Modelling of head movement in expression of disgust

Head movement modelling can be seen as a part of facial expression study because some expressions like disgust involves head movement. Head movement information can be acquired by video recording process. The recording process has to deal with image distortion correctable via plumb-line method. Unfortunately the linear fitting used in plumb-line requires piecewise function. The thesis aims to enhance the plumb-line-based image distortion correction using conic function coefficient evaluation replacing linear fitting. Experiments conducted shows that the proposed method handles various line orientations without having to rely on piecewise function. Besides distortion correction, an approach for expression movement tracking is needed. Optical flow-template matching is one of the techniques used for tracking. However, existing search algorithms did not discuss much on the looping technique of template matching. Moreover, tracking transient features during expression requires special process as the feature exists intermittently. The thesis aims to enhance the optical flow-template matching-based tracking method for tracking feature points during head movement by controlling the search loop and introducing anchoring to handle transient components. Experiment showed that the proposed method recorded a reduction in comparison of 40.1% over another similar method during worse case scenario. Besides reduction, the proposed method also lowered the lost point during searching when compared with existing method. Head movement modelling is not given proper attention in facial expression study hence affecting head model believability in computer graphics. The thesis aims to design head movement quantification method for head movement during disgust expression. The quantification method tracks movements of the head inclusive of the neck and named as ‘Dual Pivot Head Tracking’ (DPHT). To prove that it is perceptually better to use the proposed method, a perceptual study of expression with and without head movement was conducted. Results showed that subjects perceived disgust expression better if the proposed method is used ( -score of neck given head=14.9 vs. head given neck=3.59). To further support our proposal on the need to track head movement inclusive of the neck, experiments tracking subjects depicting disgust were conducted. A statistical two-tailed test to evaluate the existence of neck motion during head movement was done. Furthermore, visual comparison was made with a model without head movement approach. Results showed that neck motion was presence during head movement of disgust (z-score = 3.4 with p-value = 0.0006). Similarly the visual depictions showed that without the head movement inclusive of neck the rendering seemed to be incomplete. Having movement information, the thesis aims to design a temporal model of head movement during disgust expression. Neck motion, a part of head motion, plays a role during disgust expression. The thesis proposes spline-based function named Joint Cubic Bezier (JCB) to model neck motion during disgust. Experiments showed that using JCB, analysis and synthesis of neck motion during disgust expression is better than via cosine and exponential approach with angular separation score of JCB=0.986041, Exponential=0.897163 and Cosine=0.90773.

PDF FSKTM_2010_9_F.pdf Download (478kB)

Actions (login required)

View Item