In this paper, we successfully apply multilevel kinoform computer-generated hologram (CGH) patterns, which are recorded on a fine texture paper by a photo printer, into the hologram identification (ID) tag system. The 11-level kinoform CGH images are generated on our tag samples by a modified simulated annealing method at a low quantization error rate of 0.4% compared with the original data and then retrieved by a commercial digital camera and reconstructed in the computer to the original data with about 4% reconstruction error rate. The low reconstruction error is accomplished by suitable pattern recognition and image processing. We also propose a kinoform reconstruction algorithm that can considerably reduce reconstruction errors occurring when the tag is damaged. It is shown that the original image of a fingerprint can be reconstructed at a low error rate of about 12%, although nearly half of the tag data are lost.