Fingerprint recognition technology is widely used as a kind of powerful and effective authentication method on various mobile devices. However, most mobile devices use small-area fingerprint scanners, and these fingerprint scanners can only obtain a part of the user’s fingerprint information. Besides, traditional fingerprint recognition algorithms excessively rely on the details of fingerprints, and their recognition performance has great limitations in mobile devices which can only get partial fingerprint images due to fingerprint scanners. This paper proposes a partial fingerprint recognition algorithm based on deep learning for the recognition of partial fingerprint images. It can improve the structure of convolutional neural networks, use two kinds of loss functions for network training and feature extraction and finally improve the recognition performance of partial fingerprint images. The experimental results show that the fingerprint recognition algorithm in this paper has a better performance than the existing fingerprint recognition algorithm based on deep learning on the problem of partial fingerprint classification and fingerprint recognition in the public dataset NIST-DB4 and self-built dataset NCUT-FR.

Research on partial fingerprint recognition algorithm based on deep learning

Multimodal biometric system can be accomplished at different levels of fusion and achieve higher recognition performance than the unimodal system. This paper concerned to study the performance of different classification techniques and fusion rules in the context of unimodal and multimodal biometric systems based on the electrocardiogram (ECG) and fingerprint. The experiments are conducted on ECG and fingerprint databases to evaluate the performance of the proposed biometric systems. MIT-BIH database is utilized for ECG, FVC2004 database is utilized for the fingerprint, and further experiments are being performed to evaluate the proposed multimodal system with 47 subjects from virtual multimodal database. The performance of the proposed unimodal and multimodal biometric systems is measured using receiver operating characteristic (ROC) curve, AUC (area under the ROC curve), sensitivity, specificity, efficiency, standard error of the mean, and likelihood ratio. The findings indicate AUC up to 0.985 for sequential multimodal system, and up to 0.956 for parallel multimodal system, as compared to the unimodal systems that achieved AUC up to 0.951, and 0.866, for the ECG and fingerprint biometrics, respectively. The overall performance of the proposed multimodal systems is better than that of the unimodal systems based on different classifiers and different fusion levels and rules.

Multimodal biometric systems based on different fusion levels of ECG and fingerprint using different classifiers

Fingerprint images in crime scene are important clues to solve serial cases. In this paper we present a complete crime scene fingerprint identification system using deep machine learning with Convolutional Neural Network (CNN). Images are acquired from crime scene using methods ranging from precision photography to complex physical and chemical processing techniques and saved as the database. The images collected from the crime scene are usually incomplete and hence difficult to categorize. Suitable enhancement methods are required for pre-processing the fingerprint images. Minutiae are extracted from the fingerprint images. The features of preprocessed data are fed into the CNN as input to train and test the network. The experimental results demonstrated on database using Open CV-Python shows high accuracy of 80% recognition an partial or full fingerprints in the criminal database.

Fingerprint Image Identification for Crime Detection

espanolInspiradas en la forma en que el cerebro procesa la informacion, las redes neuronales artificiales (RNA) se crearon con el objetivo de reproducir habilidades humanas en tareas que son dificiles de resolver utilizando la programacion algoritmica clasica. El paradigma de las RNA se ha aplicado a numerosos campos de la ciencia y la ingenieria gracias a su capacidad de aprender de ejemplos, la adaptacion, el paralelismo y la tolerancia a fallas. El reservoir computing (RC), basado en el uso de una red neuronal recurrente (RNR) aleatoria como nucleo de procesamiento, es un modelo de gran alcance muy adecuado para procesar series temporales. Las realizaciones en hardware de las RNA son cruciales para aprovechar las propiedades paralelas de estos modelos, las cuales favorecen una mayor velocidad y fiabilidad. Por otro lado, las redes neuronales en hardware (RNH) pueden ofrecer ventajas apreciables en terminos de consumo energetico y coste. Los dispositivos compactos de bajo coste implementando RNH son utiles para apoyar o reemplazar al software en aplicaciones en tiempo real, como el control, monitorizacion medica, robotica y redes de sensores. Sin embargo, la realizacion en hardware de RNA con un numero elevado de neuronas, como en el caso del RC, es una tarea dificil debido a la gran cantidad de recursos exigidos por las operaciones involucradas. A pesar de los posibles beneficios de los circuitos digitales en hardware para realizar un procesamiento neuronal basado en RC, la mayoria de las implementaciones se realizan en software mediante procesadores convencionales. En esta tesis, propongo y analizo varias metodologias para la implementacion digital de sistemas RC utilizando un numero limitado de recursos hardware. Los disenos de la red neuronal se describen en detalle tanto para una implementacion convencional como para los distintos metodos alternativos. Se discuten las ventajas e inconvenientes de las diversas tecnicas con respecto a la precision, velocidad de calculo y area requerida. Finalmente, las implementaciones propuestas se aplican a resolver diferentes problemas practicos de ingenieria. catalaInspirades en la forma en que el cervell processa la informacio, les xarxes neuronals artificials (XNA) es crearen amb l’objectiu de reproduir habilitats humanes en tasques que son dificils de resoldre mitjancant la programacio algoritmica classica. El paradigma de les XNA s’ha aplicat a nombrosos camps de la ciencia i enginyeria gracies a la seva capacitat d’aprendre dels exemples, l’adaptacio, el paral·lelisme i la tolerancia a fallades. El reservoir computing (RC), basat en l’us d’una xarxa neuronal recurrent (XNR) aleatoria com a nucli de processament, es un model de gran abast molt adequat per processar series temporals. Les realitzacions en maquinari de les XNA son crucials per aprofitar les propietats paral·leles d’aquests models, les quals afavoreixen una major velocitat i fiabilitat. D’altra banda, les xarxes neuronals en maquinari (XNM) poden oferir avantatges apreciables en termes de consum energetic i cost. Els dispositius compactes de baix cost implementant XNM son utils per donar suport o reemplacar el programari en aplicacions en temps real, com ara de control, supervisio medica, robotica i xarxes de sensors. No obstant aixo, la realitzacio en maquinari de XNA amb un nombre elevat de neurones, com al cas de l’RC, es una tasca dificil a causa de la gran quantitat de recursos exigits per les operacions involucrades. Tot i els possibles beneficis dels circuits digitals en maquinari per realitzar un processament neuronal basat en RC, la majoria d’implementacions es realitzen en programari usant processadors convencionals. En aquesta tesi, proposo i analitzo diverses metodologies per a la implementacio digital de sistemes RC fent us d’un nombre limitat de recursos de maquinari. Els dissenys de la xarxa neuronal es descriuen en detall tant per a una implementacio convencional com per als distints metodes alternatius. Es discuteixen els avantatges i inconvenients de les diferents tecniques pel que fa a l’exactitud, velocitat de calcul i area requerida. Finalment, les implementacions proposades s’apliquen a resoldre diferents problemes practics d’enginyeria. EnglishInspired by the way the brain processes information, artificial neural networks (ANNs) were created with the aim of reproducing human capabilities in tasks that are hard to solve using the classical algorithmic programming. The ANN paradigma has been applied to numerous fields of science and engineering thanks to its ability to learn from examples, adaptation, parallelism and fault-tolerance. Reservoir computing (RC), based on the use of a random recurrent neural network (RNN) as processing core, is a powerful model that is highly suited to time-series processing. Hardware realizations of ANNs are crucial to exploit the parallel properties of these models, which favor higher speed and reliability. On the other hand, hardware neural networks (HNNs) may offer appreciable advantages in terms of power consumption and cost. Low-cost compact devices implementing HNNs are useful to suport or replace software in real-time applications, such as control, medical monitoring, robotics and sensor networks. However, the hardware realization of ANNs with large neuron counts, such as in RC, is a challenging task due to the large resource requirement of the involved operations. Despite the potential benefits of hardware digital circuits to perform RC-based neural processing, most implementations are realized in software using sequential processors. In this thesis, I propose and analyze several methodologies for the digital implementation of RC systems using limited hardware resources. The neural network design is described in detail for both a conventional implementation and the diverse alternative approaches. The advantages and shortcomings of the various techniques regarding the accuracy, computation speed and required silicon area are discussed. Finally, the proposed approaches are applied to solve different real-life engineering problems.

Methodologies for hardware implementation of reservoir computing systems

In this paper, we present the development and evaluation of a low-cost RGB-D camera based anthropometric measurement system for Barong Tagalog tailoring shops. Although anthropometric scanners exist in the market today, most of these current systems have high cost and complex installation process. The proposed barong anthropometric measurement system utilizes only frontal and side-view images from off-the-shelf RGB-D sensor, i.e. Kinect camera, which is inexpensive and readily available. The proposed system measures the linear body distances, i.e. arms, shoulder, and upper body lengths based on RGB-D skeleton tracking joints. We also propose three different methods for estimating the circumferential barong measurements for the hips, chest, and waist. The proposed system was evaluated by comparing the measured and actual values for absolute and relative error metrics in every measurements from 200 randomly selected participants – 100 males and 100 females. The average relative errors (%) for the male arm, body, shoulder, chest, waist, and hip girth measurements were 4.09, 3.06, 2.80, 4.51, 4.14, and 3.05, respectively, while generally higher for females, 3.93, 6.53, 5.19, 5.23, 4.37, and 5.27, respectively. The acquired body measurements were utilized to produce barong tailoring patterns and assign standard barong sizes, i.e. Extra Small (XS), Small (S), Medium (M), Large (L), and Extra Large (XL) sizes, to 130 male participants. Notably, only one participant was misallocated by the proposed system as shown in the confusion matrix results. Finally, a website was developed as a proof-of concept for e-commerce potential. Overall, the results illustrated the feasibility of the proposed system.

RGB-D Camera Based Anthropometric Measuring System for Barong Tagalog Tailoring

Traditional pattern recognition systems were implemented using neural network systems in order to recognize partial prints ranging from 100%, 75%, 50%, 40% and 30% of the whole fingerprint image. The uniqueness of the patterns of each individual fingerprint served as the motivation in pursuing this research in identifying partial fingerprints where the minutiae pattern identification method cannot be used due to incomplete information arising from the partiality of the fingerprint. Promising results are achieved during experimentations. This system is also able to identify partial prints subjected to incorporated noise and at most 10% blank spots within the prints. This partial fingerprint recognition system is intended to support forensic science particularly during crime scene investigations where full images of the fingerprints are impossible to locate.

Joan Del Espiritu

Papers

Neural network based partial fingerprint recognition as support for forensics