Vol. 31 · No. 4

We give a first report on our new software, Miller Analyzer for Matlab, which is an automatic roundoff error analyzer that extends the work of Miller et al. Using the original work of Miller, the analyzed numerical algorithm had to be expressed in a special, greatly simplified programming language. The main disadvantage of Miller's software is its own programming language and its restrictions. We have eliminated this drawback by creating a Matlab interface to the method of Miller. Applying the operator overloading based implementation technique of automatic differentiation, we have provided a means of analyzing numerical methods given in the form of Matlab m-functions. This new way of defining the analyzed numerical method is more flexible, and the integration into the Matlab environment makes the use of the software easy. Our work can be useful in analyzing the numerical stability of algorithms that use only the four basic real (not complex) arithmetic operations and square root. At the end of the paper, we discuss two such examples.

In the paper a parallel FPGA implementation of the Peer Group Filtering algorithm is described. Implementation details, results, performance of the design and FPGA logic resources are discussed. The PGF algorithm customized for FPGA is compared with the original one and Vector Median Filtering.

The paper proposes an architecture of a system automating the provisioning process of cloud computing infrastructures. Its structure and components are specified, based on an analysis of three types of requirements: infrastructure providers, service providers and end users. These considerations have led us to formulate a new infrastructural model, offered to end users as a collection of Virtual Machines (VM) connected by a dedicated Virtual Private Network (VPN) with QoS guarantees. The role of repositories in cloud provisioning systems is specified along with the relevant data acquisition processes. The applicability of the proposed system is illustrated by practical usage scenarios.

A Distributed Denial of Service attack is a large-scale, coordinated attack on the availability of services of a victim system, launched indirectly through many compromised computers on the Internet. Intrusion detection systems are network security tools that process local audit data or monitor network traffic to search for specific patterns or certain deviations from expected behavior, which indicate malicious activities against the protected network. In this study, we propose distributed intrusion detection methods to detect Distributed Denial of Service attacks in a special dataset and test these methods in a simulated-real time environment, in which the mobile agents are synchronized with the timestamp stated in the dataset. All of our methods use the alarms generated by SNORT, a signature-based network intrusion detection system. We use mobile agents in our methods on the Jade platform in order to reduce network bandwidth usage and to decrease the dependency on the central unit for a higher reliability. The methods are compared based on reliability, network load and mean detection time values.

In this paper, we utilize a novel framework for 3D face recognition, called 3D Gabor Patched Spectral Regression (3D GPSR), which can overcome some of the continuing challenges encountered with 2D or 3D facial images. In this active field, some obstacles, like expression variations, pose correction and data noise deteriorate the performance significantly. Our proposed system addresses these problems by first extracting the main facial area to remove irrelevant information corresponding to shoulders and necks. Pose correction is used to minimize the influence of large pose variations and then the normalized depth and gray images can be obtained. Due to better time-frequency characteristics and a distinctive biological background, the Gabor feature is extracted on depth images, known as 3D Gabor faces. Data noise is mainly caused by distorted meshes, varieties of subordinates and misalignment. To solve these problems, we introduce a Patched Spectral Regression strategy, which can make good use of the robustness and efficiency of accurate patched discriminant low-dimension features and minimize the effect of noise term. Computational analysis shows that spectral regression is much faster than the traditional approaches. Our experiments are based on the CASIA and FRGC 3D face databases which contain a huge number of challenging data. Experimental results show that our framework consistently outperforms the other existing methods with the distinctive characteristics of efficiency, robustness and generality.

Knowing the driver's point of gaze has significant potential to enhance driving safety, eye movements can be used as an indicator of the attention state of a driver; but the primary obstacle of integrating eye gaze into today's large scale real world driving attention study is the availability of a reliable, low-cost eye-tracking system. In this paper, we make an attempt to investigate such a real-time system to collect driver's eye gaze in real world driving environment. A novel eye-tracking approach is proposed based on low cost head mounted eye tracker. Our approach detects corneal reflection and pupil edge points firstly, and then fits the points with ellipse. The proposed approach is available in different illumination and driving environment from simple inexpensive head mounted eye tracker, which can be widely used in large scale experiments. The experimental results illustrate our approach can reliably estimate eye position with an accuracy of average 0.34 degree of visual angle in door experiment and 2–5 degrees in real driving environments.

This paper considers a genetic algorithm (GA) for a machine-job assignment with controllable processing times (MJACPT). Integer representation with standard genetic operators is used. In an objective function, a job assignment is obtained from genetic code and for this, fixed assignment processing times are calculated by solving a constrained nonlinear convex optimization problem. Additionally, the job assignment of each individual is improved by local search. Computational results are presented for the instances from literature and modified large-scale instances for the generalized assignment problem (GAP). It can be seen that the proposed GA approach reaches almost all optimal solutions, which are known in advance, except in one case. For large-scale instances, GA obtained reasonably good solutions in relatively short computational time.

In software industry, a common problem that the companies face is to decide what requirements should be implemented in the next release of the software. This paper aims to address the multi-objective next release problem using search based methods such as multi-objective evolutionary algorithms for empirical studies. In order to achieve the above goal, a requirement-dependency-based multi-objective next release model (MONRP/RD) is formulated firstly. The two objectives we are interested in are customers' satisfaction and requirement cost. A popular multi-objective evolutionary approach (MOEA), NSGA-II, is applied to provide the feasible solutions that balance between the two objectives aimed. The scalability of the formulated MONRP/RD and the influence of the requirement dependencies are investigated through simulations as well. This paper proposes an improved version of the multi-objective invasive weed optimization and compares it with various state-of-the-art multi-objective approaches on both synthetic and real-world data sets to find the most suitable algorithm for the problem.

This paper proposes a semi-fragile, robust-to-JPEG2000 compression watermarking method which is based on the Human Visual System (HVS). This method is designed to verify the content integrity of Region of Interest (ROI) in tele-radiology images. The design of watermarking systems based on HVS leads to the possibility of embedding watermarks in places that are not obvious to the human eye. In this way, notwithstanding increased capacity and robustness, it becomes possible to hide more watermarks. Based on perceptual model of HVS, we propose a new watermarking scheme that embeds the watermarks using a replacement method. Thus, the proposed method not only detects the watermarks but also extracts them. The novelty of our ROI-based method is in the way that we interpret the obtained coefficients of the HVS perceptual model: instead of interpreting these coefficients as weights, we assume them to be embedding locations. In our method, the information to be embedded is extracted from inter-subband statistical relations of ROI. Then, the semi-fragile watermarks are embedded in the obtained places in level 3 of the DWT decomposition of the Region of Background (ROB). The compatibility of the embedded signatures and extracted watermarks is used to verify the content of ROI. Our simulations confirm improved fidelity and robustness.

In the paper, several modifications to the conventional learning algorithms of the Hierarchical Temporal Memory (HTM) – a biologically inspired large-scale model of the neocortex by Numenta – have been proposed. Firstly, an alternative spatial pooling method has been introduced, which makes use of a random pattern generator exploiting the Metropolis-Hastings algorithm. The original inference algorithm by Numenta has been reformulated, in order to reduce a number of tunable parameters and to optimize its computational efficiency. The main contribution of the paper consists in the proposal of a novel temporal pooling method – the pair-wise explorer – which allows faster and more reliable training of the HTM networks using data without inherent temporal information (e.g., static images). While the conventional temporal pooler trains the HTM network on a finite segment of the smooth Brownian-like random walk across the training images, the proposed method performs training by means of the pairs of patterns randomly sampled (in a special manner) from a virtually infinite smooth random walk. We have conducted a set of experiments with the single-layer HTM network applied to the position, scale, and rotation-invariant recognition of geometric objects. The obtained results provide a clear evidence that the pair-wise method yields significantly faster convergence to the theoretical maximum of the classification accuracy with respect to both the length of the training sequence (defined by the maximum allowed number of updates of the time adjacency matrix – TAM) and the number of training patterns. The advantage of the proposed explorer manifested itself mostly in the lower range of TAM updates where it caused up to 10