Improving Automated Latent Fingerprint Identification Using Extended Minutia Types

TLDR: In this article, the authors proposed a method that exploits extended fingerprint features (unusual/rare minutiae) not commonly considered in AFIS to improve rank identification accuracies.

About: This article is published in Information Fusion.The article was published on 2019-10-01 and is currently open access. It has received 22 citations till now. The article focuses on the topics: Minutiae & Automated fingerprint identification.

Figures

Figure 7: CMC curve showing consistent improvement in the rank identification accuracies for NIST-Bozorth3, VeriFinger-SDK and MCC-SDK when the fused scores are further modified based on fitting error threshold. Left column show the parameter optimization by looking for optimal threshold. Right column show the improvement in rank identification after applying optimal threshold. 24

Table 8: Rank-1 identification (in %) for the 3 considered baseline fingerprint matchers before and after applying the proposed improvement based on rare features, on the full set of 268 latent-tenprint pairs included in GCDB database.

Figure 5: Probability density estimate of the fitting errors for match and non-match comparisons.

Table 2: Extended features defined by CDEFFS categorized into respective fingerprint feature details (not a comprehensive list).

Figure 6: CMC curve showing consistent improvement in the rank identification accuracies for NIST-Bozorth3, VeriFinger-SDK and MCC-SDK when their minutiae-based similarity scores are fused with fitting error based on the rare minutiae proposed in our algorithm.

Table 5: The probability of occurrence of minutia types present in the 268 latent fingerprints of GCDB (total number of minutiae observed = 3376). The numbers correspond to minutia types in Figure 3

Frequently asked questions

Q1. What have the authors contributed in "Improving automated latent fingerprint identification using extended minutia types" ?

In this work, the authors explore ways to improve rank identification accuracies of AFIS when only a partial latent fingerprint is available. Towards solving this challenge, the authors propose a method that exploits extended fingerprint features ( unusual/rare minutiae ) not commonly considered in AFIS. The authors report significant improvements in the rank identification accuracies when these minutiae matchers are augmented with their proposed algorithm based on rare minutiae features. 

Q2. What have the authors stated for future works in "Improving automated latent fingerprint identification using extended minutia types" ?

Also, future work may exploit the differences between minutiae types to further475 improve the performance of standard AFIS, e. g., by incorporating weighting factors in the alignment or matching steps [ 46 ] when combining those AFIS with auxiliary approaches like the one presented in this paper. 

Q3. How can the authors improve the rank identification accuracies of minutiae?

Developing more robust automatic extraction of rare minutiae can significantly improve the current state of the art in AFIS adapted for latent fingerprints. 

Q4. Why is the use of EFS in manual comparisons much debated?

The use of EFS by forensic examiners in manual comparisons is much debated, mainly due to non-repeatability by another examiner to validate the previous decision. 

Q5. What are the original minutia sets used in this study?

The original latent minutia sets provided by Guardia Civil and the postprocessed VeriFinger generated minutia sets are used in all their experiments. 

Q6. What is the way to determine the distance between the minutiae?

32016Step 6: After the alignment, all those minutia pairs which are within a threshold distance are considered to be mated pairs, and a one-to-one correspondence is established between them. 

Q7. What is the definition of latent fingerprints?

Fingerprints left at a crime scene, referred to as latent prints, are the most common type of forensic science evidence and have been used in criminal investigations for more than 100 years, but comparing latent fingerprints is not an easy task. 

Q8. How many minutiae types were reported for full fingerprints?

In related works, minutiae frequency for 20 different minutiae types were reported for full fingerprints obtained from a population of 200 Spanish individuals [36] and 278 Argentinian individuals [37]. 

Q9. How many latent fingerprint features were correctly matched?

In [15], it is reported that though the highest measured accuracy achieved by a individual matcher at Rank-1 was 71.4%, and approximately 82% of the latents were correctly matched at Rank-1 when multiple matchers were combined. 

Q10. how is the affine transformation of the minutiae set calculated?

Step 3: To establish the correspondence between latent and tenprint minutia points, the authors choose the minutia points from M that are close to the minutia points of L. The Euclidean distance is calculated between the minutia pairs todetermine whether the pairs are close or not.