Manohar Mareboyana

First evaluation of automatic image registration methods

As the need for automating registration techniques is recognized, the authors feel that there is a need to survey all the registration methods which may be applicable to Earth and space science problems and to evaluate their performances on a large variety of existing remote sensing data as well as on simulated data of soon-to-be-flown instruments. The authors present the first steps towards this quantitative evaluation: a few automatic image registration algorithms are described and first results of their evaluation are presented for three different datasets.

Similarity search in sensor networks using semantic-based caching

Sensor networks build temporary wireless connections in environments where the stationary infrastructures are either destroyed or too expensive to construct. Most of the previous research in sensor networks focuses on routing protocols that adapt to the dynamic network topologies, and not much work has been done on data accessing. One important data accessing application is similarity search, which provides the foundation of content-based retrieval. Many traditional similarity search algorithms are based on centralized or flooding mechanisms, which are not effective in wireless sensor network environments due to the multiple limitations such as bandwidth and power. In this paper we tackle the problem of similarity search by using semantic-based caching to reflect the data content distribution in the network. The basic idea is analyzing the cached results of earlier queries and trying to resolve the later queries within a small collection of content-related mobile nodes. Based on a Hilbert space-filling curve, the data points in a multi-dimensional semantic space are described as a linear representation. These data points are further cached to facilitate query processing. Through extensive simulations, we show that our method can perform similarity search with improved performance in terms of search cost and response time.

GPU accelerated one-pass algorithm for computing minimal rectangles of connected components

The connected component labeling is an essential task for detecting moving objects and tracking them in video surveillance application. Since tracking algorithms are designed for real-time applications, efficiencies of the underlying algorithms become critical. In this paper we present a new one-pass algorithm for computing minimal binding rectangles of all the connected components of background foreground segmented video frames (binary data) using GPU accelerator. The given image frame is scanned once in raster scan mode and the background foreground transition information is stored in a directed-graph where each transition is represented by a node. This data structure contains the locations of object edges in every row, and it is used to detect connected components in the image and extract its main features, e.g. bounding box size and location, location of the centroid, real size, etc. Further we use GPU acceleration to speed up feature extraction from the image to a directed graph from which minimal bounding rectangles will be computed subsequently. Also we compare the performance of GPU acceleration (using Tesla C2050 accelerator card) with the performance of multi-core (up 24 cores) general purpose CPU implementation of the algorithm.

Parallel algorithm for linear feature detection from airborne LiDAR data

Linear features from airport images correspond to runways, taxiways and roads. Detecting runways helps pilots to focus on runway incursions in poor visibility conditions. In this work, we attempt to detect linear features from LiDAR swath in near real time using parallel implementation on G5-based apple cluster called Xseed. Data from LiDAR swath is converted into a uniform grid with nearest neighbor interpolation. The edges and gradient directions are computed using standard edge detection algorithms such as Cannys detector. Edge linking and detecting straight-line features are described. Preliminary results on Reno, Nevada airport data are included.

High Resolution Image Reconstruction from Projection of Low Resolution Images DIffering in Subpixel Shifts

In this paper, we demonstrate simple algorithms that project low resolution (LR) images differing in subpixel shifts on a high resolution (HR) also called super resolution (SR) grid. The algorithms are very effective in accuracy as well as time efficiency. A number of spatial interpolation techniques using nearest neighbor, inverse-distance weighted averages, Radial Basis Functions (RBF) etc. are used in projection. For best accuracy of reconstructing SR image by a factor of two requires four LR images differing in four independent subpixel shifts. The algorithm has two steps: i) registration of low resolution images and (ii) shifting the low resolution images to align with reference image and projecting them on high resolution grid based on the shifts of each low resolution image using different interpolation techniques. Experiments are conducted by simulating low resolution images by subpixel shifts and subsampling of original high resolution image and the reconstructing the high resolution images from the simulated low resolution images. The results of accuracy of reconstruction are compared by using mean squared error measure between original high resolution image and reconstructed image. The algorithm was tested on remote sensing images and found to outperform previously proposed techniques such as Iterative Back Projection algorithm (IBP), Maximum Likelihood (ML) algorithms. The algorithms are robust and are not overly sensitive to the registration inaccuracies.

Caching for Location-Aware Image Queries in Mobile Networks

Caching has been widely used in mobile networks to improve system performance. However, conventional caching methodologies have two major drawbacks in dealing with spatial queries in a dynamic mobile network: (i) the description of cached data is defined based on the query context instead of data content ignoring the spatial or semantic locality of the data. (ii) the description of cached data does not reflect the popularity of the data, making it inefficient in providing QoS-related services. To address these issues, we propose a location-aware caching (LAC) model which reflects the distribution of images based on the analysis of earlier queries. The novelty of our method stems from several factors including: 1) describing the image data distribution based on a Hilbert space-filling curve, 2) optimizing spatial query resolution through efficient exploitation of locally cached data, and 3) reducing the cost of query resolution with restricted search scope. Through extensive simulations, we show that our model can perform spatial search with less cost. In addition, it is scalable to large environments and voluminous data.

Semantic Video Clustering in Ad Hoc Networks for Content-Based Retrieval

Traditional content-based retrieval approaches employ either centralized or flooding strategies in ad hoc networks, which may result in low fault tolerance and high search cost making them inefficient. To facilitate an efficient video retrieval, we propose a logic-based content summary framework that is able to represent semantic contents of video data using concise logic terms. In this method the video data is characterized by color and wavelet coefficients which will be converted into logical terms by using threshold operators. The logical terms are then summarized as node content descriptions. The nodes containing similar node descriptions are clustered into a virtual infrastructure according to the semantic content.

Homography based distributed video coding for a network of cameras

Networks of multiple video cameras are being deployed in several scenarios like surveillance, traffic enforcement, human motion analysis and sports telecast. The unmanageable size of the raw video sequences necessitates the use of compression schemes to efficiently encode these videos. Traditional video compression schemes account for spatial and temporal redundancy in a video sequence. In this paper, we present a compression technique that also leverages the information redundancy between video sequences across different cameras with overlapping fields of view. An algorithm based on homography for efficient compression of multiple video sequences is presented that performs significantly better than current schemes. We also derive an efficient distributed version of the algorithm that can be implemented on a large scale network of cameras. This distributed algorithm minimizes both the communication costs and the compression costs simultaneously.

Undergraduate Student Retention Prediction Using Wavelet Decomposition

In this paper, we have presented some results on undergraduate student retention using signal processing techniques for classification of the student data. The experiments revealed that the main factor that influences student retention in the Historically Black Colleges and Universities (HBCU) is the cumulative grade point average (GPA). The linear smoothing of the data helped remove the noise spikes in data thereby improving the retention results. The data is decomposed into Haar coefficients that helped accurate classification. The results showed that the HBCU undergraduate student retention corresponds to an average GPA of 2.8597 and the difference of −0.023307. Using this approach we obtained more accurate retention results on training data.

Predictive Models for Undergraduate Student Retention Using Machine Learning Algorithms

In this paper, we have presented some results of undergraduate student retention using machine learning and wavelet decomposition algorithms for classifying the student data. We have also made some improvements to the classification algorithms such as Decision tree, Support Vector Machines (SVM), and neural networks supported by Weka software toolkit. The experiments revealed that the main factors that influence student retention in the Historically Black Colleges and Universities (HBCU) are the cumulative grade point average (GPA) and total credit hours (TCH) taken. The target functions derived from the bare minimum decision tree and SVM algorithms were further revised to create a two-layer neural network and a regression to predict the retention. These new models improved the classification accuracy. Furthermore, we utilized wavelet decomposition and achieved better results.