Introduction to statistical pattern recognition (2nd ed.)
Introduction to statistical pattern recognition (2nd ed.)
Journal of Cognitive Neuroscience
| Hi-index | 0.00 |
This paper proposes a simple but powerful method for automatic classification of facial expressions from static images. Still images do not bare as much information as in video sequences which have much information I activities during the expression actions. The main aim here is to be able to classify every facial image into the six universally researched and accepted prototypic facial expressions like happy, fear, disgust, anger, surprise and sadness. We propose a powerful matching algorithm to determine the correct grouping or class of the newly received image into the already learned data in the database. Principal Component Analysis (PCA), a powerful tool in the feature extraction and dimensionality reduction is utilized to generate Eigen-components of the images. The Six representatives (Principal) Eigen-components of the individual facial expressions classes are stored and used at the time of matching. We use Mahalanobis distance algorithm to determine the similarity between the newly unknown incoming data and the already trained and known data sets in our database. This similarity is calculated between one-to-many (six) classes. The individual matching rate is recorded and compared for the evaluation and proper classification. Falsely accepted data (FAR) and the ones rejected falsely (FRR) are studied for algorithms improvement. Homomorphic filtering, one of the image enhancing methods in the frequency domain is used for pre-processing the image. With proper initial image processing and noise removal techniques used, a high proper classification is envisioned. The algorithm has worked perfectly with our local data and part of JAFFE data base. More tests are yet to be carried out on other public databases.