PSYCHOLOGICAL EVALUATION AND THE APPLICABILITY OF THE IMPRESSION TRANSFER VECTOR METHOD FOR SYNTHESIZING HIGHER-ORDER FACIAL IMPRESSIONS Yuiko Sakuta (Waseda University), Hanae Ishi (Miyagi National College of Technology), Shigeru Akamatsu (Hosei University), and Jiro Gyoba (Tohoku University) Faculty of Human Sciences Waseda University Tokorozawa, Japan E-mail: y.sakuta@aoni.waseda.jp ABSTRACT We developed a facial image generating technique that can manipulate facial impressions. The present study applied this impression transferring method to higher-order impressions such as “elegance” or “attractiveness” and confirmed the psychological validity of this method using the semantic differential method. Subsequently, we applied this method to two types of cognitive experiments. First, we examined the contributions of texture and shape on the facial impressions by using those face images for which the impressions have already been quantitatively manipulated based on this method. Second, we used such stimuli to examine the effect of facial impressions and attractiveness on the “mere exposure effect.” Thus, we concluded that the impression transfer vector method is an effective tool to quantitatively manipulate the facial impressions in various cognitive studies. Keywords: Principal component analysis, semantic differential method, face, higher-order impressions, impression transfer vector 1. INTRODUCTION The human face plays an important role in interpersonal communication. When we look at a person’s face, we readily perceive whether or not we know the person; further, we also estimate gender, age, personality, or physical attractiveness. Moreover, people perceive or form subjective facial impressions such as “elegant or inelegant” and “masculine or feminine.” In our research, we define impressions as “subjective or affective images perceived from various objects that reflect the viewer’s feelings, past experiences, knowledge, or an affective evaluation”. Some impressions are simple and directly perceived from physical information, such as gender or age. Others are more subjective, such as attractiveness. The latter are considered to be a higher level of impressions than the former. In this paper, we focused on higher-level impressions such as attractiveness or elegance and manipulated the physical properties that contribute to them. The impressions were measured by using the semantic differential method [1]. This method was originally developed by Osgood and colleagues to measure the affective meanings of words. Recently, this method has been broadly employed for quantitatively measuring various impressions from different stimuli such as face, color, music, aroma, and so on. Osgood posited that semantic differential data can generally be summarized into the following three primary factors: “Activity,” “Potency,” and “Evaluation.” In many studies, these factors have been confirmed among many materials and by people belonging to various cultures or countries. 1.1 Purpose of this study In the current research, we manipulated facial impressions by using a newly suggested image processing method based on the principal component analysis (PCA); this method is referred to as the impression transfer vector method [2][3]. We then evaluated the psychological validity by using the semantic differential method. In particular, we focused on the impressions related to the “Evaluation” factor. Since Evaluation typically contains affective appraisals or reflections of values, such as “beautiful or ugly” or “elegant or inelegant,” it represents integrative higher-order impressions more than the other factors do. Investigating relationships between higher-order impressions, such as “elegant” or “likable,” and physical features as a mathematical model is useful for creating systems that can recognize and generate affective information regarding faces in order to develop, for example, user-friendly and attractive human interfaces. The basic concepts of the impression transfer vector method and its application to higher-order impressions will be described in the first half of this paper. Thereafter, two experiments that apply the impression transfer vector method to the cognitive studies have been presented. 1.2 Impression transfer vector method based on the PCA Kobayashi and colleagues generated impression- transferred images by using the impression transfer vector method [2][3]. In this method, facial patterns are first represented by multiple high-dimensional vectors that separate the shape and texture information of the face. The shape vectors indicate the x and y coordinates of the corresponding feature points defined on the face, and the texture vectors indicate the gray level values of the corresponding pixels of the pattern. We built a separated shape and texture morphable model, which describes the variations of faces with diverse attributes in terms of a small number of shape and texture parameters, by applying the PCA independently to the sets of the shape and texture vectors [4]. Thus, shape/texture vector X m (m=1,2, …, M)