Due to the limitations of a physical memory, it is quite difficult to analyze and process big datasets. The Hadoop MapReduce algorithm has been widely used to process and mine such large sets of data using the Map and Reduce functions. The main contribution of this paper is to implement MapReduce programming algorithm to analyze large set of fingerprint images which cannot be normally processed due to a limited physical memory in order to find the features of these images at once. At first, the images are maintained in an image data store in order to be preprocessed and to extract the features for the biometric trait of each user, and then store them in a database. The algorithm preprocesses and extracts the features (ridges and bifurcation) from multiple fingerprint images at the same time. The extracted points are detected using the Crossing Number (CN) concept based on the proposed algorithm. It is validated using data taken from the National Institute of Standards and Technology’s (NIST) Special Database 4. The data consist of fingerprint images for many users. Our experiments on these large set of fingerprint images shows a significant reducing in the processing time to a nearly half when extracting the features of these images using our proposed MapReduce approach.

The rapid development of Smart Cities and the Internet of Thinks (IoT) is largely dependent on data obtained through Wireless Sensor Networks (WSNs). The quality of data gathered from sensor nodes is influenced by abnormalities that happen due to different reasons including, malicious attacks, sensor malfunction or noise related to communication channel. Accordingly, outlier detection is an essential procedure to ensure the quality of data derived from WSNs. In the modern utilizations of WSNs, especially in online applications, the high detection rate for abnormal data is closely correlated with the time required to detect these data. This work presents an investigation of different outlier detection techniques and compares their performance in terms of accuracy, true positive rate, false positive rate, and the required detection time. The investigated algorithms include Particle Swarm Optimization (PSO), Deferential Evolution (DE), One Class Support Vector Machine (OCSVM), K-means clustering, combination of Contourlet Transform and OCSVM (CT-OCSVM), and combination of Discrete Wavelet Transform and OCSVM (DWT-OCSVM). Real datasets gathered from a WSN configured in a local lab are used for testing the techniques. Different types and values of outliers have been imposed in these datasets to accommodate the comparison requirements. The results show that there are some differences in the accuracy, detection rate, and false positive rate of the outlier detections, except K-means clustering which failed to detect outlier in some cases. The required detection time for both PSO and DE is very long as compared with the other techniques meanwhile, the CT-OCSVM and DWT-OCSVM required short time and also they can achieve high performance. On the other hand CT and DWT technique has the ability to compress its used dataset where in this paper, CT can extract much less number of coefficients as compared DWT. This makes CT-OCSVM more efficient to be utilized in detecting outliers in WSNS.