This paper presents a filter-based fingerprint matching algorithm that uses a bank of Gabor filters to extract a compact, fixed-length representation called FingerCode. The algorithm captures both local and global details of a fingerprint image, and matching is based on the Euclidean distance between FingerCodes. The proposed method achieves verification accuracy that is only marginally inferior to the best results of minutiae-based algorithms. It performs better than a state-of-the-art minutiae-based system when the application does not require a very low false acceptance rate. The method also shows that combining decisions from complementary fingerprint information (minutiae-based and filter-based) can improve matching performance. The algorithm first determines a reference point in the fingerprint image, then tessellates the region of interest around this point into sectors. Each sector is filtered using a bank of Gabor filters in eight different directions. The feature vector, or FingerCode, is derived from the average absolute deviation of gray values in each sector. The FingerCode is compact, requiring only 640 (or 896) bytes of storage, and is suitable for matching and storage on a smartcard. The algorithm is tested on two databases: MSU_DBI and NIST 9. On the MSU_DBI database, the system achieves a verification accuracy that is only slightly lower than the best minutiae-based results. On the NIST 9 database, similar performance is achieved. The system is also compared with a state-of-the-art minutiae-based matcher, and the filter-based approach is shown to perform better in certain scenarios. The algorithm is also evaluated using receiver operating characteristic (ROC) curves, which show the trade-off between false acceptance rate (FAR) and false rejection rate (FRR). The filter-based matcher performs well on both databases, and combining results from multiple matchers (minutiae-based and filter-based) further improves performance. The paper concludes that the filter-based approach is a promising alternative to minutiae-based methods for fingerprint verification.This paper presents a filter-based fingerprint matching algorithm that uses a bank of Gabor filters to extract a compact, fixed-length representation called FingerCode. The algorithm captures both local and global details of a fingerprint image, and matching is based on the Euclidean distance between FingerCodes. The proposed method achieves verification accuracy that is only marginally inferior to the best results of minutiae-based algorithms. It performs better than a state-of-the-art minutiae-based system when the application does not require a very low false acceptance rate. The method also shows that combining decisions from complementary fingerprint information (minutiae-based and filter-based) can improve matching performance. The algorithm first determines a reference point in the fingerprint image, then tessellates the region of interest around this point into sectors. Each sector is filtered using a bank of Gabor filters in eight different directions. The feature vector, or FingerCode, is derived from the average absolute deviation of gray values in each sector. The FingerCode is compact, requiring only 640 (or 896) bytes of storage, and is suitable for matching and storage on a smartcard. The algorithm is tested on two databases: MSU_DBI and NIST 9. On the MSU_DBI database, the system achieves a verification accuracy that is only slightly lower than the best minutiae-based results. On the NIST 9 database, similar performance is achieved. The system is also compared with a state-of-the-art minutiae-based matcher, and the filter-based approach is shown to perform better in certain scenarios. The algorithm is also evaluated using receiver operating characteristic (ROC) curves, which show the trade-off between false acceptance rate (FAR) and false rejection rate (FRR). The filter-based matcher performs well on both databases, and combining results from multiple matchers (minutiae-based and filter-based) further improves performance. The paper concludes that the filter-based approach is a promising alternative to minutiae-based methods for fingerprint verification.