Summary: | 碩士 === 國立高雄應用科技大學 === 光電與通訊研究所 === 98 === The singular points, core and delta, are widely used in fingerprint classification and identification. In this thesis, a new singular point detection algorithm based on adaptive image enhancement, compact boundary segmentation, and 2-D non-separable wavelet extrema for localization is proposed. We perform adaptive correction for low contrast image to replace the intensity information of a given image matrix with another equalized intensity matrix of the image. With the help of binary thresholding, the impression-of-interest (IOI) region is segmented via the landmarks detected by searching and starting from both sides of the image and framed with the connected polygons. Following the orientation filed estimation and smoothing and the computation of Poincare Index, the fingerprint IOI is aligned per the singular points and the orientation filed nearby. Finally, the location of the detected core is further refined by searching the neighboring curve with the largest curvature based on Henry system. The proposed framework has been tested on the FVC2002 DB1 and DB2 databases. For the 800 images in the each database set, DB1 correctly detection rate reaches 89.5% with core false alarm rate 0.4% in average, while DB2 correctly detection rate reaches 88% with core false alarm rate 0.43% in average. The detection rates are compared favorably to the state-of-the-art singular point detectors.
|