Expression intensity measurement from facial images by self organizing maps

Facial expression recognition and inferring emotion from an expression is a challenging task. Many methods have been proposed to recognize facial expressions, but the more challenging task ldquofacial expression intensity classificationrdquo remains less focused. Here we propose a system that is able to provide an estimation of facial expression intensity from facial images. At first each image of these sequences are normalized and cropped based on a fixed template. Then, features are captured from Gabor wavelet transformation of these images followed by principle component analysis (PCA). Finally, self organizing maps (SOM) are applied to determine the intensity of emotion from these principle components. In this work we propose a heuristic; MDC (minimum distance criterion) that is able to provide a quantitative measurement about the goodness of a combination of PCs from the intensity measurement point of view. Moreover, we propose a method to represent the results of SOM in the form of membership functions to visualize the qualitative performance.