局所的に定義された印象変換ベクトルによる顔画像の印象変換の試み

摘要

A facial image is mathematically represented by a high-dimensional vector by separating the shape and texture information of the pattern. Variations in appearance of the faces are then coded in a small number of parameters indicating coordinates of the face space obtained by principal component analysis (PCA) of a set of high-dimensional vectors, and thus a morphable model of the face pattern is defined in terms of these parameters. In the impression transfer vector method, subjective impression of an input face is manipulated by changing the coordinates in the face space along the direction of the impression transfer vector which, thus far, was uniquely defined based on the Fisher's linear discriminant function. In this study, however, a new method was proposed to obtain the impression transfer vector which is independently defined by location in the face space by means of Support Vector Machine (SVM). The new method was tested on manipulating gender impression of input faces, and potential effectiveness of the method was confirmed by subjective assessments conducted in terms of the method of paired comparison.%顔画像は形状とテクスチャを分離した多次元ベクトルによって数学的に表現される.そして，これらの多次元ベクトルの集合に主成分分析（PCA）を施すことによって，顔の見かけの多様性を顔空間の位置を表す低次元のパラメータで記述できる顔のモーフィングモデルが構築される.印象変換ベクトル法は，このパラメータを印象変換ベクトルの方向に変位させることによって顔画像の印象変換を行う方法であり，本研究では従来のFisher線形判別関数の代わりにSupport Vector Machine（SVM）を応用して，印象変換ベクトルを顔空間内で局所的に定義する方法を提案しその有効性の検討を行った.具体的には顔画像に対して男女の性差の印象についての印象変換を行い，一対比較法による主観評定でその効果を調べたところ，提案手法の有効性を示唆する結果が得られた.