Personal facial expression space based on multidimensional scaling for the recognition improvement

There have been many different studies on the recognition of facial expressions so far. Most of those studies have focused on the robust recognition of facial expressions of different people. However, people make different facial motions for the same facial expression spatially and temporally. Therefore, our purpose is to study personal facial expressions more closely. In this paper, multidimensional scaling is used to derive the trajectory of a face image sequence on the proposed personal facial expression space. The multidimensional scaling maps normalized face images as points in the lowest dimensions preserving their relative positions in terms of Euclidean distances in real time. To implement the multidimensional scaling method for the facial image sequence which includes facial gesture, the estimation and compensation of 3-D head motion is employed using a 3-D face wireframe. Experimental results show that the proposed method is useful in the recognition of facial expressions as it focuses on both temporal and spatial changes of personal facial expression.