Optimized wavelets for fingerprint compression

We present opt imized b iorthogonal and orthonormal wavelets for embedded zerotree wavelet compression of fingerprint images. These were obtained by simulated annealing over the wavelet filter coefficients using rms error between original and recovered images as the cost to be minimized. Linear-phase biorthogonal wavelets optimized for fingerprint image compression provide significantly improved fidelity compared either with standard wavelets or with wavelets optimized upon general images. These results are confirmed by psychovisual evaluation by two fingerprint experts whose independent rankings were in complete agreement. Filter coefficients for optimized orthonormal and linear-phase biorthogonal wavelets are tabulated. Most large police forces use Automated Fingerprint Identification Systems (AFIS) to match fingerprints in order to identify individuals during criminal investigations. In these systems, fingerprint image compression is essential because AFIS data bases may contain several million fingerprint images. As a class of images, fingerprints have special properties: at every point in the image there is a well-defined frequency and orientation of information determined by the local orientation and spacing of dermal ridges [1,2]. The Discrete Wavelet Transform (DWT) is used widely in image analysis and coding [3-5]. The compact support of the basis functions of the DWT implies an ability to adapt to local image structures. The U.S. Federal Bureau of Investigation has specified a wavelet method for use in its fingerprint data base [6]. The aim of the present work was to obtain specialized wavelets which take advantage of the local space-frequency structures particular to fingerprint images. Digitized versions of rolled ink fingerprint images were used at 480 x 480 resolution, 8 bits per pixel. These are losslessly transformed using a DWT. To do this, the image is decomposed into four subbands by cascading horizontal and vertical two-channel critically-sampled filter banks [5]. To obtain the next scale of wavelet components, the lowest frequency subband is further decomposed and critically sampled. The process continues until some chosen final scale is reached. Four scales were used in the present work. Shapiro [3] has presented an algorithm for image compression using 'zerotrees' of wavelet transform components. This algorithm produces an embedded coding which provides a compact representation of significance maps, i.e. binary maps indicating the locations of significant wavelet components, and allows the successful prediction of insignificant samples across scales to be efficiently represented as part of exponentially growing trees. Given an initial threshold T, a component is said to be significant with respect to T …