High-accuracy fingerprint matching from poor-quality fingerprint images is important and the most challenging problem. This paper describes a new poor-quality fingerprint verification technique combining pattern matching and minutiae matching composed of the following four parts. The first part is local FFT band-pass that passes only the dominant frequencies’ components representing the periodicity of ridge patterns and in each local area and raises the norms of all frequencies’ components to the power of a specified number. The second step is generation of a mask image that extracts fingerprint region by segmenting background region and part below the second joint of finger. The third step is pattern matching that is optimal block-wise shift using gradient distribution of input and enrolled fingerprints for absorbing rotation and translation. Finally, the fourth step is minutiae matching that are extraction of endpoints and bifurcations of fingerprint ridges called minutiae and one-to-one correspondence determination between input and enrolled minutiae. Experimental results using the public FVC2004 fingerprint image database demonstrate a fairly low equal error rate (EER) of 12.91% for false rejection and false acceptance as compared to those obtained by academic groups that participated in the FVC2004.

109 情報科学部・情報科学研究科 ＞ 学位論文 ＞ 修士論文