The variation of the directional properties of an image is used to extract the iris code for human identification. In order to conserve the original information while minimizing the effect of noise, scale-space filtering is applied. Resulting binary codes have been tested on a set of 272 iris images obtained from 18 persons.