Carlos Avilés-Cruz

High-order statistical texture analysis--font recognition applied

A new optical font recognition technique is proposed in this work. The new approach is based on global texture analysis, where statistical methods are used to identify and classify font features. The font recognition is performed by taking the document as a simple image, where one or several types of fonts are present. The identification is not performed letter by letter as with conventional approaches. In the proposed method a window analysis is employed to obtain the features of the document, using fourth and third order moments. The new technique does not involve a study of local typography; therefore, it is content independent. A detailed study was performed with 8 types of fonts commonly used in the Spanish language. Each type of font can have four styles that lead, to 32 font combinations. The font recognition with clean images is 100% accurate. Also, the new method was tested by adding Gaussian noise to clean images, so as to study the impact of image degradation on font recognition. The robustness of the algorithm is also examined in terms of varying resolution.

Self organizing natural scene image retrieval

In this work we describe a new statistically-based methodology to organize and retrieve images of natural scenes by combining feature extraction, automatic clustering, automatic indexing and classification techniques. Our proposal belongs to the content-based image retrieval (CBIR) category. Our goal is to retrieve images from an image database by their content. The methodology combines randomly extracted points for feature extraction. The describing features are the mean, the standard deviation and the homogeneity (from the co-occurrence matrix) of a sub-image extracted from the three color channels (HSI). A K-means algorithm and a 1-NN classifier are used to build an indexed database. Three databases of images of natural scenes are used during the training and testing processes. One of the advantages of our proposal is that the images are not labeled manually for their retrieval. The performance of our framework is shown through several experimental results, including a comparison with several classifiers and comparison with related works, achieving up to 100% good recognition. Additionally, our proposal includes scene retrieval.

A comprehensive finite-element model of a turbine-generator infinite-busbar system

This paper shows the development of a two-dimensional finite-element magnetic model of a turbine generator coupled to an infinite busbar through a transmission line and a delta-star connected transformer. The finite-element equations of the machine are strongly coupled to the circuit equations of the transmission line and transformer. The circuit equations developed for the delta-star connection enable the simulation of several short circuit conditions at the transformer terminals with the aid of fault matrices. The calculation of electromagnetic torque is performed by surface integration of the Lorentz force, which leads to a unique result for the finite-element model used in this work. The coupling with the external controls of the machine (automatic voltage regulator and governor) is performed in a weak form. Nevertheless, any control model can be easily connected to the finite-element model. Since the use of this finite element model is meant for transient simulations, it is desirable to obtain some simplifications to speed up calculations. Hence, a three-phase current sheet is proposed in this work to avoid the explicit modelling of the stator coreback. This approach also leads to a simple approach to model motion, where remeshing is not necessary.

Distributed 3D rendering system in a multi-agent platform

In this work, we propose a 3D rendering system that distributes rendering tasks across a multi-agent platform. The new approach is based on a multi-agent platform, where the goal is to create a virtual 3D environment. The main task is the rendering of individual objects. Each 3D object must be rendered in a remote unit; the resulting rendering is sent through the network to a 3D visualization process which generates the visualization of the whole 3D environment. The object movement and remote communication requirements have been implemented using a multi-agent system platform. The distributed system is implemented in Windows O.S., using DirectX graphical libraries and JAVA programming. The multi-agent platform used is JADE. The computer connection is a LAN at 100 MBS in a star topology.

EEG Pattern Recognition: An Efficient Improvement Combination of ERD/ERS/Laterality Features to Create a Self-paced BCI System

In this paper, a new method based on an efficient improvement combination of Event-Related Desynchronization (ERD), Event-Related Synchronization (ERS) and lateral activity of sensorimotor cortex features is presented to analyze both left and right hand motor imagery tasks. Our proposal uses delta, theta, alfa and beta rhythms to BCI system. From the spectral power, an efficient combination of ERD/ERS/laterality features was used. Because electroencephalogram signals are non-stationary type and highly vary over time and frequency, a detailed time-frequency analysis is applied. Features coming from time-frequency analysis, where eight frequency bands ranging from 0 to 32 Hz were chosen. Features vectors are classified by Gaussian classifier and the final performance is evaluated in cross-validation scheme. This novel approach was tested using the BCI competition IV data set 1. The detection of the left and right hand motor imagery task was very good, with a result of \(96.4\,\%\) using BCI-Competition -IV. When comparing results from others competing methods reported in the literature, our approach resulted the best and useful to create a self-paced BCI-system.

Monocular Visual Odometry Based Navigation for a Differential Mobile Robot with Android OS

In this work, a real time Monocular Visual Odometry system to estimate camera position and orientation based solely on image measurements is proposed. The system is built on the basis of the fundamentals of Structure from Motion theory, and requires only a single camera to estimate positional information. Experiments were conducted on flat ground, under controlled light conditions environment, in which and an Android mobile device camera was employed as the processor and the system sensor due to ease of acquisition and low price. The proposed system resulted in absolute navigation error rates ranging from 0.14% to 0.4% of the travelled distance at processing rates of up to 5Hz.

Unsupervised Font Clustering Using Stochastic Versio of the EM Algorithm and Global Texture Analysis

An Unsupervised Font clustering technique is proposed in this work. The new approach is based on global texture analysis, using high order statistic features, Gaussian classifier and a stochastic version of the EM algorithm. The font recognition is performed by taking the document as a simple image, where one or several types of fonts are present. The identification is not performed letter by letter as with conventional approaches. In the proposed method a window analysis is employed to obtain the features of the document, using fourth and third order moments. The new technique does not involve a study of local typography; therefore, it is content independent. A detailed study was performed with 8 types of fonts commonly used in the Spanish language. Each type of font can have four styles that lead, to 32 font combinations. The font recognition with clean images is 100% accurate.

A Genetic Algorithm Applied to Content-Based Image Retrieval for Natural Scenes Classification

The Content-Based Image Retrieval (CBIR) techniques comprise methodologies intended to retrieve self-content descriptors over the image data set being studied according to the type of the image. The main purpose of CBIR consists in classifying images avoiding the use of manual labels related to understanding of the image by the human being vision. In this work we provide a new CBIR procedure which works with local texture analysis, and which is developed in a non supervised fashion, clustering the local achieved descriptors and classifying them with the use of a K-means algorithm supported by the genetic algorithm. This method has been deployed in LabVIEW software, programming each part of the procedure in order to implement it in hardware. The results are very promising, reaching up to 90% of recall for natural scene classification.

EEG PATTERN RECOGNITION: Application to a Real Time Control System for Android-Based Mobile Devices

This paper describes a new EEG pattern recognition methodology in Brain Computer Interface (BCI) field. The EEG signal is analyzed in real time looking for detection of “intents of movement”. The signal is processed at specific segments in order to classify mental tasks then a message is formulated and sent to a mobile device to execute a command. The signal analysis is carried out through eight frequency bands within the range of 0 to 32 Hz. A feature vector is conformed using histograms of gradients according to 4 orientations, subsequently the features feed a Gaussian classifier. Our methodology was tested using BCI Competition IV data sets I. For “intents of movements” we detect up to 95% with 0.2 associated noise, with mental task differentiation around 99%. This methodology has been tested building a prototype using an Android based mobile telephone and data gathered with an EPOC Emotive headset, showing very promising results.

Finite-Element Calculation of the SSFR of Synchronous Machines

In this paper different finite-element (FE) techniques are effectively combined to obtain the standstill frequency response of synchronous machines. As a result, actual risky and high cost tests can be largely avoided. Lagrange multipliers are employed to joint two FE meshes of different densities at the air gap region. Explicit consideration of the external circuitry connected to the machine is performed through a simultaneous and effective solving of the field and circuit equations