Relative Iris Codes

Abstract

This paper proposes a new scheme to generate iris codes based on relative measure of local iris texture. The local characteristic of iris texture is analyzed using 2D Gabor wavelets. Twelve Gabor kernels, four frequencies and three orientations, are constructed and convoluted with an iris image. To inherit relationship of local iris texture among pixels, Gabor magnitude and phase of a reference pixel is compared with Gabor magnitudes and phases of the other four pixels. These pixels are located away from the reference pixel by 8timesd pixels, where d=1, 2, ..., 4. Each comparison, a 2-bit primitive iris code is generated. Least significant bit of the primitive code describes how Gabor magnitudes of the two pixels are related. This bit is set to '1' if Gabor magnitude of a reference pixel is less than magnitude of the other pixel, otherwise it is set to '0'. Another bit of the 2-bit primitive code describes relative measure of the obtained phase values. This bit is set to '1' if difference of the obtained phases is within plusmnpi/2 , otherwise it is set to '0'. In our scheme, each pixel is described using an 8-bit iris code. Matching between two iris codes is implemented using a look-up table technique. The table contains a number of matches of the primitive code of the two iris codes. By utilizing the look-up table technique, computational time of our 1:1 matching scheme is 2.2 milliseconds. Equal- Error-Rate (EER) of the proposed system using CASIA1.0 iris database is 0.0003%EER