Latent Fingerprint Segmentation Based on Sparse Representation

A target image–oriented dictionary learning–based method for fully automated latent fingerprint forensic

“…With the human intervention involved, for latent fingerprint identification, the semiautomatic mode rather than full-automatic mode is adopted. [7][8][9] A semi-automated latent fingerprint identification procedure consists of the following 4 stages: (i) ROIs in latent images are manually labeled; (ii) on the basis of the labeled ROI, the features such as minutiae, singularity, ridge quality map, orientation field, ridge wavelength map, and skeleton are manually extracted 10 ; (iii) the marked features are uploaded to a latent fingerprint matcher then are automatically matched against the features derived from the rolled/plain fingerprints in background database; and (iv) according to the matching scores, the candidate rolled/plain prints are retrieved, and the candidates are visually verified by latent examiners. After visual verification, the most possible archived fingerprint in background database might be found.…”

“…Under the condition of fully automated minutiae extraction, the effect of ROI is evaluated depending on the benchmark, where no ROI mask is adopted but the whole query latent image is used for automated minutiae extraction. Therefore, in Equations (6) and (7), not the manually marked genuine minutiae #{MS 1 } but the automatically extracted genuine minutiae #{MS 1 ∩ MS 2 } and #{MS 2 − MS 1 ∩ MS 2 } are used as the baseline. For Minutiae Extraction Scenario 1, MS 1 − MS 1 ∩ MS 2 stands for the missing genuine minutiae, even the entire latent image is imported into the automated minutiae extractor; MS 1 ∩ MS 2 represents the genuine minutiae, which are correctly detected by the computer program; and MS 2 − MS 1 ∩ MS 2 are the spurious minutiae falsely extracted by the computer program.…”

“…For efficiency, fully automated latent fingerprint processing is desirable. With the human intervention involved, for latent fingerprint identification, the semiautomatic mode rather than full‐automatic mode is adopted …”