ 5] [ 6] [ 7] [ 8] [ 9] [ 10] [ 11] [ 12] [ 13] [ 14] [ 15] [ 16] [ 17] [ 18] [ 19] [ 20] [ 21] [ 22] [ 23] [ 24] [ 25]

¡@

Journal of Information Science and Engineering, Vol. 27 No. 1, pp. 337-351 (January 2011)

Multi-Layered Expression Synthesis

JIA-RU LIN AND I-CHEN LIN
Institute of Multimedia Engineering
College of Computer Science
National Chiao Tung University
Hsinchu, 300 Taiwan

In this paper, a feature-point-driven expression editing and synthesis framework is proposed. While the extensively used blend shape methods suffer detail losing during image blending, the proposed multi-layer method can retain the flexibility and variety of geometry editing and preserve detail features as well. For low-frequency sub-bands, optimization-based blend shape is presented for large-to-mid scale synthesis. In addition, statistics-based feature matching and enhancement are proposed for high-frequency details. Our results show that the proposed methods are adequate to high-resolution expression synthesis and detail-preserved image editing.

Keywords: facial expression, blend shape, image editing, steerable pyramid, detail preservation

Full Text (¥þ¤åÀÉ) Retrieve PDF document (201101_21.pdf)

Received November 17, 2008; revised March 17, 2009; accepted April 2, 2009.
Communicated by Tong-Yee Lee.
* This paper was partially supported by National Science Council, Taiwan with grant no. NSC 95-2221-E-009- 164-MY3.