Landslide Susceptibility Mapping through Weightages Derived from Statistical Information Value Model
An Efficient Foot Ulcer Determination System for Diabetic Patients
Statistical Wavelet based Adaptive Noise Filtering Technique for MRI Modality
Real Time Sign Language: A Review
Remote Sensing Schemes Mingled with Information and Communication Technologies (ICTS) for Flood Disaster Management
FPGA Implementation of Shearlet Transform Based Invisible Image Watermarking Algorithm
A Comprehensive Study on Different Pattern Recognition Techniques
User Authentication and Identification Using NeuralNetwork
Flexible Generalized Mixture Model Cluster Analysis withElliptically-Contoured Distributions
Efficient Detection of Suspected areas in Mammographic Breast Cancer Images
Fingerprints are widely and successfully been used in a number of applications as a preferred biometric for personal identifications. However, current fingerprint authentication systems are vulnerable to direct spoof attacks at the sensor level as fake fingerprints artificially made to replicate genuine ones are now made using common materials such as silicone, gelatin, playdoh etc. This paper therefore implements a software based deep machine learning framework for classifying fingerprints images presented to the system as either been live or fake. Since typical fingerprint images are noisy, some preprocessing on the images were first of all performed using a decision based adaptive median filtering algorithm for de-noising and min-max normalization for enhancement. Features were then extracted using pre- trained Deep Convolutional Neural Network (DCNN) and their dimensionality reduced using Principal Component Analysis (PCA). Resulting features were then used to train a Support Vector Machine with Gaussian kernel optimized by Genetic Algorithm. The developed GA-SVM method was evaluated on the 3993 Biometrika datasets from the LivDet2009 database. The results obtained demonstrate robustness and effectiveness of the developed method in achieving good average liveness classification accuracy.
Skin diseases are conditions that irritate, or affect the skin causing huge impact on a person’s day-to-day life. The tight schedule of people has greatly affected their availability to routine check-ups, thus keeping them away from visiting a doctor. The reputation of web-based medical systems is slowly becoming a paradigm to help people know the severity level of a disease. Acne skin disease ranks among the most popular skin disease and upsets the sebaceous glands, hence, routine check-ups could help prevent burns. In this paper, fuzzy based method is proposed for the identification of acne skin disease. This method is proposed to overcome the shortcoming of expert systems in previous methods. From literature, expert system reasoning is associated with uncertainty. Our proposed expert system uses fuzzy rules to resolve imprecision in the expert system reasoning. According to the evaluation results from the confusion matrix, modeled for evaluating the performance of the proposed fuzzy expert scheme, it was established that the scheme got 82% accuracy, which is indicative of a good performance. The designed fuzzy expert system showed a high level of recommendation, treatment advice and suggests the severity of acne skin disease in the patient.
Lower Respiratory Tract Infection (LRTI) is a common infection among children in both tropical and subtropical regions which includes Africa, America and Asia. World Health Organization reported more than 2.5 million of deaths as a result of LRTI in 2012, late and untimely diagnosis of this infection is one of the factors responsible for its high mortality rate. This paper employed the use of machine learning techniques to diagnose the presence of LRTI in infants. The LRTI dataset obtained from Federal Medical centre (FMC) Owo in Ondo State was preprocessed and relevant attributes obtained from it as well as the whole preprocessed dataset were used to implement a Naïve bayes and K- nearest neighbor machine learning models using java programming language. The performance of the models were evaluated based on accuracy, sensitivity, specificity and precision. The result of Naïve bayes and k-nearest neighbour with all features (18) used shows 94.25% and 94.43% respectively. Naïve Bayes with information- based feature selection method shows accuracy of 99.60% while k-nearest neighbour shows 94.35% with 10 features. Also, Naïve Bayes with Correlation-based feature selection method shows accuracy of 95.40% while k-nearest neighbour shows 95.40% too with just six (6) features. The comparative results shows that Naïve bayes with information- based feature selection method performs stronger and better than others.
This design work presents a proposed replacement to the current system used by the Federal Road Safety Commission (FRSC) for checking licensed/unlicensed drivers. It gives a faster and less tedious way of identifying registered and licensed road users using biometric captures. The system employs the use of an Arduino board to control and process the functioning of other peripherals: the fingerprint scanner and the Organic Light Emitting Diode (OLED) screen connected to it to achieve its purpose. The prototype system developed was able to displays driver's information on the OLED screen (Age, Name, Sex and License ID); the average response time of the system was also calculated to be 1.41 seconds, which is a good response time considering the system in question. The tested false acceptance rate and false rejection rates were relatively low (after a sample test with 25 individuals); at 4% and 8% respectively. Also, for its implementation, the components are readily available, relatively cheap and the system is one that can be easily adopted by the FRSC if access to their already existing database is granted. Consequently, it is safe to say that the developed system measured up to the design expectations; it meets the aim of a proposed replacement for the present analogue and easy to beat system employed by the FRSC.
Content based Music Information Retrieval (MIR) has been a study matter for MIR research group since the inception of the group. Different pattern recognition paradigms are used for the diverse application for content-based music information retrieval. Music is a multidimensional phenomenon posing severe investigation tasks. Diverse tasks such as automatic music transcription, music recommendation, style identification, music classification, emotion modeling etc. requires quantitative and qualitative analysis. In spite of noteworthy efforts, the conclusions revealed shows latency over correctness achieved in different tasks. This paper covers different feature learning techniques used for music data in conventional audio pattern in different digital signal processing domains. Considering the remarkable improvements in results for applications related to speech and image processing using deep learning approach, similar efforts are attempted in the domain of music data analytics. Deep learning applied for music analytics applications are covered along with music adversaries reported. Future directions in conventional and deep learning approach with evaluation criteria for pattern recognition approaches in music analytics are explored.
