Advances in Computational Intelligence

Growing models have been widely used for clustering or topology learning. Traditionally these models work on stationary environments, grow incrementally and adapt their nodes to a given distribution based on global parameters. In this paper, we present an enhanced unsupervised self-organising network for the modelling of visual objects. We first develop a framework for building non-rigid shapes using the growth mechanism of the self-organising maps, and then we define an optimal number of nodes without overfitting or underfitting the network based on the knowledge obtained from information-theoretic considerations. We present experimental results for hands and we quantitatively evaluate the matching capabilities of the proposed method with the topographic product.

The topological features of the communication network between computing nodes in Parallel Genetic Algorithms, under the framework of the island model, is discussed in the context of both the local rate of information exchange between nodes, and the global exchange rate that measures the level of information flow in the entire network. For optimal performance of parallel genetic algorithm for a set of benchmark functions, the connectivity of the network can be found, corresponding to a global information exchange rate between 40-70%. This range is obtained by statistical analysis on the search for solutions of four benchmark problems: the 0-1 knapsack, the Weierstrass’s function, the Ackley’s function, and the Modified Shekel’s foxholes function. Our method is based on the cutting of links of a fully connected network to gradually decrease the connectivity, and compare the performance of the genetic algorithm on each network. Suggestions for the protocol in applying this general guideline in the design of a good communication network for parallel genetic algorithms are made, where the islands are connected with 40% of links of a fully connected network before fine tuning the parameters of the island model to enhance performance in a specific problem.

Clustering Search (

*CS

) has been proposed as a generic way of combining search metaheuristics with clustering to detect promising search areas before applying local search procedures. The clustering process may keep representative solutions associated to different search subspaces (search areas). In this work, a new approach is proposed, based on Artificial Bee Colony (ABC), observing the inherent characteristics of detecting promissing food sources employed by that metaheuristic. The proposed hybrid algorithm, performing a Hooke & Jeeves based local, is compared against other versions of ABC: a pure ABC and another hybrid ABC, exploring an elitist criteria.

The service time of the container vessels is the main indicator of the competitiveness of a maritime container terminal. This work proposes two Variable Neighbourhood Searches (VNS) in order to tackle the Tactical Berth Allocation Problem and the Quay Crane Scheduling Problem, which are the main operational problems in the seaside. These metaheuristics are integrated into a framework that provides an overall planning for the vessels arrived to port within a given time horizon. The performance of the proposed VNSs is compared with the most highlighted solution methods published in the related literature. In addition, the effectiveness of the framework is assessed in real size environments.

This paper tackles a Restricted Dynamic Heterogeneous Fleet Vehicle Routing Problem with Time Windows as a real-world application of a courier service company in the Canary Islands, Spain. In this particular application of the Vehicle Routing Problem with Time Windows (VRPTW), customer requests can be either known at the beginning of the planning horizon or dynamically revealed over the day. Moreover, a heterogeneous fleet of vehicles has to be routed in real time. In addition, some other constraints required by the company, such as the allowance of extra hours for the vehicles, as well as, the use of several objective functions, are taken into account. This paper proposes a metaheuristic procedure to solve this particular problem. It has already been installed in the fleet management system of the company. The computational experiments indicate that the proposed method is both feasible to solve this real-world problem and competitive with the literature.

This work is motivated by a real problem posed to the authors by a company in Tenerife, Spain. Given a set of service orders,daily routes have to be designed in order to minimize the total traveled distance while balancing the workload of drivers in terms of required time. A bi-objective mixed-integer linear model for the problem is formulated and a solution approach, based on metaheuristics, is proposed. One the main handicaps associated to this approach is the fact that it is very time consuming for non-standard literature instances, mainly due to the the initial solution generation method. Therefore, the goal of this work is to study the performance of three different ways to build the initial solutions and observe what is their impact on the approximations of the Pareto Front for Solomon instances of 100 customers. Results obtained on a real instance are also discussed.

We consider a path planning problem wherein an agent needs to safely and swiftly navigate from a given source location to a destination through an arrangement of disk-shaped obstacles. The agent possesses a limited neutralization capability in the sense that it can neutralize a certain number of obstacles enroute and pass through them safely upon neutralization. Optimal utilization of such a capability is called the neutralization problem. This problem is essentially a shortest path problem with resource constraints, which has been shown to be NP-Hard except for some trivial variants. In this study, we propose an ant system algorithm for the neutralization problem. In the proposed algorithm, the state transition rule makes use of certain problem-specific information to guide the ants. We show how the parameters of the algorithm can be fine-tuned for enhanced performance and we present limited computational experiments including a real-world naval minefield dataset. Our experiments suggest that the proposed algorithm finds high quality solutions in general with reasonable computational resources.

Statistical fraud detection problem is a very difficult problem in that there are very few examples of fraud. The great majority of transactions are legitimate. On the other hand, for this binary classification problem the costs of the two types of classification errors (FP=false positive and FN=false negative) are not the same. Thus, the classical data mining algorithms do not fit to the problem exactly. Departing from this fact, we have solved this problem by genetic algorithms and scatter search. Now, we apply the recently developed new metaheuristics algorithm namely the migrating birds optimization algorithm (MBO) to this problem. Results show that it outperforms the former approach. The performance of standard MBO is further increased by the help of some modified benefit mechanisms.

This paper presents a comparison of several methods of measuring the quality of a subset of features that characterise kidney’s graft so they can be evaluated to be transplanted. First, two non-parametric methods, Delta Test and Mutual Information, are used isolated and in a multiobjective manner using a genetic algorithm and comparing the solutions will all the possible solutions obtained by brute force. Afterwards, LSSVM are used to approximate the score of the graft so, for smaller approximation errors, the subset of features is considered better. The results obtained are confirmed from the clinical perspective by an expert.

There is a need for fast and cost-effective leukemia identification methods, because early identification could increase the likelihood of recovery. Currently, diagnostic methods require sophisticated expensive laboratories such as immune-phenotype and cytogenetic abnormality. Therefore, we propose an identification method based on using blood smear images of normal and cancerous cells, in addition to a neural network classifier. We focus in this paper on identifying Acute Lumphoblastic Leukemia (ALL) cases, and implement our experiments following three learning schemes for a neural model. The neural classifiers distinguish between normal blood cells and ALL-infected cells. The experimental results show that the proposed novel leukemia identification system can be effectively used for such a task, and thus could be implemented for identifying other leukemia types in real life applications.

The development of accurate models to describe and predict pressure inactivation kinetics of microorganisms is very beneficial to the food industry for optimization of process conditions. The need for methods to model highly nonlinear systems is long established. The architecture of a novel clustering-based fuzzy wavelet neural network (CB-FWNN) model is proposed. The objective of this research is to investigate the capabilities of the proposed scheme, in predicting the survival curves of

Listeria monocytogenes

inactivated by high hydrostatic pressure in UHT whole milk. The proposed model is obtained from the Takagi–Sugeno–Kang fuzzy system by replacing the THEN part of fuzzy rules with a “multiplication” wavelet neural network. Multidimensional Gaussian type of activation functions have been used in the IF part of the fuzzy rules. The performance of the proposed scheme has been compared against neural networks and partial least squares models usually used in food microbiology.

This paper presents the simulation and learning of soft tissue temperature dynamics when exposed to laser radiation. Monte Carlo simulation is used to represent the photon distribution in the tissue while machine learning techniques are used to obtain the mapping from controllable laser inputs (power, pulse rate and exposure time) to the correspondent changes in temperature. This model is required to predict the effects of laser-tissue interaction during surgery, i.e., tissue incision depth and carbonization.

Detection of Alzheimer’s disease based on Magnetic Resonance Imaging (MRI) still is one of the most sought goals in the neuroscientific community. Here, we evaluate a ensemble of classifiers each independently trained with disjoint data extracted from a partition of the brain data volumes performed according to the 116 regions of the Anatomical Automatic Labeling (AAL) brain atlas. Grey-matter probability values from 416 subjects (316 controls and 100 patients) of the OASIS database are estimated, partitioned into AAL regions, and summary statistics

per

region are computed to create the feature sets. Our objective is to discriminate between control subjects and Alzheimer’s disease patients. For validation we performed a leave-one-out process. Elementary classifiers are linear Support Vector Machines (SVM) with model parameter estimated by grid search. The ensemble is composed of one SVM per AAL region, and we test 6 different methods to make the collective decision. The best performance achieved with this approach is 83.6% accuracy, 91.0% sensitivity, 81.3% specificity and 0.86 of area under the ROC curve. Most discriminant regions for some of the collective decision methods are also provided.

In this paper, we compare the performance of two different methods for the task of electrooculogram saccadic points classification in patients with Ataxia SCA2:

Multilayer Perceptrons

(MLP) and

Random Forest

. First we segment the recordings of 6 subjects into ranges of saccadic and non-saccadic points as the basis of supervised learning. Then, we randomly select a set of cases based on the velocity profile near each selected point for training and validation purposes using percent split scheme. Obtained results show that both methods have similar performance in classification matter, and seems to be suitable to solve the problem of saccadic point classification in electrooculographic records from subjects with Ataxia SCA2.

In this paper, phonocardiography (PCG) segmentation methodology based on envelope detection is developed by using a time-scale representation and a synthetic electrocardiogram signal (EKG). The heart cycle duration is calculated by autocorrelation of S1-S2 sounds that are synchronized with the synthetic EKG. Two algorithms for noisy signal removal are implemented to ensure the detection of signals with low signal to noise ratio. Approach is tested in a PCG database holding 232 recordings. Results show an achieved accuracy up of 90%, thus, overperforming three state-of-the-art PCG segmentation techniques used to compare the proposed approach. Additionally, the synthetic EKG is built by estimation of heart rate length, thus it does not use a patient recording EKG, reducing the computational cost and the amount of required devices.

Ataxia SCA2 is a neurological disorder among a group of inherited diseases of the central nervous system. In SCA2, genetic defects lead to impairment of specific nerve fibers, resulting in degeneration of the cerebellum and its afferent connections. As anomalies in the oculomotor system are well known symptoms in SCA2, electro-oculographic records become a useful technique for SCA2 diagnosis. This work presents a novel technique for determining how many decomposition levels are necessary to perform signal de-noising, based on the evaluation of the shape correspondence between the wavelet approximation coefficients and the EOG record, focusing in noise cancellation in order to obtain a clean velocity profile. Experimental results show the validity of the approach.

Analysis of human movement is an important research area, specially for health applications. In order to assess the quality of life of people with mobility problems like Parkinson’s disease (PD) or stroke patients, it is crucial to monitor their daily life activities. The main goal of this work is to characterize basic activities and their transitions using a single sensor located at the waist. This paper presents a novel postural detection algorithm which is able to detect and identify 6 different postural transitions, sit to stand, stand to sit, bending up/down and lying to sit and sit to lying transitions with a sensitivity of 86.5% and specificity of 95%. The algorithm has been tested on 31 healthy volunteers and 8 PD patients who performed a total of 545 and 176 transitions respectively. The proposed algorithm is suitable to be implemented in real-time systems for on-line monitoring applications.

Multiple sequence alignment (MSA) is an essential approach to apply in other outstanding bioinformatics tasks such as structural predictions, biological function analyses or phylogenetic modeling. However, current MSA methodologies do not reach a consensus about how sequences must be accurately aligned. Moreover, these tools usually provide partially optimal alignments, as each one is focused on specific features. Thus, the same set of sequences can provide quite different alignments, overall when sequences are less related. Consequently, researchers and biologists do not agree on how the quality of MSAs should be evaluated in order to decide the most adequate methodology. Therefore, recent evaluations tend to use more complex scores including supplementary biological features. In this work, we address the evaluation of MSAs by using a novel supervised learning approach based on Least Square Support Vector Machine (LS-SVM). This algorithm will include a set of heterogeneous features and scores in order to determine the alignment accuracies. It is assessed by means of the benchmark BAliBASE.

The reception of vibrations is a special sense of touch, important for many insects and vertebrates. The latter realize this reception by means of hair-shaped vibrissae in the mystacial pad, to acquire tactile information about their environments. The system models have to allow for stabilizing and tracking control while nevertheless being able to detect superimposed solitary excitations. Controllers have to be adaptive in view of both the randomness of the external signals to be suppressed and the uncertainty of system data. We presents mechanical models and an improved adaptive control strategy that avoids identification but renders the system sensitive.

A new approach to continuous authentication is presented. The method is based on a combination of statistical decision machines for brain signals. Functional Near InfraRed Spectroscopy (NIRS) is used to measure brain oxyhemoglobin changes for each subject to be authenticated. Such biosignal authentication is expected to be a viable complementary method to traditional static security systems. The designed system is based on a discriminant function which utilizes the average weight vector of one-versus-one support vector machines for NIRS spectra. By computing a histogram of Mahalanobis distances, high separability among subjects was recognized. This experimental result guarantees the utility of brain NIRS signals to the continuous authentication.

Screening femoral neck osteoporosis is important to prevent fractures of the femoral neck. We developed machine learning models with the aim of more accurately identifying the risk of femoral neck osteoporosis in postmenopausal women and compared those to a conventional clinical decision tool, osteoporosis self-assessment tool (OST). We collected medical records based on the Korea National Health and Nutrition Surveys. The training set was used to construct models based on popular machine learning algorithms using various predictors associated with osteoporosis. The learning models were compared to OST. Support vector machines (SVM) had better performance than OST. Validation on the test set showed that SVM predicted femoral neck osteoporosis with an area under the curve of the receiver operating characteristic of 0.874, accuracy of 80.4%, sensitivity of 81.3%, and specificity of 80.5%. The machine learning methods may be effective tools for identifying postmenopausal women at high risk for osteoporosis.

Bioinformatics is facing a post-genomic era characterized by the release of large amounts of data boosted by the scientific revolution in high throughput technologies. This document presents an approach to deal with such a massive data processing problem in a paradigmatic application from which interesting lessons can be learned. The design of an out-of-core and modular implementation of traditional High-scoring Segment Pairs (HSPs) applications removes the limits of genome size and performs the work in linear time and with controlled computational requirements. Regardless of the expected huge I/O operations, the full system performs faster than state-of-the-art references providing additional advantages such as monitoring and interactive analysis, the exploitation of important intermediate results, and giving the specific nature of the modules, instead of monolithic software, enabling the plugging of external components to squeeze results.

Technological advances in biological and biomedical data acquisition are creating mountains of data. Existing legacy applications are unable to process this data without using new strategies. However, some workloads in bioinformatics are easily parallelized by splitting the data, running legacy applications in parallel and then join the partial results into one final result. In this paper, we present Bio-Cirrus, a software package which facilitates this process. Our software consists of a user-friendly client (jORCA) for accessing Web Services and enacting workflows, and a module (Mr. Cirrus) for processing the data with a map/reduce style approach. Bio-Cirrus binaries and documentation are freely available at

http://www.bitlab-es.com/cloud

under the Creative Commons Attribution-No Derivative Works 2.5 Spain License and its source code is available under request. (GPL v3 license).

Ensuring ubiquity, robustness and continuity of monitoring is of key importance in activity recognition. To that end, multiple sensor configurations and fusion techniques are ever more used. In this paper we present a multi-sensor meta-classifier that aggregates the knowledge of several sensor-based decision entities to provide a unique and reliable activity classification. This model introduces a new weighting scheme which improves the rating of the impact that each entity has on the decision fusion process. Sensitivity and specificity are particularly considered as insertion and rejection weighting metrics instead of the overall accuracy classification performance proposed in a previous work. For the sake of comparison, both new and previous weighting models together with feature fusion models are tested on an extensive activity recognition benchmark dataset. The results demonstrate that the new weighting scheme enhances the decision aggregation thus leading to an improved recognition system.

Activity recognition has recently gained a lot of interest and appears to be a promising approach to help the elderly population pursue an independent living. There already exist several methods to detect human activities based either on wearable sensors or on cameras but few of them combine the two modalities. This paper presents a strategy to enhance the robustness of indoor human activity recognition by combining wearable and depth sensors. To exploit the data captured by those sensors, we used an ensemble of binary one-vs-all neural network classifiers. Each activity-specific model was configured to maximize its performance. The performance of the complete system is comparable to lazy learning methods (

k

-NN) that require the whole dataset.

After recalling the different interpretations usually assigned to the term Galois connection, both in the crisp and in the fuzzy case, we survey on several of their applications in Computer Science and specifically, in Soft Computing.

In this work, a proximity-based generic method for discovery of generalized knowledge is presented and implemented in the framework of a fuzzy logic programming language with a weak unification procedure that uses proximity relations to model uncertainty. This method makes use of the concept of

λ

-block characterizing the notion of equivalence when working with proximity relations. When the universe of discourse is composed of concepts which are related by proximity, the sets of

λ

-blocks extracted from that proximity relation can be seen as hierarchical sets of concepts grouped by abstraction level. Then, each group (forming a

λ

-block) can be labeled, with user help, by way of a more general descriptor in order to simulate a generalization process based on proximity. Thanks to this process, the system can learn concepts that were unknown initially and reply queries that it was not able to answer. The novelty of this work is that it is the first time a method, with analogous features to the one aforementioned, has been implemented inside a fuzzy logic programming framework. In order to check the feasibility of the method we have developed a software tool which have been integrated into the

Bousi

~

Prolog

system.

Fuzzy property-oriented concept lattices are a formal tool for modeling and processing incomplete information in information systems. This paper relates this theory to fuzzy mathematical morphology, which scope, for instance, is to process and analyze images and signals. Consequently, the theory developed in the concept lattice framework can be used in these particular settings.

Fuzzy set theory has proved to be a successful paradigm to extend the database relational model, augmenting its skill to capture uncertaintly. This capability may be consider in two levels: the data itself and the constraints defined to adjust the database schema to the real system. When constraints are considered, it is necessary to design methods to reason about it and not only a way to express them. This situation leads to a ambitious goal: the design of automated reasoning methods. Highly-expressive data models are not useful without an automated reasoning method. In this work we introduce an automated method to infer with fuzzy functional dependencies over a high level generalization of the relational model and provide its completeness result.

In formal concept analysis, attribute reduction is an important preprocessing in order to obtain concept lattices, which provides fundamental information of the attributes, as well. This importance is increased in the fuzzy case.

This paper presents, in the general fuzzy framework of multi-adjoint concept lattices, a classification of the fuzzy-attributes of a context, which provides interesting properties of the attributes and its application to reduce the computational complexity for building this kind of concept lattices.

The overcapacity of the European fishing fleets is one of the recognized factors for the lack of success of the Common Fisheries Policy. Unwanted non-targeted species and other incidental fish likely represent one of the causes for the overexploitation of fish stocks; thus there is a clear connection between this problem and the type of fishing gear used by vessels. This paper performs an environmental impact study of the Spanish Fishing Fleet by means of ordinal classification techniques to emphasize the need to design an effective and differentiated common fish policy for “artisan fleets”, that guarantees the maintenance of environmental stocks and the artesan fishing culture.

Determine an appropriate distribution of reservoir parameters is a challenge in reservoir engineering. Permeability is one of special reservoir parameters which its modeling is more complicated because there is no direct tool to determine permeability distribution. This problem is more critical in carbonate reservoir because of the fracture effects on measurements. The most reliable way of permeability calculation is laboratory analysis of cores, but this method could not provide a thorough permeability profile in the desired field. In recent years, different methods and algorithms used to predict permeability. One of the most common methods is artificial intelligent methods such as ANN, FL and GA. This paper provides a way to compare the ability of different fuzzy methods to predict permeability from well logs in one of southern Iranian carbonate reservoirs. Sugeno type fuzzy inference system (SFIS), adaptive neuro-fuzzy inference system (ANFIS) and locally linear neuro-fuzzy (LLNF) used to predict permeability. One third of all data used for test the fuzzy systems. Mean square error (MSE) and correlation coefficient (CC) of the test dataset used to select the best method in permeability determination. In final step genetic algorithm is applied to combine different method results to obtain a final model. This algorithm minimizes an error function. This function consists of SFIS, ANFIS and LLNF model predictions. The ability of different methods are compared to find an appropriate method for permeability prediction.

In this paper we present a method for debugging XPath queries which has been implemented with the fuzzy logic language MALP by using the FLOPER tool developed in our group. We describe how XPath expressions can be manipulated for obtaining a set of alternative queries matching a given XML document. For each new proposed query, we give a “chance degree” that represents an estimation on its deviation w.r.t. the initial expression. Our work is focused on providing to the programmers a repertoire of paths which can be used to retrieve answers.

This work describes the design of a bot for the first person shooter Unreal Tournament

TM

2004 (UT2K4), which behaves as a human expert player in 1 vs. 1 death matches. This has been implemented modelling the actions (and tricks) of this player, using a state-based IA, and supplemented by a database for ‘learning’ the arena. The expert bot yields excellent results, beating the game default bots in the hardest difficulty, and even being a very hard opponent for the human players (including our expert). The AI of this bot is then improved by means of three different approaches of evolutionary algorithms, optimizing a wide set of parameters (weights and probabilities) which the expert bot considers when playing. The result of this process yields an even better rival; however the noisy nature of the fitness function (due to the pseudo-stochasticity of the battles) makes the evolution slower than usual.

This work studies the performance and the results of the application of Evolutionary Algorithms (EAs) for evolving the decision engine of a program, called in this context

agent

, which controls the player’s behaviour in an real-time strategy game (RTS). This game was chosen for the Google Artificial Intelligence Challenge in 2011, and simulates battles between teams of ants in different types of maps or mazes. According to the championship rules the agents cannot save information from one game to the next, which makes impossible to implement an EA ‘inside’ the agent, i.e. on game time (or on-line), that is why in this paper we have evolved this engine off-line by means of an EA, used for tuning a set of constants, weights and probabilities which direct the rules. This evolved agent has fought against other successful bots which finished in higher positions in the competition final rank. The results show that, although the best agents are difficult to beat, our simple agent tuned with an EA can outperform agents which have finished 1000 positions above the untrained version.

User interaction experience has been lately the main focus of the industry of entertainment. High rendered graphics and huge computing process have been lagging behind for paving the way to the user experience interaction in which the way the user gets involved with the system is a key point. For this reason, we present along with this paper an enhancement for video games and museum applications that is able to increase user experience by separating the display from the controller side, using mobile phones as tracking devices. For this purpose, we have implemented a system architecture based on a Bluetooth peer-to-peer model that establishes a strong connection between mobile phones and desktop applications. The utilization of mobile phones has been revealed as a fundamental element in user experience due to its ease of use and its widespread adoption among society, which makes possible to enter the competitive market of entertainment.

Car racing is a successful genre of videogames, as proved, for example, by the racing simulator saga, Gran Turismo. In this genre of games, players not only race but they are also involved in the process of setting up the car, assuming the role of a technician/mechanic/engineer. Generally, this configuration deals with a large set of parameters that range from the amount of fuel loaded into the car to the tire pressure and type. This article compares different proposals for optimizing this process using evolutionary computation techniques to make several suggestions for a simulated international competition for car racing setup optimization.

A numerical method for computational implementation of gradient dynamical systems is presented. The method is based upon the development of geometric integration numerical methods, which aim at preserving the dynamical properties of the original ordinary differential equation under discretization. In particular, the proposed method belongs to the class of discrete gradients methods, which substitute the gradient of the continuous equation with a discrete gradient, leading to a map that possesses the same Lyapunov function of the dynamical system, thus preserving the qualitative properties regardless of the step size. In this work, we apply a discrete gradient method to the implementation of Hopfield neural networks. Contrary to most geometric integration methods, the proposed algorithm can be rewritten in explicit form, which considerably improves its performance and stability. Simulation results show that the preservation of the Lyapunov function leads to an improved performance, compared to the conventional discretization.

The paper proposes a neurocomputing approach for numerical solving of a hyperbolic partial differential equation (PDE) arising from a system of conservation laws. The main idea is to combine the method of lines (transforming the mixed initial boundary value problem for PDE into a high dimensional system of ordinary differential equations (ODEs)) with a cellular neural network (CNN) optimal structure which exploits the inherent parallelism of the new problem in order to reduce the computational effort and storage. The method ensure from the beginning the convergence of the approximation and preserve the stability of the initial problem.

The structures of the Artificial Intelligence (AI) are sometimes “created” in order to solve specific problems of science and engineering. They may be viewed as dedicated signal processors, with dedicated, in particular repetitive, structure. In this paper such structures of Neural Networks (NN)-like devices are considered, having as starting point the problems in Mathematical Physics. Both the ways followed by such inferences and their outcomes may be quite diverse - one of the paper’s aims is to illustrate this assertion. Next, ensuring global stability and convergence properties in the presence of several equilibria is a common feature of the field. The general discussion on the “emergence” of AI devices with NN structure is followed by the presentation of the elements of the global behavior for systems with several equilibria. The approach is illustrated on the case of the M-lattice; in tackling this application there is pointed out the role of the high gain to ensure both gradient like behavior combined with binary outputs which are required e.g. in image processing.

Biological vision systems have become highly optimized over millions of years of evolution, developing complex neural structures to represent and process stimuli. Moreover, biological systems of vision are typically far more efficient than current human-made machine vision systems. The present report describes a non-task-dependent image representation schema that simulates the early phase of a biological neural vision mechanism. We designed a neural model involving multiple types of computational units to simulate ganglion cells and their non-classical receptive fields, local feedback control circuits and receptive field dynamic self-adjustment mechanisms in the retina. We found that, beyond the pixel level, our model was able to represent images self-adaptively and rapidly. In addition, the improved representation was found to substantially facilitate contour detection. We propose that this improvement arose because ganglion cells can resize their receptive fields, enabling multiscale analysis functionality, a neighborhood referring function and a localized synthesis function. The ganglion cell layer is the starting point of subsequent diverse visual processing. The universality of this cell type and its functional mechanisms suggests that it will be useful for designing image processing algorithms in future.

Most companies recognize the need for the integration and coordination of various components in logistics and supply chain management as an important factor. This paper presents an evolutionary approach to modeling and optimization on inventory routing problem of inventory management, logistics distribution and supply chain management. The aim of this research is to present different individual evolutionary approach, and to obtain power extension of these hybrid approaches. In general, these evolutionary hybrid approaches are more competitive than classic problem-solving methodology including improved heuristics methods or individual bio-inspired methods and their solutions in inventory management, logistics distribution and supply chain.

A Brain-Computer Interface (BCI) is a system that enables people to control an external device with their brain activity, without the need of any muscular activity. Researchers in the BCI field aim to develop applications to improve the quality of life of severely disabled patients, for whom a BCI can be a useful channel for interaction with their environment. Some of these systems are intended to control a mobile device (e. g. a wheelchair). Virtual Reality is a powerful tool that can provide the subjects with an opportunity to train and to test different applications in a safe environment. This technical review will focus on systems aimed at navigation, both in virtual and real environments.

Speller is an important application in brain-computer interface researching. In this study, we developed a novel motor imagery based braincomputer interface speller which integrates a 2-D cursor control strategy into a hex-o-spell paradigm to spell a character in two-step. The experimental results (five subjects participated) showed that the average spelling speed is 14.64 characters per minute and that its average information transfer rate is 73.96 bits per minute.

Research on Brain Computer Interfaces (BCI) covers a wide specturm of applications. In this paper we tackle the problem of spelling words which could be used by people with harmful motor skills. The P300 has been widely used to communicate with the machine using a set of stimuli. For the word spelling problem, several approaches have been proposed although the most popular consists on an array of characters.

The novelty of the work proposed in this paper is that the panel that estimulates the user generating the P300 usually performs better in terms of efficiency and allows a better visual focalisation.

At this preliminary stage of the research, several experiments were carried out using simulated signals showing that, indeed, the new way of stimulating the user could make word spelling faster.

We investigate in this paper the activity-dependent person verification method using electroencephalography (EEG) signal from a person performing motor imagery tasks. Two tasks were performed in our experiments were performed. In the first task, the same motor imagery task of left hand or right hand was applied to all persons. In the second task, only the best motor imagery task for each person was performed. The Gaussian mixture model (GMM) and support vector data description (SVDD) methods were used for modelling persons. Experimental results showed that lowest person verification error rate could be achieved when each person performed his/her best motor imagery task.

Pressure ulcer is a clinical pathology of localized damage to the skin and underlying tissue caused by pressure, shear or friction. Diagnosis, care and treatment of pressure ulcers can result in extremely expensive costs for health systems. A reliable diagnosis supported by precise wound evaluation is crucial in order to success on the treatment decision and, in some cases, to save the patient’s life. However, current evaluation procedures, focused mainly on visual inspection, do not seem to be accurate enough to accomplish this important task. This paper presents a computer-vision approach based on image processing algorithms and supervised learning techniques to help detecting and classifying wound tissue types which play an important role in wound diagnosis. The system proposed involves the use of the k-means clustering algorithm for image segmentation and a standard multilayer perceptron neural network to classify effectively each segmented region as the appropriate tissue type. Results obtained show a high performance rate which enables to support ulcer diagnosis by a reliable computational system.

Image segmentation is a typical task in the field of image processing. There is a great number of image segmentation methods in the literature, but most of these methods are not suitable for multispectral images and they require a priori knowledge. In this work, a hierarchical self-organizing network is proposed for multispectral image segmentation. An advantage of the proposed neural model is due to the hierarchical architecture, which is more flexible in the adaptation process to input data. Experimental results show that the proposed approach is promising for multispectral image processing.

Most of object detection algorithms do not yield perfect foreground segmentation masks. These errors in the initial stage of video surveillance systems could cause that the subsequent tasks like object tracking and behavior analysis, can be extremely compromised. In this paper, we propose a methodology based on self-organizing neural networks and histogram analysis, which detects unusual objects in the scene and improve the foreground mask handling occlusions between objects. Experimental results on several traffic sequences found in the literature show that the proposed methodology is promising and suitable to correct segmentation errors on crowded scenes with rigid objects.

Global illumination methods based on stochastically techniques provide photo-realistic images. However, they are prone to noise that can be reduced by increasing the number of paths as proved by Monte Carlo theory. The problem of finding the number of paths that are required in order to ensure that human observers cannot perceive any noise is still open. In this paper, a novel approach to predict which image highlights perceptual noise is proposed based on Fast Relevance Vector Machine (FRVM). This model can then be used in any progressive stochastic global illumination method in order to find the visual convergence threshold of different parts of any image. A comparative study of this model with experimental psycho-visual scores demonstrates the good consistency between these scores and the model quality measures. The proposed model has been compared also with other learning model like SVM and gives satisfactory performance.

3D sensors provides valuable information for mobile robotic tasks like scene classification or object recognition, but these sensors often produce noisy data that makes impossible applying classical keypoint detection and feature extraction techniques. Therefore, noise removal and downsampling have become essential steps in 3D data processing. In this work, we propose the use of a 3D filtering and down-sampling technique based on a Growing Neural Gas (GNG) network. GNG method is able to deal with outliers presents in the input data. These features allows to represent 3D spaces, obtaining an induced Delaunay Triangulation of the input space. Experiments show how the state-of-the-art keypoint detectors improve their performance using GNG output representation as input data. Descriptors extracted on improved keypoints perform better matching in robotics applications as 3D scene registration.

This paper presents a semi-parametric Algorithm for parsing football video structures. The approach works on a two interleaved based process that closely collaborate towards a common goal. The core part of the proposed method focus perform a fast automatic football video annotation by looking at the enhance entropy variance within a series of shot frames. The entropy is extracted on the Hue parameter from the HSV color system, not as a global feature but in spatial domain to identify regions within a shot that will characterize a certain activity within the shot period. The second part of the algorithm works towards the identification of dominant color regions that could represent players and playfield for further activity recognition. Experimental Results shows that the proposed football video segmentation algorithm performs with high accuracy

Registration is a main task in 3D objects reconstruction. Different approaches have been developed in order to solve specific problems in scenarios, objects or even the source of data. Recently, new problems have been appeared with the increasing use of low-cost RGB-D sensors. Registering small objects acquired by these cameras using traditional methods is a hard problem due to their low resolution and depth sensing error. In this paper, we propose a model-based registration method for objects composed by small planes using multi-view acquisition. It is able to deal with the problem of low resolution and noisy data. Experiments show very good promising results registering small objects acquired with low-cost RGB-D sensors compared to ICP variants.

We propose the design of a real-time system to recognize and interprethand gestures. The acquisition devices are low cost 3D sensors. 3D hand pose will be segmented, characterized and track using growing neural gas (GNG) structure.The capacity of the system to obtain information with a high degree of freedom allows the encoding of many gestures and a very accurate motion capture. The use of hand pose models combined with motion information provide with GNG permits to deal with the problem of the hand motion representation. A natural interface applied to a virtual mirrorwriting system and to a system to estimate hand pose will be designed to demonstrate the validity of the system.