Howida A. Shedeed

Brain EEG signal processing for controlling a robotic arm

Researchers recently proposed new scientific methods for restoring function to those with motor impairments. one of these methods is to provide the brain with a new non-muscular communication and control channel, a direct Brain-Machine Interface (BMI). This paper presents a Brain Machine Interface (BMI) system based on using the brain electroencephalography (EEG) signals associated with 3 arm movements (close, open arm and close hand) for controlling a robotic arm. Signals recorded from one subject using Emotive Epoc device. Four channels only were used, in our experiment, AF3, which located at the prefrontal cortex and F7, F3, FC5 which located at the supplementary motor cortex of the brain. Three different techniques were used for features extraction which are: Wavelet Transform (WT), Fast Fourier Transform (FFT) and Principal Component Analysis (PCA). Multi-layer Perceptron Neural Network trained by a standard back propagation algorithm was used for classifying the three considered tasks. Classification rates of 91.1%, 86.7% and 85.6% were achieved with the three used features extraction techniques respectively. Experimental results show that the proposed system achieved high classification rates than other systems in the same application.

New technique for online object tracking-by-detection in video

Object detection and tracking is an important task within the field of computer vision, because of its promising application in many areas, such as video surveillance. The need for automated video analysis has generated a great deal of interest in the area of motion tracking. A new technique is proposed for online object tracking-by-detection capable of achieving high detection and tracking rates, using a stationary camera, in a particle filtering framework. The fundamental innovation is that the detection technique integrates the local binary pattern texture feature, the red green blue (RGB) colour feature and the Sobel edge feature, using ‘Choquet’ fuzzy integral to avoid uncertainty in the classification. This is performed by extracting the colour and edge grey scale confidence maps and introducing the texture confidence map. Then, the tracking technique makes use of the continuous confidence detectors, extracted from those confidence maps, along with another three introduced classifier confidence maps, extracted from an online boosting classifier. Finally, both the confidence detectors and the classifier maps are integrated in the particle filtering framework, using the Choquet integral. Experimental results for both indoor and outdoor dataset sequences confirmed the robustness of the proposed technique against illumination variation and scene motion.

Cloud‐based parallel solution for estimating statistical significance of megabyte‐scale DNA sequences

Confidence in a pairwise local sequence alignment is a fundamental problem in bioinformatics. For huge DNA sequences, this problem is highly compute‐intensive because it involves evaluating hundreds of local alignments to construct an empirical score distribution. Recent parallel solutions support only kilobyte‐scale sequence sizes and/or are based on sophisticated infrastructures that are not available for most of the research labs. This paper presents an efficient parallel solution for evaluating the statistical significance for a pair of huge DNA sequences using cloud infrastructures. This solution can receive requests from various researchers via web‐portal and allocate resources according to their demand. In this way, the benefits of cloud‐based services can be achieved. The fundamental innovation of this research work is proposing an efficient solution that utilizes both shared and distributed memory architectures via cloud technology to enhance the performance of evaluating the statistical significance for pair of DNA sequences. Therefore, the restriction on the sequence sizes is released to be in megabyte‐scale, which was not supported before for the statistical significance problem. The performance evaluation of the proposed solution was carried out on Microsofts cloud and compared with the existing parallel solutions. The results show that the processing speed outperforms the recent cluster solutions that target the same problem. In addition, the performance metrics exhibit linear behavior for the addressed number of instances. Copyright

An efficient solution for aligning huge DNA sequences

Recently, many parallel solutions were proposed in order to accelerate the exact methods for optimal alignment of DNA sequences. However, most of these solutions calculate only the alignment similarity score without finding the actual alignment. This paper presents an efficient solution to find the optimal alignment of the huge DNA sequences. This solution releases the condition of the sequence size to be in megabyte-scale instead of few kilobytes. The fundamental innovation in this work is developing an efficient, linear space complexity, parallel solution to achieve the optimum alignment with relatively good performance. The shared memory parallel architecture is the focus of this work therefore; we have considered off-the-shelf systems like multi-core CPUs as well as advanced shared memory platforms. Experimental results show that the proposed solution achieved high records compared to other solutions that targeted the same goal with less hardware requirements.

A Study of Action Recognition Problems: Dataset and Architectures Perspectives

Action recognition field has recently grown dramatically due to its importance in many applications like smart surveillance, human–computer interaction, assisting aged citizens or web-video search and retrieval. Many research trials have tackled action recognition as an open problem. Different datasets are built to evaluate architectures variations. In this survey, different action recognition datasets are explored to highlight their ability to evaluate different models. In addition, for each dataset, a usage is proposed based on the content and format of data it includes, the number of classes and challenges it covers. On other hand, another exploration for different architectures is drawn showing the contribution of each of them to handle different action recognition problem challenges and the scientific explanation behind their results. An overall of 21 datasets is covered with 13 architectures that are shallow and deep models.

EEG-EOG based Virtual Keyboard: Toward Hybrid Brain Computer Interface

The past twenty years have ignited a new spark in the research of Electroencephalogram (EEG), which was pursued to develop innovative Brain Computer Interfaces (BCIs) in order to help severely disabled people live a better life with a high degree of independence. Current BCIs are more theoretical than practical and are suffering from numerous challenges. New trends of research propose combining EEG to other simple and efficient bioelectric inputs such as Electro-oculography (EOG) resulting from eye movements, to produce more practical and robust Hybrid Brain Computer Interface systems (hBCI) or Brain/Neuronal Computer Interface (BNCI). Working towards this purpose, existing research in EOG based Human Computer Interaction (HCI) applications, must be organized and surveyed in order to develop a vision on the potential benefits of combining both input modalities and give rise to new designs that maximize these benefits. Our aim is to support and inspire the design of new hBCI systems based on both EEG and EOG signals, in doing so; first the current EOG based HCI systems were surveyed with a particular focus on EOG based systems for communication using virtual keyboard. Then, a survey of the current EEG-EOG virtual keyboard was performed highlighting the design protocols employed. We concluded with a discussion of the potential advantages of combining both systems with recommendations to give deep insight for future design issues for all EEG-EOG hBCI systems. Finally, a general architecture was proposed for a new EEG-EOG hBCI system. The proposed hybrid system completely alters the traditional view of the eye movement features present in EEG signal as artifacts that should be removed; instead EOG traces are extracted from EEG in our proposed hybrid architecture and are considered as an additional input modality sharing control according to the chosen design protocol.

PFastNCA: Parallel Fast Network Component Analysis for Gene Regulatory Network

One of the gene expression data analysis tasks is the Gene regulatory network analysis. Gene regulatory network is concerned in the topological organization of genes interactions. Moreover, the regulatory network is important for understanding the normal cell physiology and pathological phenotypes. However, the main challenge facing gene regulatory network algorithms is the data size. Where, the algorithm runtime is proportional to the data size. This paper presents a parallel algorithm for gene regulatory network (PFastNCA) which is an improved version of FastNCA. PFastNCA enhanced the main core of FastNCA which is the connectivity matrix estimation using a distributed computing model. Where, the work is divided among N processing nodes, PFastNCA is more efficient than FastNCA. It also achieved a better performance and speedup reached 1.91.

Improving Land-Cover and Crop-Types Classification of Sentinel-2 Satellite Images

Land cover and crop-types classification are of great importance for monitoring agricultural production and land-use patterns. Many classification approaches have used different parameters settings. In this paper, we investigate the modern classifiers using the most effective parameters to improve the classification accuracy of the major crops and land covers that exist in Sentinel-2 images for Fayoum region of Egypt. Four major crop-types and four major land-cover types are classified. This paper investigates the k-Nearest Neighbor (k-NN), Artificial Neural Network (ANN), Support Vector Machine (SVM), and Random Forest (RF) supervised classifiers. The experimental results show that the SVM and the RF report more robust results. The k-NN reports the least accuracy especially for crop types. The RT, K-NN, ANN, and SVM record 92.7%, 92%, 92.1% and 94.4% respectively. The SVM classifier out-performs the k-NN, ANN and RF classifiers.

Content Based Image Retrieval Using Local Feature Descriptors on Hadoop for Indoor Navigation

This paper demonstrates Content Based Image Retrieval (CBIR) algorithms implementation on a huge image set. Such implementation will be used to match query images to previously stored geotagged image database for the purpose of vision based indoor navigation. Feature extraction and matching are demonstrated using the two famous key-point detection CBIR algorithms: Scale Invariant Feature Transformation (SIFT) and Speeded Up Robust Features (SURF). The key-points matching results using Brute Force and FLANN (Fast Library for Approximate Nearest Neighbors) on various levels for both SIFT and SURF algorithms are compared herein. The algorithms are implemented on Hadoop MapReduce framework integrated with Hadoop Image Processing Interface (HIPI) and Open Computer Vision Library (OpenCV). As a result, the experiments shown that using SIFT with KNN (4, 5, and 6) levels give the highest matching accuracy in comparison to the other methods.

Comparing Multi-class Approaches for Motor Imagery Using Renyi Entropy

One of the main problems that face Motor Imagery-based system is addressing multi-class problem. Various approaches have been used to tackle this problem. Most of these approaches tend to divide multi-class problem into binary sub problems. This study aims to address the multi-class problem by comparing five multi-class approaches; One-vs-One (OVO), One-vs-Rest (OVR), Divide & Conquer (DC), Binary Hierarchy (BH), and Multi-class approaches. Renyi entropy was examined for feature extraction. Three linear classifiers were used to implement these five-approaches: Support Vector Machine (SVM), Multinomial Logistic Regression (MLR) and Linear Discriminant Analysis (LDA). These approaches were compared according to their performance and time consumption. The comparative results show that, Renyi entropy demonstrated its robustness not only as a feature extraction technique but also as a powerful dimension reduction technique, for multi-class problem. In addition, LDA proved to be the best classifier for almost all approaches with minimum execution time.