A Clustering-Based Approach for the Extraction of ROI from Fingerprint Images

Fingerprint-based verification systems require a certain amount of pre-processing on the fingerprint images before they can be applied. The complete fingerprint image is usually never used during authentication. A specific region of interest (ROI) is extracted for the feature extraction, which is then used for matching. In this paper, the ROI is extracted using a clustering-based approach. The entire fingerprint is first segmented into blocks; then, several features are extracted from each block using a Sobel filter. These features are clustered based on similarity, after which an agglomerative clustering algorithm combines similar clusters and separates dissimilar clusters leading to an accurate ROI. When used in a fingerprint recognition pipeline, the ROI extracted improves the matching accuracy significantly. The extracted ROI will always contain a core point if it exists in the initial fingerprint. A generalized algorithm is proposed to find the ROI consistently on fingerprints while invariant to translation and rotation.