An image, such as a fingerprint, can be decomposed into a linear combination of polar shapelet-base functions. This publication describes a fingerprint-matching algorithm that uses polar shapelet-base functions. Using polar shapelet-base functions, a fingerprint image block can be separated into components with explicit rotational symmetries. Polar shapelet-base functions can represent the fingerprint image through compact parameterization or encoder representation due to their interpretation in terms of the rotational angle πœƒ, and due to their separability by a distance r. Therefore, the use of polar shapelets enables a convenient and robust method to perform fingerprint image manipulation, analysis, and matching. Polar shapelet-base functions are special types of steerable filters that have rotational symmetry. When the fingerprint image is convolved with a polar shapelet-base function, the magnitude of the convolution output is rotationally invariant, and the relative rotation between two fingerprint images is the phase shift in the convolution output, which enables calculating the rotation angle between two matching image blocks with relative ease. Polar shapelet-base functions can be utilized to create a machine-learned (ML) model that is composed of harmonic and rotationally symmetric convolution filters. The fingerprint-matching algorithm pre-specifies the rotation order of each filter, but the size and shape of the convolution filter is optimized using the ML model. Also, the fingerprint-matching algorithm optimizes a TensorFlow implementation for each convolution filter in the radial direction r. The ML model determines an optimized set of filters that can increase the matching between two rotated images. The described fingerprint-matching algorithm offers high-resolution fingerprint images, low computation latency, low image energy residuals, and high matching rates.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.