Background

Human fingerprints, since long have been used as crucial evidence for criminal investigation. Advancements in technology have enabled to improve the efficiency of the scientific procedure for evidence collection and analysis. Simultaneously, the rise in the number and diversity of crimes committed by criminals has become a challenging task for intelligence agencies to convict a criminal. It has been observed that perpetrators of the crime have also changed their methods of committing a crime, and they equally exploit technological advancements. With the increased digitization, criminals are now more into hacking, phishing, malware attacks, etc.. To deal with these upcoming security threats, it became imperative to secure ourselves from these new-age threats. One such method of defending ourselves is biometrics, which relies on intrinsic physical or behavioural traits of human beings for authentication purposes. Unique physical characteristics like fingerprints, palm prints, iris, facial recognition, etc. are widely used today for solving criminal cases in today’s digital society (Singla et al., 2020). Solo or multiple traits can be used for authentication purposes. Even today, fingerprints are appreciated as highly significant and remain the most commonly accepted traits, among all, due to their uniqueness. Therefore, fingerprint recognition is widely used in the banking industry, securing areas of national interest, passport control, securing E-commerce, identifying missing children, etc.. In most of the above applications, the fingerprints are captured in a controlled environment for recognition purposes.

In real-world scenarios, the fingerprints recovered, particularly by law enforcement agencies, are unintentional and are left at crime scenes by chance. In such circumstances, latent fingerprinting is the mechanism that is available to recover the chance impression from a crime scene by legal authorities. These prints require further processing for the identification of criminals. Due to the unintentional and uncontrolled nature of these impressions, we encounter a whole lot of challenges like inefficient capturing and upliftment of fingerprints, partial prints, complex background noise, manual marking of minutiae, one-time upliftment of prints in some techniques, enhancement of poor-quality ridge, reconstruction of the incomplete image, etc.. These challenges provide a lot of scope in improving the performance of the fingerprint recognition system. Recently, India launched the world’s largest fingerprint database (i.e. Aadhaar), signifying the importance of fingerprint-based recognition even today (Singla et al., 2020; Krishna & Sudha, n.d.) The key objective of the paper is to acquaint the reader with the basic concepts of latent fingerprinting, along with some of the latest available approaches that are required for the enhancement, reconstruction, and matching of the lifted fingerprints. The research gaps and limitations are highlighted, providing scope for further improving the latent fingerprinting tools and techniques.

The primary task related to latent fingerprinting technology involves matching, reconstruction and enhancement. Matching deals with comparing the ground truth latent features with the features recovered from the sample under consideration. For matching to be performed efficiently, it is imperative to extract quality features from the sample which could be ensured by applying novel reconstruction and enhancement techniques which are discussed in the following sections.

Matching of the latent fingerprint is done using unique features which are categorized into three different levels, namely, level 1, level 2 and level 3(Fig. 1) (Jain & Feng, 2010a). Level 1 features are the most basic features that can be derived from a latent fingerprint sample like the arch, left loop, right loop, whorl, etc.. They are visible to our naked eyes and helps in visual inspection and manual matching of fingerprints. Further, we have level 2 features comprising ridge endings, bifurcations, hook, etc.. They are more sophisticated features than level 1 features. Sometimes due to poor quality of evidence, these features may not be extracted efficiently due to smudging of ridges. Hence, an appropriate reconstruction and enhancement is required to eliminate spurious features. Level 3 features are the most defining features that can help us enhance our performance enormously. They are permanent features that we can recover from a sample like pores, line-shape, scars, etc.. However, it is difficult to extract such features because of resolution constraints. Usually, a combination of the above features is used for appropriate matching results.

Fig. 1
figure 1

Different levels of features in a latent fingerprint (Krishna & Sudha, n.d.)

Full size image

The processing of latent fingerprint images follows a sequence of steps as depicted in Fig. 2. The first step is the image acquisition phase, wherein we uplift the latent fingerprint using various techniques, discussed in the Main text section of our paper. This captured image is further used in the enhancement phase in which the quality of an image is improved by noise removal, sharpening of an image, adjusting the brightness of the image, etc.. Image enhancement makes it easier to identify key features in an image. The next step is image restoration in which an image that is degraded due to blur, noise, dirt, scratches, etc. is recovered to extract accurate features from the image. Matching is the final step in which the features that are recovered from an image is matched with the ground truth using various matching techniques and algorithms.

Fig. 2
figure 2

The basic flow diagram of latent fingerprint processing (Jain & Feng, 2010a)

Full size image

Main text

Latent fingerprint upliftment approaches

Latent fingerprint upliftment from different surfaces is the first step in the processing of latent fingerprints. Since different surfaces possess different properties (texture, porosity, etc.), we require different techniques for latent fingerprint upliftment which are discussed in Table 1. This is the most vital step among all the preprocessing steps because the quality of latent prints uplifted at this stage is further used for enhancement, reconstruction and matching. If the uplifted prints are of good quality, the chances are that the results after preprocessing will be far better than if the prints are of poor quality. Further, the number of minutiae that we are able to extract from an image directly depends on the quality of prints obtained, which further affects the matching performance. To get quality results, we must be handing our evidence with the utmost care and uplift the prints with as much care as we can. In this section, we are going to discuss some of the available techniques for fingerprint upliftment.

Table 1 Various approaches for fingerprint upliftment
Full size table

Latent fingerprint enhancement approaches

After capturing the fingerprint evidence using various methods as discussed above, the next step is to enhance the image. In a real-world crime scenario, it is commonly observed that the uplifted evidence is not of good quality. So to get relevant information from the image, we need to enhance it using various approaches as discussed in Table 2.

Table 2 Available latent fingerprint enhancement approaches
Full size table

In 2021, a generative adversarial network (GAN)–based latent fingerprint enhancement model was proposed (Joshi et al., 2021). The advantage of the proposed approach is that it helps preserve the ridge structure along with the minutiae details which helps in improving the enhancement of the fingerprint sample. Further, a novel Lindeberg’s automatic scale selection method (Agarwal & Bansal, 2021) is introduced by the author. This method is based on the utility of level 3 features for the enhancement of latent fingerprints. In a recent paper by Gupta et al., it introduces a new approach for enhancement and reconstruction of using two dictionaries. First dictionary is orientation based, while another is composed using continuous phases. The ridge pattern is reconstructed using a continuous phase-based dictionary (Gupta et al., 2020). Further, the AM–FM model is used for field correction. A novel approach for enhancement using progressive generative adversarial network (GAN) is proposed in (Gupta et al., 2020). A high-quality latent fingerprint image is obtained using two stages. In the first stage, Progressive Offline Training (POT) is used, while in the second phase, the Iterative Online Testing (IOT) module is used. Next, an algorithm is proposed by Horapong et al. based on matched filter and sparse autoencoder (Horapong et al., 2020). This method is devised for poor-quality or partially missing fingerprints. The given algorithm improves on the friction ridges using the frequency domain of the latent fingerprint. Further, a conditional generative adversarial network-based latent fingerprint enhancement algorithm is proposed by (Joshi et al., 2021). The proposed latent fingerprint enhancement model preserves ridge structure including minutiae and removes structured and nonstructured background noise present in a latent fingerprint.

In 2019, a fingerprint enhancement approach was proposed by Jhansirani et al. in which a combination of total variation model and sparse representation with multi-scale patching is used. In this method, the image is divided into two components, texture and cartoon components, using the total variation (TV) model (Jhansirani & Vasanth, 2019). In this algorithm, cartoon components are removed as non-fingerprint patterns, and texture components are classified as the informative structure of small patterns. Attributes of ridge structures like ridge frequency and orientation are obtained with the help of the Gabor function. Further, using a set of distinct fingerprint pattern dictionaries are created. Enhancement and restoration of ridge structures are done using multiscale patch-based sparse representation along with the understanding of dictionaries. For matching and identification purposes, the author used the Levenberg–Marquardt algorithm (Jhansirani & Vasanth, 2019) for training the neural networks. The advantage of the proposed algorithm is that it reduces the distortion and further enhances the fingerprint pattern which leads to increasing the recognition rate. A generative adversarial network-based latent fingerprint enhancement algorithm is proposed by Joshi et al.. The main objective of the proposed approach is to boost the quality of ridge structure quality. Using this approach the ridge structures are preserved along with improving the quality of fingerprint images. The IIITD Multisensor Optical and Latent Fingerprint database (IIITD-MOLF) and the IIITD Multi-surface Latent Fingerprint database (IIITD-MSLFD) (Joshi et al., 2019a) are the datasets that are used in this paper for conducting experiments. The performance of the latent fingerprint recognition can be improved by making use of enhanced images with standard feature extraction as suggested by the author. Further, an enhancement approach was proposed by (Manickam & Devarasan, 2019) using an intuitionistic fuzzy set. For matching and enhancement purposes, the model proposed by the author requires the manual marking of the region of interest. The given approach is divided into two stages. Firstly, fingerprint contrast enhancement is done using an intuitionistic fuzzy set. Further, the level 2 features are extracted for matching purposes. The core of the given technique is based on minutia points which looks over n number of images. The matching score is calculated by the author using the Euclidean distance.

A novel approach was proposed by Manickam et al. which is based on Scale-Invariant Feature Transformation (SIFT) (Manickam et al., 2019a). The model deals with two phases. In the first phase, contrast enhancement of latent prints is done using an intuitionistic type 2 fuzzy set. In the next phase, the SIFT features are extracted which are further used for matching purposes. With the help of the Euclidean distance, the matching scores are calculated. A hybrid model is presented by (Liban & Hilles, 2018) which is a fusion of the edge directional total variation (EDTV) model and quality image enhancement with lost minutia reconstruction. The database used by the author for testing purposes is NIST SD27. The objective of the paper was to enhance input image as well as de-noise latent fingerprints. The observation made by the author is that the performance of the proposed technique is superior to good-quality latent fingerprint as compared with bad and ugly–quality images. Also, it was perceived that the matching accuracy is improved by about 30% using the given approach. The algorithm proposed by Xu et .al. in 2017 constructs minutia and ridge dictionaries (Liu et al., 2014). The prior knowledge of both ridge and minutia are utilized along with the proposed two-step multiscale patch-based sparse representation for enhancement purposes. Enhancement of ridges is done using ridge dictionaries, whereas minutia is enhanced using both the dictionaries. The main objective of the author was to overcome the limitations of the widely used Gabor function. One of the major limitations is that Gabor functions are not capable of capturing the details of bifurcation of ridges as well as endpoints. From the results, it is evident that the two-step SR algorithm exceeds the performance of SR only by using the Gabor dictionary.

The algorithm proposed by Yoon et al. is based on the reconstruction of an image using orientation guided sparse representation and a TV image decomposition model (Feng et al., 2012). The first step of the proposed approach is to disintegrate the latent image into cartoon and texture components. In the next step, computation of the reliability and orientation field of the texture image is done. In the final step, to deal with low-reliability regions, a redundant dictionary that is based on sparse representation is used iteratively to reconstruct the image. This dictionary is created using the Gabor function and local ridge orientations. The enhancement algorithm proposed by (Yoon et al., 2011) is based on a multiscale patch-based sparse representation and total variation model. Firstly, the latent fingerprint is decomposed into texture and cartoon components using a total variation model. The cartoon component is removed as structural noise because it contains most of the patterns that are not required. In the next stage, weak latent fingerprints are enhanced, with the proposed multiscale patch-based sparse representation method, which is present in texture components. Using the Gabor elementary functions, dictionaries are constructed to capture ridge structures. Good-quality latent images are reconstructed using multiscale patch-based sparse representation. The advantage of using this algorithm is that along with the removal of overlapping noise, it also helps to enhance and restore the distorted ridge structures. The algorithm proposed by the author is based on prior knowledge of latent fingerprints. A dictionary is created using good-quality reference patches. Loopy belief propagation is used for orientation field estimation. This prior knowledge helps us to reconstruct our latent fingerprint.

A robust orientation field estimation algorithm is proposed in which an image is divided into multiple image blocks using a short-time Fourier transform. Further in this approach, a set of hypothesized orientation fields are created using randomized Ransac (Chaidee et al., 2018). The author has proposed an algorithm that is used in the pre-enhancement phase to obtain better results (Cao et al., 2014). In this approach, a dictionary is created using spectral responses of the Gabor filter. This dictionary helps improve the high curved ridges. Most of the present algorithms are not able to achieve and preserve this information. The approach proposed in this paper is dictionary based. The paper aims to achieve “lights-out” latent identification systems. Background noise is removed using the total variation (TV) decomposition model (Yoon et al., 2010). Ridges are reconstructed using the dictionary which is created using good-quality patches. The author in this approach proposed a novel orientation estimation algorithm for enhancement of latent fingerprints. A commercial fingerprint SDK is used in this approach for estimation purposes. An adaptive directional total variation (ADTV) model is proposed by the author in this approach of enhancement of latent fingerprints (Zhang et al., 2013). In this approach, the latent images are divided into two layers (i.e. cartoon and texture). The latent print is present in the texture component whereas unwanted noise is present in the cartoon layer. This decomposition helps in the enhancement and segment of the latent print.

Latent fingerprint reconstruction approaches

Image reconstruction is a fundamental step in improving the quality of an image. Generally, the evidence recovered from crime scenes is of poor quality, blured, incomplete, etc.. So to extract minutiae efficiently from the evidence, it becomes essential to first reconstruct the image. Various reconstruction techniques are discussed in this section along with their comparison in Table 3.

Table 3 Available latent fingerprinting reconstruction approaches
Full size table

Wong and Lai in 2020 proposed a CCN-based method for reconstruction and enhancement of latent fingerprints. The recovery of ridge structures is done by learning from corruption and noises encountered at various stages in fingerprint processing (Wong & Lai, 2020). The CNN model consists of two streams that help in reconstruction. The enhancement of an image is improved using orientation fields. A generative adversarial network (GAN)–based data augmentation scheme to improve reconstruction is proposed by (Lee et al., 2020). In the given approach, the clean fingerprints are converted to their corresponding latent one which is augmented with an unpaired large-scale clean dataset for the reconstruction purpose. Further, a novel algorithm is proposed by (Xu et al., 2020) which uses machine learning and skeleton image features for the reconstruction of the image. Also, a new method is proposed by the author for generating more natural images using the Pix2Pix model. The work proposed by (Joshi et al., 2019b) is based on generative convolutional networks. This approach helps in predicting the gaps, holes, and missing parts of the ridge structures, as well as helps in filtering the noise from minutiae. The testing of the proposed method is done using various standard methods of feature extraction like MINDTCT followed by MCC and BOZORTH3.

A conditional generative adversarial network (cGAN) approach is given by Liu et al. which helps in the direct reconstruction of latent fingerprints (Dabouei et al., 2018). The cGAN approach has been modified by the author so that it can be used for the task of reconstruction. In order to ensure that the orientation and frequency information is used in the generation process, three additional ridge maps are created. This prevents the model from generating false minutiae as well as avert the model from filling missing areas that are large in size. To protect ID information in the course of the reconstruction process, a perpetual ID preservation approach is used. An artificially generated latent fingerprint database is used for guessing missing information. An algorithm based on dictionary-based learning and sparse coding for the latent fingerprint is proposed by (Li et al., 2018). Also, an algorithm has been proposed for the estimation of orientation fields. In the first step using the total variation model, the texture image is acquired by decomposing the latent fingerprint image. It has been observed that a great reduction in the structural noise is observed from a texture image. To estimate local ridge orientation for texture images, a multiscale sparse coding method is presented. In order to create a dictionary, good-quality fingerprint patches of multiscale are used, to get prior information. Also, sparse coding is repeatedly applied with varying patch sizes to amend the distorted and corrupted orientation fields. The advantage of using this approach is that it helps to repair corrupted orientations as well as reduce noise. This algorithm helps to preserve the details of singular regions. Further, a convolutional neural network (ConvNet)–based approach is proposed by (Cao & Jain, 2015) for estimating latent orientation field. In order to achieve it, ConvNets are trained using 128 representative orientation patterns.

The authors Zhou et al. present an analytical framework for latent fingerprints (Kaushal et al., 2016). The reconstruction approach adopted by the paper is based on a combination of two approaches (i.e. exemplar inpainting and partial differential equation). These two approaches are used for the reconstruction of distorted images. The binarization approach is used for the matching of fingerprints. In this approach, the author (Zhou et al., 2016) proposes triplets of minutiae to improve the performance of the algorithm. Author claims of improvement in the performance after the addition of new triplet features. Further performance has been improved by combining global features and triplet features. The paper (http://www.ijirset.com/upload/2017/may/269_Criminal.pdf, 2019) proposes an algorithm based on prior knowledge. In this approach, two dictionaries are created. One is based on a continuous phase patch and another is prepared using an orientation patch. For correction of orientation field, the latter of the two dictionaries is used and for the reconstruction of ridge pattern, the former is used. A model-based partial fingerprint reconstruction algorithm is proposed by the author (Zhou et al., 2013). The objective of the approach is to complete ridge information. This approach helps to reduce the index list before matching.

A fingerprint orientation model based on 2D Fourier expansions (FOMFE) is proposed in this paper (Wang et al., 2007) which is independent of prior knowledge. The biggest advantage of the proposed approach is its low computational cost and also that it can handle a very large database. This approach is very helpful in applications such as fingerprint indexing.

Latent fingerprint matching approaches

Latent fingerprint matching is the final step in the processing of our fingerprint image. At this stage, the matching between the original and the ground truth image is done using various approaches as mentioned in Table 4.

Table 4 Available latent fingerprint matching approaches
Full size table

Malemath et al. proposed a latent minutiae similarity (LMS) algorithm and clustered latent minutiae pattern (CLMP) algorithm (Deshpande et al., 2020). The former algorithm is used for solving the geometrical constraints between the pairs of nearest points around a minutia, whereas the latter one is based on the arrangement of minutia and its patterns.

The matching technique proposed by (Manickam et al., 2019b) uses Scale-Invariant Feature Transformation (SIFT) for matching and enhancement purposes. The approach comprises two stages—in the first stage, contrast enhancement is performed using type 2 fuzzy sets. In the next step, the SIFT features are extracted for further matching purposes. A deep learning-based approach is put forward by Zheng et al. for matching latent with rolled fingerprints (Ezeobiejesi & Bhanu, 2018). This approach is based on the resemblance of patches and the minutiae which are present on the consistent patches. For enhancing the learning, the deep learning network is used. The distance metric learned with a convolution neural network is used for calculating the similarity score. With the fusion of minutiae and patch similarity score, the matching score has been calculated. The Minutia Spherical Coordinate Code (MSCC)–based matching algorithm is proposed by (Lin & Kumar, 2018). This algorithm is the improvement of the Minutia Cylinder Code (MCC). Every minutia is represented by a binary vector using 288 bits. The MCC algorithm was represented using 448 or 1792 bits. The advantage of using this approach is its compact representation. A greedy alignment approach is used to restore minutiae pairs that are lost at the original stage.

A robust descriptor–based alignment algorithm is proposed by Paulino et al. which is based on the Hough transform (Ezhilmaran & Adhiyaman, 2017). Minutiae along with orientation fields are used by the author to draw a similarity between the fingerprints. Manual marking of the minutiae is performed in this algorithm due to which it is easy for application purposes. The orientation fields of latent fingerprints are reconstructed from minutiae. A novel fingerprint matching system is proposed by (Zhou et al., 2017). In the proposed approach the latent fingerprint images found at crime scenes are matched to the rolled fingerprint database of law enforcement agencies. Along with minutiae, other features like ridge wavelength map, skeleton, singularity, etc. are used to enhance the performance.

Further, a novel approach is proposed by Cao et al. in which extended features are used for improving the matching performance (Medina-Pérez et al., 2016). An automated latent fingerprint recognition system is proposed by (Zheng et al., 2015). Convolutional neural networks (ConvNets) are used for enhancing the matching performance. Fusion of rank, score and feature–based approach is proposed by (Jain & Cao, 2015) to boost the performance of the proposed approach. The approach proposed by the author (Cao et al., 2014) uses extended features like ridge quality map, ridge wavelength map, etc. along with minutiae. This system is created for matching crime scene fingerprints with rolled fingerprints. To gain insights into how performance changes with the addition of extended features, these features are added incrementally to the system. The conclusion drawn by the author is that among extended features, the most useful are singularity, ridge quality map, and ridge flow map. In this paper, a descriptor-based Hough transform algorithm is proposed (Feng, 2012). In this method, the comparison between latent prints is done after aligning the fingerprints using the proposed algorithm. One of the disadvantages of this approach is the requirement of manual markup. The approach proposed by the author is exclusively for matching partial fingerprints. In this paper, a new fingerprint feature is proposed by the author (i.e. Distinctive Ridge Point (DRP)) (Lan et al., 2019). This feature along with existing features are used for matching performance improvement. A novel algorithm is proposed in this paper (Jain & Feng, 2010b) for latent fingerprint matching. The core of the proposed algorithm is directional information. Estimation of distortion is done by merging image fields with the traditional model. This approach leads to a simple model with effective use of directional information.

The matching approach proposed in this paper (Jain & Feng, 2010b) merges manually marked minutiae with minutiae that are extracted automatically. The reconstruction is done using singular points and manually marked minutiae. Ridge frequency is used for the enhancement of latent prints. The main objective of the proposed approach is to enhance the speed of the matching system. Three filtering stages are proposed in this algorithm (Feng & Jain, 2008). Singular points, pattern type and orientation fields are utilized in this filtering system. The approach proposed in this paper fuses rank, score and features (Feng et al., 2009) to enhance the performance of the system as followed in many existing fusion-based approaches followed in image and video forensics (Kaur & Gupta, 2019). The main aim of fusion is to retrieve a high-quality fingerprint. Along with minutiae, the author proposes to use some extended features like quality maps, etc. to improve the performance of the system. An automatic fingerprint verification method is proposed by Feng et al.. Two minutiae-based descriptors are proposed by the author that are histograms of gradients and binary gradients. The false minutiae are handled using an orientation descriptor. Fusion of scores obtained from all the descriptors are done to achieve the desired performance.

.

Databases available

The fingerprint database is generally classified into three categories – rolled, plain and latent fingerprint database (Singla et al., 2020). For forensic applications, mainly rolled and latent fingerprints are used, whereas for commercial applications, plain fingerprints are used. To capture latent fingerprints, range of methods like chemical, powder or simply photography is done. Plain fingerprints are prints of our fingers taken using sensors that are mostly used as ground truth. Rolled prints, on the other hand, are obtained by simply rolling fingers from one side to another. Various databases available related to latent fingerprints are listed in Table 5 as follows—NIST27 (https://www.nist.gov/itl/iad/image-group/nist-special-database-2727a, 2019), WVU latent databases (https://databases.lib.wvu.edu/, 2019), FVC2004 databases (http://bias.csr.unibo.it/fvc2004/download.asp, 2019), IIIT latent fingerprint database (http://www.iab-rubric.org/resources/molf.html, 2019), IIITD Multi-surface Latent Fingerprint database (IIITD-MSLFD) (http://www.iab-rubric.org/resources/molf.html, 2019), IIIT Simultaneous Latent Fingerprint (SLF) database (http://www.iab-rubric.org/resources.html, 2019), Multisensor Optical and Latent Fingerprint database (Sankaran et al., 2015), Tsinghua Latent Overlapped Fingerprint database (http://ivg.au.tsinghua.edu.cn/dataset/TLOFD.php, 2019) and ELFT-EFS Public Challenge database (https://www.nist.gov/itl/iad/image-group/nist-evaluation-latent-fingerprint-technologies-extended-feature-sets-elft-efs, 2019).

Table 5 Available latent fingerprint datasets
Full size table

Research gaps and challenges

To improve the authentication results and reliability of fingerprint recognition, we need a lot of improvement at various stages like enhancement, reconstruction, and matching. Some of the major challenges encountered are as follows.

  • Even today, the marking of fingerprint features is done by an expert which opens a new sphere for improvement (i.e. automation of fingerprint marking) (Jhansirani & Vasanth, 2019).

  • The fingerprints recovered from the crime scenes are generally of very poor quality (background noise, partial prints, etc.) which requires a lot of preprocessing to get desired results (Feng et al., 2012).

  • Another major challenge is concerning the surface from which the fingerprints are uplifted. Different surfaces require different methods based on their texture, colour, porous/nonporous surface, etc.

  • Fingermark age determination is among the recent challenges that have attracted many researchers as its reliable estimation is a difficult task. Factors like environmental conditions, substrate properties, donor features, etc. influence the composition and components of the fingerprint which hinders its effective determination (Chen et al., 2021).

Conclusions

To enhance the robustness and efficiency of various security applications, there is a dire need for a novel approach for latent fingerprint recognition. Various image processing techniques can be applied at the enhancement and reconstruction phase to improve robustness and efficiency at the matching stage. Some of the recent methods are trying to utilize deep learning techniques like GAN’s to enhance the quality of fingerprint features. In addition, researchers are also trying to improve the results of latent fingerprint matching using various fusion techniques. This paper presents various aspects of latent fingerprinting which can be used to improve recognition and authentication results. Research in this domain may help us fortify ourselves from emerging digital era threats which is imperative to maintain the security and integrity of any nation.