Ion Marqués

Face recognition with lattice independent component analysis and extreme learning machines

We focus on two aspects of the face recognition, feature extraction and classification. We propose a two component system, introducing Lattice Independent Component Analysis (LICA) for feature extraction and Extreme Learning Machines (ELM) for classification. In previous works we have proposed LICA for a variety of image processing tasks. The first step of LICA is to identify strong lattice independent components from the data. In the second step, the set of strong lattice independent vector are used for linear unmixing of the data, obtaining a vector of abundance coefficients. The resulting abundance values are used as features for classification, specifically for face recognition. Extreme Learning Machines are accurate and fast-learning innovative classification methods based on the random generation of the input-to-hidden-units weights followed by the resolution of the linear equations to obtain the hidden-to-output weights. The LICA-ELM system has been tested against state-of-the-art feature extraction methods and classifiers, outperforming them when performing cross-validation on four large unbalanced face databases.

Spatially regularized semisupervised Ensembles of Extreme Learning Machines for hyperspectral image segmentation

This paper explores the performance of Ensembles of Extreme Learning Machine classifiers for hyperspectral image classification and segmentation in a semisupervised and spatially regularized process. The approach assumes that we have available only a small training set of labeled samples, which we enrich with a set of guessed labelings on selected samples from the vast pool of unlabeled image pixels. Selection and label guessing is conditioned to an unsupervised classification of the image pixel spectra, and to the spatial proximity to the labeled samples in the image domain. Unlabeled pixels falling in the spatial neighborhood of a labeled training sample, and belonging to the same unsupervised class, acquire its label. Unsupervised classification can be performed by any clustering technique, in this paper we have resorted to the classical K-means. The classifier built from the enriched training dataset is applied to the entire hyperspectral image. Finally, we perform a spatial regularization of the classification label image, maximizing a rather general prior smoothness criterion, by the selection of the most frequent class in each pixel neighborhood. This paper reports experiments with homogeneous ensembles of ELM, rELM, and OP-ELM classifiers, including a sensitivity analysis over the ensemble size and the number of hidden nodes. Computational experiments on four well known benchmarking hyperspectral images give state-of-the-art results.

Undesired state-action prediction in multi-agent reinforcement learning for linked multi-component robotic system control

The paper deals with the problem of learning the control of Multi-Component Robotic Systems (MCRSs) applying Multi-Agent Reinforcement Learning (MARL) algorithms. Modeling Linked MCRS usually leads to over-constrained environments, posing great difficulties for efficient learning with conventional single and multi-agent reinforcement algorithms. In this paper, we propose a hybrid learning algorithm composed of a modified Q-Learning algorithm embedding an Undesired State-Action Prediction (USAP) module trained by a supervised learning approach which learns a model predicting undesired transitions to states breaking physical constraints. The USAP modules output is used by the Q-Learning algorithm to prevent these undesired transitions, therefore boosting learning efficiency. This hybrid approach is extended to the multi-agent case embedding the USAP module in Distributed Round-Robin Q-Learning (D-RR-QL), which requires very little communications among agents. We present results of computational experiments conducted in the classical multi-agent taxi scheduling task and a hose transportation task. Results show a considerable learning gain both in time and accuracy, compared to the state-of-the-art Distributed Q-Learning approach in the deterministic taxi scheduling task. In the hose transportation task, USAP module introduces a significant improvement in learning convergence speed.

Image security and biometrics: a review

Imaging security and biometrics are two heavily connected areas. The quick evolution of biometrics has raised the need of securing biometric data. A majority of this data is visual, which has lead to intensive development of image security techniques for biometric applications. In this paper we give a fast fly over image security approaches and imaging-related biometrics. We present the current state-of-the-art of the interplay between both areas. The emphasis in this paper is the computational methods.

A Domestic Application of Intelligent Social Computing: The SandS Project

This paper introduces principal ideas of new ways to mediate the interaction between users and their domestic environment, namely the set of household appliances owned by the user. These ideas are being developed in the framework of the Social and Smart (SandS) project, which elaborates on the idea of a social network of home appliance users that exchange information and insights about the use of their appliances. This interaction constitutes the conscious social computing layer of the system. The system has a subconscious computing layer consisting of a networked intelligence that strives to provide innovative solutions to user problems, so that the system goes beyond being a recollection of appliance recipes. This paper discusses the structure of the system, as well as some data representation issues that may be instrumental to its development, as part of the development work leading to the final implementation of the project ideas.

An experiment of subconscious intelligent social computing on household appliances

Subconscious Social Intelligence refers to the design of social services oriented towards user problem solving, providing an underlying innovation layer is able to generate new solutions to yet unknown problems. The innovation layer is achieved by Computational Intelligence techniques, encompassing machine learning to build models of user satisfaction over solution quality, and stochastic search as the means for innovation generation. The SandS project provides an instance of such paradigm, where household appliances are the subject of the social service. This paper proposes a specific architecture, reporting results on a synthetic database build according to SandS project current designs. Database synthesis for system tuning and validation is a critical issue, hence the paper details the considerations guiding its design and generation, as well as the validation procedure ensuring the ecological validity of the innovation process simulation. The architecture is composed of a Support Vector Regression (SVR) module for user satisfaction modeling, and an Evolution Strategy (ES) achieving recipe innovation. The paper reports some computational experiments that may guide the real life implementation. The reported results are methodologically sound as far as they are independent of the generation process.

Fusion of lattice independent and linear features improving face identification

Abstract This paper proposes the fusion of lattice independent component analysis (LICA) features with linear features obtained from conventional methods. Specifically, we compute class conditional LICA, where separate endmembers are extracted from each class dataset. We find that this fusion approach improves systematically the recognition accuracy in face recognition applications. We report experimental results using seven state-of-the-art linear feature extraction algorithms on four public face databases using Extreme Learning Machines (ELMs) for the classification building algorithm.

Hybrid Sparse Linear and Lattice Method for Hyperspectral Image Unmixing

Linear spectral unmixing aims to estimate the fractional abundances of spectral signatures in each pixel. The Linear Mixing Model (LMM) of hyperspectral images assumes that pixel spectra are affine combinations of fundamental spectral signatures called endmembers. Endmember induction algorithms (EIA) extract the endmembers from the hyperspectral data. The WM algorithm assumes that a set of Affine Independent vectors can be extracted from the rows and columns of dual Lattice Autoassociative Memories (LAAM) built on the image spectra. Indeed, the set of endmembers induced by this algorithm defines a convex polytope covering the hyperspectral image data. However, the number of endmembers extracted can be huge. This calls for additional endmember selection steps, and to approaching the unmixing problem with linear sparse regression techniques. In this paper, we combine WM algorithm with clustering techniques and Conjugate Gradient Pursuit (CGP) for endmember induction. Our experiments are conducted using hyperspectral imaging obtained by the Airborne Visible/Infrared Imaging Spectometer of the NASA Jet Propulsion Laboratory. The limited length of the paper limits the experimental depth to the confirmation of the validity of the proposed method.

Face Processing for Security: A Short Review

In this paper we give a fast fly over the face image preocessing issue, taking special care to highlight the security related applications. Face detection is the first step for the face recognition systems, posing its own challenges. Face recognition is essentially a classification problem, which can be a large multiclass problem. The emphasis in this paper is the of review the different computational approaches instead of the concrete applications.

Greedy Sparsification WM Algorithm for Endmember Induction in Hyperspectral Images

The Linear Mixing Model (LMM) of hyperspectral images asumes that pixel spectra are affine combinations of basic spectral signatures, called endmembers, which are the vertices of a convex polytope covering the image data. Endmember induction algorithms (EIA) extract the endmembers from the image data, obtaining a precise spectral characterization of the image. The WM algorithm assumes that a set of Affine Independent vectors can be extracted from the rows and columns of dual Lattice Autoassociative Memories (LAAM) built on the image spectra. Indeed, the set of endmembers induced by this algorithm defines a convex polytope covering the hyperspectral image data. However, the number of induced endmembers obtained by this procedure is too high for practical purposes, besides they are highly correlated. In this paper, we apply a greedy sparsification algorithm aiming to select the minimal set of endmembers that explains the data in the image. We report results on a well known benchmark image.