Mojtaba Bandarabadi

Combining wavelet transforms and neural networks for image classification

A new approach for image classification based on the color information, shape and texture is presented. In this work, we use the three RGB bands of a color image in RGB model to extract the describing features. All the images in image database are divided into 6 parts. We use the Daubechies 4 wavelet transform and first order color moments to obtain the necessary information from each part of the image. The proposed image classification system is based on Back propagation neural network with one hidden layer. Color moments and wavelet decomposition coefficients from each part of the image are used as an input vector of neural network. 150 color images of aircrafts were used for training and 250 for testing. The best efficiency of 98% was obtained for training set, and 90% for the testing set.

Mean shift-based object tracking with multiple features

This paper presents visual features for tracking of moving object in video sequences using Mean Shift algorithm. The features used in this paper are color, edge and texture. Mean shift Algorithm is expanded based on mentioned multiple features, which are described with highly nonlinear models. In the proposed method, firstly all the features is extracted from first frame and the histogram of each feature is computed then the mean shift algorithm is run for each feature independently and the output of the mean shift algorithm for each feature is weighted based on the similarity measure. In last step, center of the target in the new frame is computed through the integration of the outputs of mean shift. We show that tracking with multiple weighted features provides more reliable performance than single features tracking.

Airplane detection and tracking using wavelet features and SVM classifier

In this paper we explain a fully automatic system for airplane detection and tracking based on wavelet transform and Support Vector Machine (SVM). By using 50 airplane images in different situations, models are developed to recognize airplane in the first frame of a video sequence. To train a SVM classifier for classifying pixels belong to objects and background pixels, vectors of features are built. The learned model can be used to detect the airplane in the original video and in the novel images. For original video, the system can be considered as a generalized tracker and for novel images it can be interpreted as method for learning models for object recognition. After airplane detection in the first frame, the feature vectors of this frame are used to train the SVM classifier. For new video frame, SVM is applied to test the pixels and form a confidence map. The 4th level of Daubechiess wavelet coefficients corresponding to input image are used as features. Conducting simulations, it is demonstrated that airplane detection and tracking based on wavelet transform and SVM classification result in acceptable and efficient performance. The experimental results agree with the theoretical results.

Target Tracking Using Wavelet Features and RVM Classifier

In this paper, a new method is proposed for target tracking based on wavelet transform and relevance vector machine (RVM). Considering tracking as a classification problem, we train a RVM classifier to distinguish an object from its background. This is done by constructing feature vector for every pixel in the reference image and then training a RVM classifier to separate pixels which belong to the object from those related to the background. Receiving new video frame, RVM is employed to test the pixels and form a confidence map. In this work, the features we use the 4th level Daubechiespsilas wavelet coefficients corresponding to input image. Conducting simulations, it is demonstrated that target tracking based on wavelet transform and RVM classification result in acceptable and efficient performance. The experimental results agree with the theoretical results.

Modify kernel tracking using an efficient color model and active contour

In this paper, a new method for contour tracking of mobile target in video sequences is presented. Proposed method helps to track variety of targets exactly while the camera is moving. In this study, a new type of active contour is used with a way for estimating motion model of the target. The novelty of this paper comes from using a new color to gray level transform instead of conventional counterpart. Estimating motion model of the target uses color and location information together. This information helps the proposed method to be robust against the aspect change. Our method is compared with two other methods: tracking via single active contour and estimating motion model by using affine transform. Experimental results illustrate our method to outperform the prior ones.

Lightweight blocking coordinated checkpointing for cluster computer systems

In this paper we introduce a new approach for blocking coordinated checkpointing using two level checkpointing for high performance cluster computing systems. First level of checkpointing is local checkpointing and computing nodes save the checkpoints in local disk based on transient failure rates. If a transient failure occurs in the computing node, process can recover from local disk. Second level of checkpointing is global checkpointing and computing nodes send their checkpoints to high reliable global stable storage in network based on the permanent failure rate. If a permanent failure occurs in the computing node, computing node can not be used and process can recover from global storage in a new computing node. Transient failures are probable than permanent failures and the number of global checkpointingis very lower than local checkpointing. Based on this method, coordinated checkpointing overhead is reduced and it is proportional to transient and permanent failure rates of cluster systems.

Proactive blocking coordinated checkpointing with dynamic intervals

In this paper we introduce a new proactive blocking coordinated checkpointing for cluster computing systems with dynamic interval. Many current schemes to increase the availability of cluster computing systems either make use of redundancy in space or redundancy in time (reactive methods). These methods induce the overhead to the cluster computing system in failure free execution time. In order to minimize the performance loss (rollback and checkpoint overheads) due to unexpected failures or unnecessary overhead of fault tolerant mechanisms, we present a proactive method for the blocking coordinated checkpointing strategy. Existing checkpointing methods are static with constant checkpointing interval. These methods are based on the exponential distribution function. In this paper we use the weibull distribution function to find the dynamic interval. Our method is based on the failure data analysis of LANL cluster system. Experimental results show that average execution time of NAS application is significantly reduced by using the proposed method.

Adaptive two-level blocking coordinated checkpointing based on recovery cost

In this paper we introduce a new adaptive two-level blocking coordinated checkpointing for cluster computing systems. First level of checkpointing is local checkpointing and computing nodes save the checkpoints in local disk based on transient failure rates. If a transient failure occurs in the computing node, process can recover from local disk. Second level of checkpointing is global checkpointing and computing nodes send their checkpoints to high reliable global stable storage in network based on the expected recovery time in the case of permanent failure. If a permanent failure occurs in the computing node, computing node can not be used and process can recover from global storage in a new computing node. Transient failures are probable than permanent failures and the number of global checkpointingis very lower than local checkpointing. Based on this method, coordinated checkpointing overhead is reduced and it is proportional to transient and permanent failure rates of cluster systems.

Metric distance transform for kernel based object tracking

An object tracking algorithm that uses the flexible kernels based on the normalized Metric dα Distance Transform for the Mean shift procedure is proposed and tested. This replaces the more usual Epanechnikov kernel (E-kernel), improving target representation and localization without increasing the processing time, minimizing the similarity measure using the Bhattacharya coefficient. The target shape which defines the dα Distance Transform is found either by regional segmentation or background-difference imaging, dependent on the nature of the video sequence .The algorithm is tested on several image sequences and shown to achieve robust and reliable frame-rate tracking.