Suet-Peng Yong

Image processing based vehicle detection and tracking method

Vehicle detection and tracking plays an effective and significant role in the area of traffic surveillance system where efficient traffic management and safety is the main concern. In this paper, we discuss and address the issue of detecting vehicle / traffic data from video frames. Although various researches have been done in this area and many methods have been implemented, still this area has room for improvements. With a view to do improvements, it is proposed to develop an unique algorithm for vehicle data recognition and tracking using Gaussian mixture model and blob detection methods. First, we differentiate the foreground from background in frames by learning the background. Here, foreground detector detects the object and a binary computation is done to define rectangular regions around every detected object. To detect the moving object correctly and to remove the noise some morphological operations have been applied. Then the final counting is done by tracking the detected objects and their regions. The results are encouraging and we got more than 91% of average accuracy in detection and tracking using the Gaussian Mixture Model and Blob Detection methods.

Unscented Kalman Filter for Noisy Multivariate Financial Time-Series Data

Kalman filter is one of the novel techniques useful for statistical estimation theory and now widely used in many practical applications. In this paper, we consider the process of applying Unscented Kalman Filtering algorithm to multivariate financial time series data to determine if the algorithm could be used to smooth the direction of KLCI stock price movements using five different measurement variance values. Financial data are characterized by non-linearity, noise, chaotic in nature, volatile and the biggest impediment is due to the colossal nature of the capacity of transmitted data from the trading market. Unscented Kalman filter employs the use of unscented transformation commonly referred to as sigma points from which estimates are recovered from. The filtered output precisely internments the covariance of noisy input data producing smoothed and less noisy estimates.

Scaled UKF---NARX hybrid model for multi-step-ahead forecasting of chaotic time series data

Accurate forecasting is critically important in many time series applications. In this paper, we consider forecasting chaotic problems by proposing a hybrid model composed of scaled unscented Kalman filter with reduced sigma points and non-linear autoregressive network with exogenous inputs, trained using a modified Bayesian regulation backpropagation algorithm. To corroborate developments of the proposed hybrid model, real-life chaotic and simulated time series which are both non-linear in nature are applied to validate the proposed hybrid model. Experiment results show that the proposed hybrid model outperforms other forecasting models reported in the literature in forecasting of chaotic time series.

Modality classification of medical images with distributed representations based on cellular automata reservoir computing

Modality corresponding to medical images is a vital filter in medical image retrieval systems. This article presents the classification of modalities of medical images based on the usage of principles of hyper-dimensional computing and reservoir computing. It is demonstrated that the highest classification accuracy of the proposed method is on a par with the best classical method for the given dataset (83% vs. 84%). The major positive property of the proposed method is that it does not require any optimization routine during the training phase and naturally allows for incremental learning upon the availability of new training data.

Ensemble classification with modified SIFT descriptor for medical image modality

The increasing number of medical images of various imaging modalities is challenging the accuracy and efficiency of radiologists. In order to retrieve the images from medical databases, radiologists will confine their search to the image modality. In this paper, we present an improved image feature to represent medical images for image modality classification. The proposed image descriptor is an ensemble descriptor that combines the Harris Corner encoded by the SIFT algorithm fused with Local Binary Pattern. Furthermore, we propose an ensemble classifier with surrogate splits to be used in medical image modality classification in order to improve the performance. It is shown that the proposed ensemble classifier with surrogate splits and ensemble descriptor encoded with bag-of-visual-words representation outperforms other conventional approaches applied in medical image modality classification.

Hybridization of Ensemble Kalman Filter and Non-linear Auto-regressive Neural Network for Financial Forecasting

Financial data is characterized as non-linear, chaotic in nature and volatile thus making the process of forecasting cumbersome. Therefore, a successful forecasting model must be able to capture long-term dependencies from the past chaotic data. In this study, a novel hybrid model, called UKF-NARX, consists of unscented kalman filter and non-linear auto-regressive network with exogenous input trained with bayesian regulation algorithm is modelled for chaotic financial forecasting. The proposed hybrid model is compared with commonly used Elman-NARX and static forecasting model employed by financial analysts. Experimental results on Bursa Malaysia KLCI data show that the proposed hybrid model outperforms the other two commonly used models.

Text Summarization in Android Mobile Devices

This paper presents a text summarization in Android mobile devices. With the proliferation of small screen devices and advancement of mobile technology, the text summarization research has been inspired by the new paradigm shift in accessing information ubiquitously at anytime, anywhere and anyway on mobile devices. However, it is a challenge to browse large documents in a mobile device because of its small screen size and information overload problems. In this paper, a semantic and syntactic based summarization was attempted and implemented in a text summarizer. The objectives of the paper are two-fold. (1) To integrate WordNet 3.1 into the proposed system called TextSumIt which condenses single lengthy document into shorter summarized text. (2) To provide better readability to Android mobile users by displaying the salient ideas in bullets points. Documents were collected from DUC 2002 and Reuter news datasets. Experimental results show that the text summarization model improves the accuracy, readability and time saving in the text summarizer as compared with MS Word AutoSummarize.

An Evaluation of Convolutional Neural Nets for Medical Image Anatomy Classification

Classification of the anatomical structures is an important precondition for several computer aided detection and diagnosis systems. Attaining extraordinary precision for automatic classification is a stimulating job because of vast amount of variation in the anatomical structures. Current trend in object recognition is driven by “Deep learning” methods that are outperforming the contemporary methods in classification of images. Till now these “Deep learning” methods have been applied on natural images. In this study, we compare the performance of three main Deep learning architectures i.e. LeNet, AlexNet, GoogLeNet on medical imaging data containing five anatomical structures for anatomic specific classification.

Text Summarization Using Latent Semantic Analysis Model in Mobile Android Platform

This paper presents the Latent Semantic Analysis (LSA) Model in Automatic Text Summarization (ATS) on single English document in mobile Android platform. Readers are drowned in information while starved of knowledge. Millions of articles are uploaded into the website every day. Quite often, lengthy text are presented in online articles but shorter summarized texts are preferred by readers. There exists research gap as most of the extractive text summarizations are based on syntactic appearance of words. Thus, the objective of this paper is to investigate the LSA Model by examining the semantic relationship between terms and sentences in a document for text summarization. We intend to shift our research paradigm to summarize text to infer the semantic contextual cues using the co-occurrence of terms in text. The input text documents were downloaded from Document Understanding Conference 2002 dataset. The preliminary results show that the LSA model yields an average F-Score of 0.386 in text summarization.

Surrogate reservoir modeling-prediction of Bottom-Hole Flowing Pressure using Radial Basis Neural Network

Reservoir simulation provides information about the behaviour of a reservoir in various production and injection conditions. Reservoir simulator is used to predict the future behaviour and performance of a reservoir field. However, the heterogeneity of reservoir and uncertainty in the reservoir field cause some obstacles in selecting the best calculation of oil, water and gas components that lead to the production system in oil and gas. Due to intrinsic uncertainty in the reservoir simulation models, large number of computational resources such as simulation runs and long processing time are required to predict the properties in a reservoir. This paper presents an application of Surrogate Reservoir Model (SRM) for predicting the Bottom-Hole Flowing Pressure (BHFP) at different time step for an initially under-saturated reservoir. The developed SRM is based on Artificial Neural Network to regenerate the results of a numerical simulation model in considerable amount of time. The output of the reservoir simulation consists of oil production, gas rate, average reservoir pressure, saturation and BHFP etc. The proposed SRM adopted Radial Basis Neural Network to predict the BHFP based on the output data extracted from the Black Oil Applied Simulation Tool (BOAST). It is found that the developed SRM is capable in supporting fast track analysis, decision optimization and manage to generate the results in a shorter time as compared to the conventional reservoir model.