摘要 |
PROBLEM TO BE SOLVED: To provide an object recognizing apparatus robust to a blurred image. SOLUTION: Object area images of an object to be recognized are detected from a degraded image degraded by a motion blur or a focus blur. The object area images are converted into frequency areas, respectively, and a feature vector indicating the amount of blur is extracted. A plurality of blurred images generated by applying each PSF (point spread function) in a plurality of PSFs to each of a plurality of training images without blur are formed as a cluster, and each cluster is stored in association with each PSF. A feature vector is compared with a plurality of blurred images belonging to each cluster to find the cluster most similar to the feature vector, and one PSF associated with the most similar cluster is selected. COPYRIGHT: (C)2009,JPO&INPIT
|