The 256-Byte Iris Code

The 2-D Gabor filters used for iris recognition are defined in the doubly dimensionless polar coordinate system (r, ) as follows:

Both the real and imaginary members of such quadrature filters are employed, so the resulting image projections are complex. The real parts of the 2-D Gabor filters are slightly adjusted through truncation to give them zero volume, and hence no DC response, so that computed iris code bits do not depend upon strength of illumination. (The imaginary parts of the filters inherently have no DC response because of odd symmetry.) The parameters and co-vary in inverse proportion to to generate a self-similar, multi-scale wavelet family of 2-D frequency-selective quadrature filters with constant logarithmic bandwidth, whose locations, specified by _o and r_o, range across the zones of analysis of the iris.

Each bit h in an iris code can be regarded as a coordinate of one of the four vertices of a logical unit square in the complex plane. It is computed by evaluating, at one scale of analysis, the sign of both the real and imaginary parts of the quadrature image projections from a local region of the iris image I(,) onto a particular complex 2-D Gabor filter:

Thus a single complex 2-D Gabor filter (13), having a particular set of size and position parameters ( r_o, _o, , , ) in the dimensionless iris domain (r,) , performs a coarse phase quantization of the local texture signal by approximating it as one vertex (h_Re,h_Im) of the logical unit square associated with this filter through conditionals (14) - (17). The time required for computing a complete iris code of 2,048 such paired bits (256 bytes) on a RISC-based CPU, once an iris has been located within the image, is about one-tenth of a second (100 msec) with optimized integer code.