Person-independent facial expression analysis by fusing multiscale cell features
Date of Issue2013
School of Electrical and Electronic Engineering
Automatic facial expression recognition is an interesting and challenging task. To achieve satisfactory accuracy, deriving a robust facial representation is especially important. A novel appearance-based feature, the multiscale cell local intensity increasing patterns (MC-LIIP), to represent facial images and conduct person-independent facial expression analysis is presented. The LIIP uses a decimal number to encode the texture or intensity distribution around each pixel via pixel-to-pixel intensity comparison. To boost noise resistance, MC-LIIP carries out comparison computation on the average values of scalable cells instead of individual pixels. The facial descriptor fuses region-based histograms of MC-LIIP features from various scales, so as to encode not only textural microstructures but also the macrostructures of facial images. Finally, a support vector machine classifier is applied for expression recognition. Experimental results on the CK+ and Karolinska directed emotional faces databases show the superiority of the proposed method.
DRNTU::Engineering::Electrical and electronic engineering
© 2013 SPIE. This paper was published in Optical Engineering and is made available as an electronic reprint (preprint) with permission of SPIE. The paper can be found at the following official DOI: [http://dx.doi.org/10.1117/1.OE.52.3.037201]. One print or electronic copy may be made for personal use only. Systematic or multiple reproduction, distribution to multiple locations via electronic or other means, duplication of any material in this paper for a fee or for commercial purposes, or modification of the content of the paper is prohibited and is subject to penalties under law.