Biomedical Image Processing / Medical Image Processing
Amin Mohammadian; Hasan Aghaeinia; Farzad Towhidkhah
Volume 6, Issue 3 , June 2012, , Pages 207-218
Abstract
In this paper, a method is proposed based on the prior knowledge from a new subject to improve the performance of person-independent facial expression recognition. First, in order to obtain a basic system, a combination of geometric features and texture descriptor is compared with global features (i.e., ...
Read More
In this paper, a method is proposed based on the prior knowledge from a new subject to improve the performance of person-independent facial expression recognition. First, in order to obtain a basic system, a combination of geometric features and texture descriptor is compared with global features (i.e., mapped face images using the Kernel-PCA and raw data of face images). The results of comparison under noisy conditions were investigated and evaluated by person-dependent/independent cross-validation method. The obtained basic system was evaluated by leave-one-subject-out cross-validation. Since the same subjects are not introduced in both training and test phases, the basic recognition system is person-independent and its performance is substantially lower than that of person-dependent cross-validation case. To improve the performance of the basic system, a method is proposed in which virtual samples are generated based on the prior knowledge from the new subject and are used in learning process. The results show that the recognition rate increases up to 96% for the person-dependent basic system, kernel-PCA method is more sensitive than the others to interpersonal variability, and the recognition rate is significantly (P<0.05) improved up to 91.39% compared to that of person-independent case.
