遺伝的アルゴリズムを用いた画像中の任意方向を向いた顔の候補領域の抽出

抄録

Positioning human facial regions in images can be used for various fields such as person identification, counting of humans, security system, database and so on. In this paper, we propose a method to extract facial candidate regions in still gray images in which the directions, positions, sizes, and rotational angles on a vertical axis to the images, of human faces are unknown. This method is based on pattern matching that magnified and rotated facial templates in various directions are overlapped with an original image. To avoid a vast number of matching, we apply a genetic algorithm to this process, and to improve accuracy of extraction, we use differential images of original ones. We confirmed by computer simulations that the proposed method could extract facial candidate regions from various images.