Image Processing and Communications Challenges 2

In the paper is presented the brief history of research conducted in Poland (at AGH University) in the area of image processing, analysis and recognition. The history is connected with changes of the technology used for problem solving. First systems are build from the separate integrated circuits of low and medium scale of integration, after this we develop some systems build on the base of VLSI elements, and now the new systems are constructed on the base of FPGA technology. Nevertheless all the time scientific group from AGH was conducted research in technological applications of vision systems. At 70th and 80th years of XX century it was pioneer works devoted to building of second in Poland and one of the first in Europe systems for computer processing of the images, now research is dedicated mainly toward real time image processing systems.

In this paper, we are interested by the different sides of the visual learning and the visual machine learning, as well as the development of the ”visual cognitive” evolution cycle. For this purpose, we present an expected cognitive architecture framework to highlight all the visual learning functionalities. Despite the fact that our investigations were based on the conception of a cognitive processor as a high interpreter of object recognition tasks, we strongly emphasize on a novel evolutionary pyramidal learning. Indeed, this elaborated learning approach based on association rules enables to learn highest concepts induced from concepts of lower level in order to progressively understand the highest semantic content of an input image.

In this paper an application of DirectShow software technology in a computer vision system is described. DirectShow (DS) is well suited for image processing and analysis tasks. In the reported study it was successfully applied in a computer stereovision system. Physical cameras of the system are represented by DS source filters connected to image analysis procedures. Original filter prototypes were designed for stereo disparity estimation and scene analysis tasks. Image analysis procedures for scene depth estimation were build and tested. The developed system has proved usefulness of the DirectShow technology in computer vision applications.

Computer vision needs suitable methods of shape representation and contour reconstruction. One of them, invented by the author and called method of Hurwitz-Radon Matrices (MHR), can be used in representation and reconstruction of shapes of the objects in the plane. Proposed method is based on a family of Hurwitz-Radon (HR) matrices. The matrices are skew-symmetric and possess columns composed of orthogonal vectors. Shape is represented by the set of nodes. It is shown how to create the orthogonal and discrete OHR operator and how to use it in a process of shape representation and reconstruction. MHR method is interpolating the curve point by point without using any formula or function.

In this paper the new combined video quality metric is proposed, which may be useful for the quality assessment of the compressed video files, especially transmitted using wireless channels. The proposed metric is the weighted combination of three state-of-the-art image quality metrics, which are well correlated with the subjective evaluations. A simple extension of those metrics for the video quality assessment is the averaging of their values for all video frames. Nevertheless, such approach may not lead to satisfactory results for all types of distortions. In this paper the typical distortions introduced during the wireless video transmission have been analyzed using the 160 files available as the LIVE Wireless Video Quality Assessment Database together with the results of subjective quality evaluation. Obtained results are promising and the proposed metric is superior to each of the analyzed ones in the aspect of the linear correlation with subjective scores.

Lossy compression attacks in digital watermarking are one of the major issues in digital watermarking. Cheddad et al. proposed a robust secured self-embedding method which is resistant to a certain amount of JPEG compression. Our experimental results show that the self-embedding method is resistant to JPEG compression attacks and not resistant to other lossy compression attacks such as Block Truncation Coding (BTC) and Singular Value Decomposition (SVD). Therefore we improved Cheddad et al’s. method to give better protection against BTC and SVD compression attacks.

In this paper the new method of view representation generation is proposed. View points in the method are located on spiral trajectory enlaced view sphere. Additionally the arrangement of view point is more evenly then in other known similar methods and can be formulated by only one parameter. The number of view points follows from the required scanning resolution.

This paper presents gradient based methods of color edge detection in various color spaces. First of all it briefly describes various color spaces and differences between them. Besides it shows edge detection in described spaces. Some examples of applying Sobel’s algorithm to real images in various color spaces are presented.

Vector Quantization is an efficient method for image compression. It has been developed as one of the most efficient image coding techniques. It is a process that maps the blocks of high rate digital pixel intensities into a relatively small number of symbols. The aim of this work is to use different ways to encode the homogenous/ heterogeneous or edge/smooth part of the image with the improvement of the existing Vector Quantization algorithms and reduce its complexity. Many techniques in this paper have been examined to improve the quality and the compression ratio for the compressed images, such as the block rotation process, the mean and mode operation, block classification, and random blocks selection. High PSNR results obtain when using scalar quantization as a pre processing with rand selection blocks and blocks rotation.

In this paper we present human identification method based on knuckle biometrics also termed as FKP (finger-knuckle-print). Knuckle is a part of hand, and therefore, is easily accessible, invariant to emotions and other behavioral aspects (e.g. tiredness) and most importantly is rich in texture features which usually are very distinctive. The major contribution of this paper are texture-based knuckle features and their evaluation using IIT Delhi knuckle image database.

In this paper a novel method of biometric protection of private keys is consider. This method is based on enhanced Juels and Sudan ”fuzzy vault” scheme. Introduced algorithms utilize minutiae based fingerprint data for key locking procedures. It is proposed to include derived cryptographic primitive into special ≪biometric≫ extension of X.509.v3 certificates, which are used in grid environment authentication procedures.

In this paper we propose two methods of human vein pattern segmentation from low quality images, called frequency high pass filtration and local minima analysis, witch are compared with the often used local thresholding algorithm. These methods are evaluated using manually assigned bifurcation points which is also proposed in this paper. Evaluation was carried out on 480 collected images, and shows that proposed methods are worth to consider in human vein pattern segmentation.

In this paper the authors modify their previous kNN algorithm and present a modification to improve the algorithm by considering key inner and interclass distinguishability. The suggested approach is tested on a large group of individuals with data gathered over Internet using browser-based WWW application. The obtained results are promising and encouraging for further development in this area.

In this paper Iris and Retina features are combined for recognition in biometric system. In this multimodal biometric system two biometrics can be taken from the same acquisition process and image. Gabor transform to extract the features from Iris and Retina is used. Feature fusion is performed.

In this paper we take into consideration group of decision making methods formed by the classifier fusion on the level of their discriminates . For such models we analyze what is the best way of assigning weights for them. Some analytical properties are of aforementioned methods are shown. Evaluation of proposed concept is done on the basis on computer experiment results.

The article presents different sets of Haar-like features defined for adaptive boosting (AdaBoost) algorithm for face detection. Apart from a simple set of pixel intensity differences between horizontally or vertically neighboring rectangles, the features based on rotated rectangles are considered. Additional parameter that limits the area on which the features are calculated is also introduced. The experiments carried out on the set of MIT 19×19 face and non-face examples showed the usefulness of particular types of features and their influence on generalization.

Major problem in real world scenarios is lack of suficient image samples per person in database for successful face recognition. In most cases insuficient number of samples per an individual in database is present. This makes face classification almost impossible for larger number of people. This problem is commonly described as ’one sample problem’. Recent state-of-art in face recognition allows to achieve high accuracy using face images with frontal pose. However, recognizing faces with rotations in depth, increases error rate significantly. In this paper we present a method to expand database using 3D morphable models to reconstruct 3D face from a single frontal image sample. By rotating reconstructed face to different views we create series of novel virtual images with pose variations for every individual in database. This approach can help to decrease error rate from pose variations and resolves ’one sample problem’.

The paper considers the partition problem of feature space in classification. The partition is based on information energy for fuzzy events. In this paper we use Bayes rule for classification with fuzzy observations and exact classes. Additionally a probability of misclassifications is derived for fuzzy information on object features. The results show deterioration the quality of classification when we use fuzzy information on object features instead of exact information and are compared with the partition of feature space. Numerical example concludes the work.

The paper considers automatic visual recognition of signed expressions. The proposed method is based on modeling gestures with subunits, which is similar to modeling speech by means of phonemes. To define the subunits a data-driven procedure is applied. The procedure consists in partitioning time series, extracted from video, into subsequences which form homogeneous groups. The cut points are determined by an evolutionary optimization procedure based on multicriteria quality assessment of the resulting clusters. In the paper the problem is formulated, its solution method is proposed and experimentally verified on a database of 100 Polish words.

The increasing need of providing information about the surface of complex 3D objects accounts for new approaches in shape, color and reflectivity measurements. A method integrating 3D shape and reflectance measurement has been developed, based on multispectral imaging and directional illumination. The unified data representation of objects under investigation can aid machine vision, digital documentation and expand means of realistic imaging of unique items in virtual reality. While there already exist methods of merging image based reflectance measurements with 3D data collected independently, this is a step toward integration and increase of data correspondence. Also, the proposed data processing method allows further extension, and fulfills needs of both an accurate and mobile solution.

In this paper we present a novel software framework for efficient representation and manipulations of tensors which aims in minimizing data copying. Tensors are stored in the matricized form with simultaneous abstraction superimposed on tensor indices thanks to the proxy design pattern. The proposed software pattern was then used in computation of the Higher- Order Singular Value Decomposition. Finally, the whole framework was tested in the problem of static gesture recognition.

In this paper the generalization capabilities of the Hierarchical Temporal Memory (HTM), the new computing paradigm based on cortical theory, has been exploited in order to recognize the hand shape in real images while training is done on synthetically generated data. It has been shown that the HTM trained in this way and tuned up with a small number of real examples gives pretty good recognition rates. Additionally the good scalability of the proposed solution has been observed while analyzing the recognition rates for the class that is ’almost unknown’ because only few examples are shown during training.

In this paper we investigate the recently developed dual tree complex wavelet transform (DT-CWT) and the single tree wavelet transform (ST-CWT) and compared them with Gabor wavelet transform for the face recognition problem. Experiments are carried out on standard databases. The resulting feature vectors of complex wavelets were applied to PCA and LDA for dimensionality reduction. In all experiments, complex wavelets equaled or surpassed the performance of Gabor wavelets in recognition rate when equal number of orientations and scales are used. Moreover, generally ST-CWT results outperformed DT-CWT. Obtained results indicate that complex wavelets can provide a successful alternative to Gabor wavelets for face recognition both using PCA and LDA.

The standardization method of immunohistchemically staining tissue section images prior to the image processing and analysis is described in this paper. The effectiveness of the proposed standardization method is examined on thin tissue slices of breast cancer stained with DAB & H. The image analysis results after the initial image standardization are more closer to the results of traditional methods of cells nuclei quantification than for original images.

To enable an effective treatment, the prostate cancer (PCa) must be detected early enough. Unfortunately, the diagnostic methods are insufficient. The hope for improve the PCa diagnosis lies in the perfusion computed tomography (p-CT) method. However, the p-CT prostate images are not easy to interpret.

The presented work describes the technique of computational analysis of such images using the textural features of the Haralick’s co-occurrence matrices. The research based on the material from over 50 patients concentrated on selection of proper preprocessing procedures, optimal feature space and the best decision function. A serious problem was also to choose regions of interest - especially important areas in the gland.

It seems that the improvement of detectability of PCa with the p-CT technology is possible by creating a dedicated computational system to CT scanners, that could point out the cancerous lesions automatically, faster, and more reliable than in traditional methods.

In this article we present our research into the subject of reducing the influence of noise on evaluation of perfusion parameters, such as CBF, CBV or MTT. Noise can be present on some pixels of study slices, therefore it can lead to artifacts in calculated concentration time curves and blur the final results.

To minimize influence from these factors we propose method that is different from commonly used. Generally noise reduction is done by filtering (smoothing, blurring), which is not always producing good results, as many information from image is lost. Therefore more effective is using the interpolation methods.

We have studied different interpolation techniques and compared them numerically. Tests have proven that using our method leads to better, more accurate estimation of perfusion parameters. It also seems that large window

Sinc

interpolation gives the best results.

Magnetocardiography is a promising diagnostic technique that is a magnetic equivalent of electrocardiography. This article presents a novel approach to interpretation of the sequences of magnetocardiographical images that could support diagnostics of cardiac infarctions. Algorithm presented in this paper was used to create a system that classifies to or excludes patients from cardiac infarction risk group.

An automatic left ventricle (LV) segmentation method for T2 weighted Cardiac Magnetic Resonance (CMR) image is presented. The method takes multi-slice T2 weighted CMR images from the basal to the apex of the heart. Inter-slice and intra-slice fuzzy reasoning is used to guide the centre point detection for each slice. Morphological filtering is used in the reconstruction to homogenise the blood pool region. Then radial search Fuzzy Multiscale Edge Detection (FMED) is used to segment the endocardium and the epicardium of the LV. Evaluation of the method is performed on 6 patient with approximately 42 slices of real T2 weighted MRI data. The quantitative result of the automatic method compared to those obtained from manual segmentation by a skilled clinician are very encouraging, with correlation scores of 96.2% correlation for the endocardium and 85.7% correlation for the epicardium.

Assessing muscular activity during gait in CP persons could provide valuable information in prescribing appropriate treatment to reduce the consequences of cerebral palsy as well as limiting further complication in cerebral palsy children. The main goal of this study was explored working regularities of muscle pairs in children population to show dependencies and variation on gait parameters. Functional evaluation was carried out on 20 cerebral palsy patients. The research have been done by using the system EMG. A surface electrode picked up on the main groups of muscles of lower limbs: the Rectus Femoris, the Vastus Lateralis, the Medial Hamstrings, the Lateral Gastrocnemius, and the Anterior Tibialis. There were several phases to the signal approach such as: data acquisition, data pre-processing, data modeling, data analysis and interpretation. From the results seen that for each task subjects have different strategies for keeping balance during walk depending on the basic level of muscle contraction or antagonistic and synergistic contraction required for that activity.

In this paper we study the problem of classification of wireless capsule endoscopy images (WCE). We aim at developing a computer system that would aid in medical diagnosis procedure. The goal is to automatically detect images showing pathological alterations in an 8-hour-long WCE video. We focus on three classes of pathologies, ulcers, bleedings and petechia, since they are typical for several diseases of intestines. Utilized are methods for image texture and color analysis to obtain numerical description of images. Then, three methods for selection of most discriminative descriptors are used, namely Vector Supported Convex Hull, Support Vector Machines and Radial Basis Function Networks. The results produced by the three methods are compared.

Most of the publications in the field of augmented reality in medicine concentrates on developing new types of systems but not on optimization of speed or portability of existing ones. The intention of authors of this article is to fill this gap by showing a possibility of creating augmented reality interface for visualization of volumetric medical data for Windows OS environment that can be run on off - the - shelf computer and test its capability. Tests of the authors application was performed on three 3D models generated from computer tomography data stored in collection of DICOM files. Authors novel and efficient solution can be easily attached to nearly all medical application system running Windows OS and give medical personnel support of a new level by presenting more informative and realistic three dimensional visualizations for low cost.

Computer algebra is applied to biomedical computer vision. Specifically certain biomedical images resulting from a case of rhinocerebral mucormysocis in a diabetic boy are analyzed using techniques in computational geometry and in algebraic-geometric topology. We apply convolution and deblurring via diffusion equation from the side of computational geometry and knot theory, graph theory and singular homology form the side of algebraic-geometric topology. Our strategy consists in to represent the biomedical images using algebraic structures in such way that the peculiarities of the images are represented using algebraic complexities. With our strategy we obtain an automatic procedure for the analysis and the diagnostic based on biomedical images.

This paper describes a probabilistic region-based deformable model using a new adaptive scheme for B-spline representation. The idea is to adapt the number of spline control points which are necessary to describe an object with complex shape. For this purpose, the curve segment length (CSL) is used as criterion. The proposed split and merge strategy on the spline model consists in: adding a new control point when CSL is greater than a certain splitting threshold so that the contour tracks all the concavities and, removing a control point when CSL is less to a certain merging threshold so that the contour aspect maintains its smoothness. Noise on synthetic and real weld radiographic images is assumed following Gaussian or Rayleigh distribution. The experiments carried out confirm the adequacy of this approach, especially in tracking pronounced concavities contained in images.

In this paper a fast orthogonal neural network (FONN) is used to construct an image classifier invariant to basic affine transformations (rotation, translation, scaling). The shift-invariance property of the Fourier amplitude spectrum in conjunction with the log-polar transform is applied for this purpose. Two image databases are built and used for testing the proposed classifier.

Reducing time of application execution is significant for the quality of image processing and communication systems. Automatic coarse-grained parallelization of program loops is of a great importance for multi-core computing. This paper presents Iteration Space Slicing algorithms aimed at extracting coarse grained parallelism available in arbitrarily nested parameterized loops. We demonstrate that Iteration Space Slicing permits us to generate parallel code for image analysis, encoding and communication solutions. Experimental results are carried out with UTDSP benchmark.

New computer vision solutions dedicated for blind and partially sighted people have been recently introduced as a result of significant progress in computer science. Also the growing computation power of mobile and portable devices together with development of information systems allow to adopt and apply new and robust solutions that are able to work in nearly in a real-time and share and use information spread over IP network. Many of currently developed solutions are dedicated to support the user, giving the information about divert obstacles located in the environment. However many of them are using simple detectors (commonly ultrasonic echo-location) for obstacles tracking without its classification and recognition. Therefore the solution presented in this paper engages the stereo camera and image processing algorithms to facilitate its user with object detection and recognition mechanisms. The inference engine combined together with ontology based problem modeling allows to handle the risk, predict possible user’s moves and provide the user with appropriate set of tips that will eliminate or reduce the discovered risk.

In this paper, a new neural network capable of extracting knowledge from empirical data [1]–[6] is presented. The network utilizes the idea proposed in [2] and developed in [3,4]. Two variants of the network are shown that differ in relationships describing activation functions of neurons in the network. One variant utilizes logarithmic and exponential functions as the activation ones and the other is based on reciprocal activation functions. The first network variant is similar that proposed in [3]. The difference is that in our network the logarithmic activation function works with hidden layer neurons while in [3] with input signals. In the second variant, all activation functions are of 1/

x

type. To the author’s knowledge, such a network has not been published in the literature so far. Like that of [3], our network provides a real valued symbolic relationship between input and output signals, resulting from numerical data describing the signals. The relationship is a continuous function created on the basis of a given set of input–output numerical data when learning the network. Extraction of the symbolic function expression is carried out after the training in finished. By forming the symbolic expression, the neural network structure and synaptic connection weights associated with the neurons are taken into account. The ability of knowledge extraction, also called law discovery, is a consequence of applying proper activation functions of neurons included in hidden and output layers of the network. The neural network under consideration can also play an inverse role to the above mentioned. Instead of extracting the symbolic relation, it can also be used as a neural realization of continuous functions expressed in a symbolic way. The presented theory is illustrated by an example.

Mathematical Morphology is a tool for extracting image components that are useful for representation and description. This article shows part of process of automatic Optical Music Recognition.. It suggest effective methods used to remove staff line and preparing image to symbol identification. This method based on mathematical morphology. The experimental results are showing.

In this work we present an object tracking algorithm running on GPU. The tracking is achieved by a particle filter using appearance-adaptive models. The main focus of our work is parallel computation of the particle weights. The tracker yields promising GPU/CPU speed-up. We demonstrate that the GPU implementation of the algorithm that runs with 256 particles is about 30 times faster than the CPU implementation. Practical implementation issues in the CUDA framework are discussed. The algorithm has been tested on freely available test sequences.

The paper describes a method for rectification of video in 3D television parallel multi-camera systems. Using the camera calibration data gathered, a new coordinate system is calculated, in which virtual camera positions and orientations are calculated. A rectifying perspective transform is calculated that performs transformation from the real camera coordinate system to a new system. Transformed images correspond to images from virtual rectified camera setup. The results obtained using the method described are verified by computation of depth maps using the rectified video.

In this paper a new method of cracks detection is introduced. The proposed algorithm is applied to detect the cracks in the pavement image. Local minimum and linear relation between them was proposed. The proposed method is classify into two stages: linear local minimum and verification of detecting of pavement cracking. This method is fast although is complex. Additionally, the proposed method eliminates slight and strong variations like irregularly illuminated conditions, shading and road signs painted on pavement surface.

Algorithm of eyes and iris detection for mouse cursor control is proposed. Non-invasive method analyzes images from usb camera. After eye detection, coordinates of iris position are determined. After short calibration, iris movement is converted into cursor position. Application based on OpenCV library has been created.

In this paper, we present some results of long-term measurement in the 433 MHz, wireless telemetric network dedicated to watt-hour meter reading. The RSSI of received packets was measured by the sink. The sink was equipped with a half wave, 4 dBi antenna. Received and recorded data enabled us to make analysis of PER versus RSSI from all nodes within range of the sink, simultaneously. With knowledge of how many packets sink expected to receive, during the query and how long frames were, we propose the designation of the value of FER estimator. Lack of relationship between FER and frame length prompted us to investigate errors at bits level. Both the sink and other nodes were equipped with CC1101 low power RF transceiver. The paper concludes with the areas of the results application.

In this paper we show how Google Earth can be used for visualizing network plans, in particular for FTTH networks, supporting automated GIS-based planning by reducing the need for physically visiting sites in order to inspect and verify the plans.We develop and present a simple procedure for converting Danish GIS data into the KML/KMZ formats viewable by Google Earth, and show that it produces mappings which are precise enough for the current planning purposes. With the conversion tool in place, Google Earth is easy to use, free of charge and a strong tool for presenting and reviewing network plans. Since the same formats can be used also for Google Maps and Google Maps Mobile, it is possible to view the data from handheld devices, providing a tool that is available on-site during the network deployment.

Early detection of forest fires is the primary way of minimizing their damages. Compared with the traditional techniques of forest fire detection, a wireless sensor network paradigm based on a ZigBee technique was proposed. The proposed technique is in real time, given the exigencies of forest fires. The architecture of a wireless sensor network for forest fire detection is described. The hardware circuitry of the network node is designed based on a Crossbow node. The process of data transmission is discussed in detail. Environmental parameters such as temperature and humidity in the forest region can be monitored in real time. From the information collected by the system, decisions for fire fighting or fire prevention can be made more quickly by the relevant government departments.

Currently, the broadband networks, co-financed with the European funds, are being deployed in many different regions of Poland. In previous financing period, that took place in years 2004 - 2006, the only one broadband network, namely in Kujawsko - Pomorskie region, has been successfully implemented. In the current financing period the broadband networks are intended to be implemented in some other regions of Poland. The main goal of the regional broadband networks implementation in Poland is to fulfill the gap which appeared as a lack of investment in broadband networks infrastructure.

Reliability evaluation is important task in a network design process. The GMPLS network supports physical separation of functional planes and therefore reliability evaluation becomes more difficult. The method of GMPLS network reliability evaluation is proposed in this paper. The idea of this method is based on the GMPLS as a multistate system. In order to demonstrate usefulness of this approach, proposed method is implemented and reliability evaluation for test network is preformed.

The paper compares existing algorithms of choice of LSPs in IP networks with MPLS protocol and proposes a new algorithm, which is an improved version of the Least Interference Routing Algorithm. The study showed that in dynamic conditions of the network (only short lived connections) proposed algorithm rejects a smaller number of requests of LSPs choice in relation to the compared algorithms, irrespective of the requests rate and the network topology and size.

In this paper we introduce a number of variants of modified degree 6 chordal rings, and we evaluate and compare their transmission properties in terms of average distance and diameter.We present theoretical models for calculating the distances, using optimal and ideal graphs, which are also shown to provide fairly good estimates. When comparing the distances, it turns out that the new suggestions for modification of the chordal rings results in lower distances, making them potentially interesting for use in communication networks. In the end of the paper we suggest directions for future research, in particular to investigate to what extend the topologies are suitable to implement.

In this paper we present results of evaluating and validating the three novel measurement based admission control algorithms for WiMAX dealing with connection requests that arrive in batches (ARAC, nscARAC, EMAC). We focus on the capacity of a future 4G system using non binary LDPC codes. Simulations rely solely on the use of VoIP traffic flows (with and without voice detection). In order to avoid actual FEC decoder implementation we use L2S interface described in IEEE 802.16 EMD. We use performance metrics characteristic to admission control to assess the system performance. Simulated nodes move according to the Leavy Walk distribution. All algorithms presented in the paper have been simulated using the ns2 platform with ViMACCS extension developed by authors as part of previous work.

This paper studies the difference between packet level evaluation of mobile WWW and novel image based method for objective quality of experience (QoE) assessment. Authors, based on the reference web browser architecture, show the two kinds quantitative results of measurements with mobile phone and different browsers (Opera Mini and default browser). The designed test methodology proves that there is a gap in the key performance indicators values depending on the measurement method used. Image based assessment of the WWW QoE is the right choice for mobile operators when evaluating end user satisfaction.

There are a number of techniques used in modern Location aware systems such as Received Signal Strength Indicator (RSSI), Time of Arrival (TOA), Time Difference of Arrival (TDOA) and Angle of Arrival (AOA). However the benefit of RSSI-based location positioning technologies, is the possibility to develop location estimation systems without the need for specialised hardware.

The human body contains more than 70% water which is causing changes in the RSSI measurements. It is known that the resonance frequency of the water is 2.4 GHz. Thus a human presence in an indoor environment attenuates the wireless signal. Device-free Passive (DfP) localisation is a technique to detect a person without the need for any physical devices i.e. tags or sensors. A DfP Localisation system uses the Received Signal Strength Indicator (RSSI) for monitoring and tracking changes in a Wireless Network infrastructure. The changes in the signal along with prior fingerprinting of a physical location allow identification of a person’s location. This research is focused on implementing DfP Localisation built using a Wireless Sensor Network (WSN). The aim of this paper is the evaluation of various smoothing algorithms for the RSSI recorded in a Device-free Passive (DfP) Localisation scenario in order to find an algorithm that generates the best output. The best output is referred to here as results that can help us decide if a person entered the monitored environment. The DfP scenario considered in this paper is based on monitoring the changes in the wireless communications due to the presence of a human body in the environment. Thus to have a clear image of the changes caused by human presence indoors, the wireless recordings need to be smoothed.We show results using algorithms such as five-point Triangular Smoothing Algorithm, 1-D median filter, Savittzky-Golay filter, and Kalman filter.

Anomaly detection approach is a new, emerging trend for network security especially for high-security networks (such as military or critical infrastructure monitoring networks). In our previous work we proposed a new methodology for such intrusion detection systems. We proposed new signal based algorithm for intrusion detection on the basis of the Matching Pursuit algorithm. As to our best knowledge, we are the first to use Matching Pursuit for intrusion and anomaly detection in computer networks. Hereby, we report further, more extensive, evaluation of the proposed methodology. We show results for 15 metrics characterizing network traffic (previously we tested our system using packets-per-second only). Moreover, we provided the comparison of our method with state-of-the-art DWT-based anomaly detection system and proved that our solution gives better results in terms of detection rate and false positives.