Agata Manolova

Classification of dissimilarity data with a new flexible Mahalanobis-like metric

Statistical pattern recognition traditionally relies on feature-based representation. For many applications, such vector representation is not available and we only possess proximity data (distance, dissimilarity, similarity, ranks, etc.). In this paper, we consider a particular point of view on discriminant analysis from dissimilarity data. Our approach is inspired by the Gaussian classifier and we defined decision rules to mimic the behavior of a linear or a quadratic classifier. The number of parameters is limited (two per class). Numerical experiments on artificial and real data show interesting behavior compared to Support Vector Machines and to kNN classifier: (a) lower or equivalent error rate, (b) equivalent CPU time, (c) more robustness with sparse dissimilarity data.

Challenges in designing and implementation of an effective Ambient Assisted Living system

In this paper some challenges in the design and realization of an effective Ambient Assisted Living (AAL) system are discussed. Solutions to meet those challenges are proposed. Example of the practical implementation of the architecture of an AAL system - “eWall for Active Long Living” (eWALL) and the related context-aware services are presented.

Pain detection from facial characteristics using supervised descent method

Current methods of assessing pain depend almost entirely on verbal report such as clinical interview or questionnaires of the patients. Pain being a symptom that can neither be felt nor seen, poses a major problem for the medical personnel involved in pain management since there are no accurate objective measures to establish the extent of pain the patient is suffering from when using a remote assistive medical system. The verbal report grading pain has obvious discrepancies, especially when it comes to children or people with limited ability to communicate (i.e. the mute, mentally impaired, or patients having assisted breathing). When designing a medical assistive system measuring pain in an efficient way is of great importance. In this paper we proposes an algorithm for both automatic pain recognition (i.e. pain/no pain presence in human) and continuous pain intensity estimation based on facial expression analysis. To locate specific landmarks in the face we used Supervised Descent Method (SDM) and then extract feature vectors using Scale Invariant Feature Transform (SIFT). For the recognition task we build a classier based on Support Vector Machines (SVM) and for the continuous pain intensity estimation task we trained linear regressor. The experiments with patients with shoulder pain show very good recognition rate (more than 95.7%). For the pain intensity estimation we reached an average Mean Squared Error of 1.28 and Correlation coefficient of 0.59. The recorded results demonstrate performance that exceeds state-of-the-art results on a standard data set.

Recognition of facial images with subspace projection and dissimilarity representation

In this work, we present a framework for face recognition, combining face detection algorithm, dimensionality reduction method and a dissimilarity-based classifier. The face detection algorithm is intended to detect and extract faces in complex scenes, prior to face recognition. The Spectral Regression method, in sparse setting, is used for dimensionality reduction. The classification problem is solved by the Proximity Index ”Shape Coefficient” with SVM decision rules and Prototype Selection based classification. The results with real world experiments encourage us to propose this framework as good alternative to other face recognition methods.

Non-intrusive sleep analyzer for real time detection of sleep anomalies

Solutions for caring for the elderly both efficacious and cost-effective are given by Ambient Assisted Living (AAL) systems that combine the research fields of intelligent systems and communication technologies. These systems are promising for the improvement of the quality of life of elderly and disabled people. One important characteristic of health and well-being is sleep. While sleep quantity is directly measurable, its quality has traditionally been assessed with subjective methods such as questionnaires. In this paper, we propose a non-intrusive sleep analyzer for real time detection of sleep anomalies, part of an effective AAL system. The proposed solution is based on combination of non-invasive sensors and an algorithm for sleep analysis with two stages - low and high level reasoning. It also offers the opportunity to include third party devices. Using the analyzer we can monitor basic sleep behavior and to detect sleep anomalies, which can serve as an important indicator for both mental and physical health.

Human Gaze Tracking With An Active Multi-Camera System

This paper presents a framework for determining the direction of human gaze with an active multi-camera system. A fixed camera is employed in order to estimate the position of the human face and its features, like the eyes. By means of the Supervised Descent Method (SDM) for minimizing a Non-linear Least Squares (NLS) function we can compute correctly the position of the two eyes using 6 landmarks for each of them and the pose of the head. Then an active pan-tilt camera is oriented to one of the users eyes. This way a high precision gaze direction determination is accomplished.

Facial Expression Classification Using Supervised Descent Method Combined With PCA and SVM

It has been well known that there is a correlation between facial expression and person’s internal emotional state. In this paper we use an approach to distinguish between neutral and some other expression: based on the displacement of important facial points (coordinates of edges of the mouth, eyes, eyebrows, etc.). Further the feature vectors are formed by concatenating the landmarks data from Supervised Descent Method, applying PCA and use these data as an input to Support Vector Machine (SVM) classifier. The experimental results show improvement of the recognition rate in comparison to some state-of-the-art facial expression recognition techniques.

Level-Set Based Algorithm for Automatic Feature Extraction on 3D Meshes: Application to Crater Detection on Mars

The knowledge of the origin and development of all bodies in the solar system begins with understanding the geologic history and evolution of the universe. The only approach for dating celestial body surfaces is by the analysis of the crater impact density and size. In order to facilitate this process, automatic approaches have been proposed for the impact craters detection. In this article, we propose a novel approach for detecting craters’ rims. The developed method is based on a study of the Digital Elevation Model (DEM) geometry, represented as a 3D triangulated mesh. We use curvature analysis, in combination with a fast local quantization method to automatically detect the craters’ rims with artificial neural network. The validation of the method is performed on Barlow’s database.

Classification of Mental Tasks from EEG Signals Using Spectral Analysis, PCA and SVM

Abstract Signals provided by the ElectroEncephaloGraphy (EEG) are widely used in Brain-Computer Interface (BCI) applications. They can be further analyzed and used for thinking activity recognition. In this paper we proposed an algorithm that is able to recognize five mental tasks using 6 channel EEG data. The main idea is to separate the raw EEG signals into several frames and compute their spectrums. Next, a second-order derivative of Gaussian is applied to extract features and an optimum Gaussian kernel parameters grid search is performed with the help of cross-validation. The extracted features are further reduced by Principal Component Analysis. The processed data is utilized to train SVM classifier which is used for mental tasks recognition afterwards. The performance of the algorithm is estimated on publically available dataset. In terms of 5 folds cross-validation we obtained an average of 82.7% recognition rate (accuracy). Additional experiments were conducted using leave-one-out cross-validation where 67.2% correct classification was reported. Comparison to several state-of-the art methods reveals the advantages of the proposed algorithm.

Feature selection in affective speech classification

The increasing role of spoken language interfaces in human-computer interaction applications has created conditions to facilitate a new area of research — namely recognizing the emotional state of the speaker through speech signals. This paper proposes a text independent method for emotion classification of speech signals used for the recognition of the emotional state of the speaker. Different feature selection criteria are explored and analyzed, namely Mutual Information Maximization (MIM) feature scoring criterion and its derivatives, to measure how potentially useful a feature or feature subset may be when used in a classifier. The proposed method employs different groups of low-level features, such as energy, zero-crossing rate, frequency bands in Mel scale, fundamental frequency or pitch, the delta- and delta-delta regression and statistical functions such as regression coefficients, extremums, moments etc., to represent the speech signals and a Neural Network classifier for the classification task. For the experiments the EMO-DB dataset is used with seven primary emotions including neutral. Results show that the proposed system yields an average accuracy of over 85% for recognizing 7 emotions with 5 of the best performing feature selection algorithms.