2012 年 20 巻 3 号 p. 102-107
Face images taken by a digital camera were identified by use of the Mahalanobis-Taguchi system. Because of the high pixel count of the images, the multilevel MT system was used. The images were mosaicked by treating the average of multiple pixels as a single pixel, and the amount of data was further reduced by double processing by the MT system. Effectiveness of identification of face images taken from different camera angles was also examined. Even with background present, identification proved possible unless the faces were disguised in a way that concealed most of the face. When the unit space was defined from a face image photographed from the front, it was found that identification was possible with camera angles up to about 45 degrees. When the definition range of the unit space was extended, identification accuracy became lower, differing from the result in a previous report. This indicated that the unit space must be defined on the basis of a study of the variance within the unit space and distance from the signal data.