... local approaches, all of the training local patchesare not necessarily in the same class, and their eigenspace tends not to be obtainedaccurately. In addition, Kanemura et al. proposed a framework ... class such as face images and fingerprint images.However, since the global approach requires the assumption that all of the training imagesare in the same class, it is difficult to apply it to arbitrary ... all of the training images should be globally similar, i.e.,they should represent a similar class of objects [7, 23, 24]. Therefore, the global approach issuitable for images of a particular...