Abstract
Face perception studies support a holistic/part-based model of face perception, where holistic and part-based processing make parallel and separable contributions to face perception [1]. However, it remains unclear which exact aspects are included in holistic processing. Many studies have concluded that facial expressions are processed holistically [e.g. 2,3,4], but most of them have used static stimuli, opposed to the fact that natural facial expressions are highly dynamical. Using a novel technique to control the dynamics of individual face parts separately, we studied the integration of dynamic facial components in expression recognition. METHODS. From a video database with natural dynamic facial expressions we constructed a morphable model that permits to control the emotional style and timing of different face parts separately. The technique exploits active appearance models [5] that were trained separately on different parts of the face, so that shape variations (e.g. of the eye or mouth) could be characterized by low dimensional vectors. In a classification experiment, by morphing between neutral and emotionally expressive parts, we determined the recognition thresholds for face parts and the holistic perception of the whole face. RESULTS AND CONCLUSION. The thresholds for the perception of facial expressions were equal for holistic faces and combinations of spatial parts, as long as the eyes were present (all p's > 0.14). In a quantitative analysis we did not find indications of a holistic processing of dynamic emotional expressions. References: [1] Pieper et al., Front Psychol, 3, 2012 [2] White, Am J Psychol, 112(3), 1999 [3] Calder et al. J Exp Psychol Hum Percept Perform, 26(2), 2000 [4] Tanaka et al., Cogn Emot, 26(6) , 2012 [5] Cootes et al., IEEE Trans Pattern Anal Mach Intell, 1998.
Meeting abstract presented at VSS 2014