RESEARCH

My research interests are in medical image analysis, computer vision, artificial intelligence, and computer graphics. My current and future research might provide interdisciplinary foundation from collaborations with researchers in life science, pathology, radiology, bio-nanotechnology, human-computer interaction, and visualization in virtual reality environment.

3D reconstruction technology development for scene of car accident using multi view black box image

Device and User-purpose Oriented Contents Reconstruction/Re-augmentation Software Framework from Massive-scale Streetview Information

We propose an integration framework based on information analysis, hierarchical data representation and reconstruction techniques on massive-scale multimedia-spatial data. Our research sub-goals include the followings: Hyper-perspective panoramic image construction technique, object type data reconstruction, interactive re-augmentation technique, user-centric, device optimized data representation technique, and large-scale multimedia data processing technique.

Content based video assesment based on color and optical flow pattern analyses

The main goal of the proposed research is in quantifying the factors that can be potentially harmful for video observers during watching video streams from camera input, computer games, and video contents. Previous work mainly focused on clinical trials about why and how people feel simulation sickness. In this research, we propose a contents based analysis framework that can efficiently quantify the factors about causing simulation sickness. In particular, we tackle the problem by analyzing color distribution and optical flows in consecutive video frames.

3D Volume Reconstruction from Confocal Laser Scanning Microscopy Imagery

(Joint work with Department of Pathology, University of Illinois at Chicago)

We study a three-dimensional volume reconstruction framework which consists of volume reconstruction procedures using multiple automation levels, feature types, and feature dimensionalities, a data-driven registration decision support system, an evaluation study of registration accuracy, and a novel intensity enhancement technique for 3D CLSM volumes.

The motivation for developing the framework came from the lack of 3D volume reconstruction techniques for CLSM image modality. The 3D volume reconstruction problem is challenging due to significant variations of intensity and shape of cross sectioned structures, unpredictable and inhomogeneous geometrical warping during medical specimen preparation, and an absence of external fiduciary markers. The framework addresses the problem of automation in the presence of the above challenges as they are frequently encountered during CLSM-based 3D volume reconstructions used for cell biology investigations.

The broader impact of the presented work is in providing the algorithms in a form of web-enabled tools to the medical community so that medical researchers can minimize laborious and time intensive 3D volume reconstructions using the tools and computational resources at NCSA.

3D Volume Reconstruction of Extracellular Matrix Proteins in Uveal Melanoma

We developed a method for the three-dimensional reconstruction of volume for these extravascular matrix proteins from serial paraffin sections cut at 4 µm thicknesses and stained with a fluorescent-labeled antibody to laminin. Each section was examined with confocal laser-scanning focal microscopy (CLSM) and 13 images were recorded in the Z-dimension for each slide. The input CLSM imagery is composed of a set of 3D sub-volumes (stacks of 2D images) acquired at multiple confocal depths, from a sequence of consecutive slides. Steps for automated reconstruction process included (1) unsupervised methods for selecting an image frame from a sub-volume based on entropy and contrast criteria, (2) a fully automated registration technique for image alignment, and (3) an improved histogram equalization method that compensates for spatially varying image intensities in CLSM imagery due to photo-bleaching. We compared image alignment accuracy of a fully automated method with registration accuracy achieved by human subjects using a manual method. In this study, automated 3D volume reconstruction provided significant improvement in accuracy, consistency of results, and performance time for CLSM images acquired from serial paraffin sections.

The distribution of looping patterns of laminin in uveal melanomas and other tumors has been associated with adverse outcome. Moreover, these patterns are generated by highly invasive tumor cells through the process of vasculogenic mimicry and are not therefore blood vessels. Nevertheless, these extravascular matrix patterns conduct plasma. The three-dimensional configuration of these laminin-rich patterns compared with blood vessels has been the subject of speculation and intensive investigation.

Accuracy Evaluation for Region Centroid-Based Registration

We introduce an accuracy evaluation of a semi-automatic registration technique for 3D volume reconstruction from fluorescent confocal laser scanning microscope (CLSM) imagery. The presented semi-automatic method is designed based on our observations that (a) an accurate point selection is much harder than an accurate region (segment) selection for a human, (b) a centroid selection of any region is less accurate by a human than by a computer, and (c) registration based on structural shape of a region rather than based on intensity-defined point is more robust to noise and to morphological deformation of features across stacks. We applied the method to image mosaicking and image alignment registration steps and evaluated its performance with 20 human subjects on CLSM images with stained blood vessels. Our experimental evaluation showed significant benefits of automation for 3D volume reconstruction in terms of achieved accuracy, consistency of results and performance time. In addition, the results indicate that the differences between registration accuracy obtained by experts and by novices disappear with the proposed semi-automatic registration technique while the absolute registration accuracy increases.

Three-dimensional Volume Reconstruction Based on Trajectory Fusion

We address the problem of 3D volume reconstruction from depth adjacent sub-volumes (i.e., sets of image frames) acquired using confocal laser scanning microscopy (CLSM). Our goal is to align sub-volumes by estimating an optimal global image transformation which preserves morphological smoothness of medical structures (called features, for instance blood vessels) inside of a reconstructed 3D volume. We approached the problem by learning morphological characteristics of structures inside of each sub-volume (e. g., centroid trajectories of features). Next, adjacent sub-volumes are aligned by fusing the morphological characteristics of structures using ex-trapolation and model fitting. Finally, a global sub-volume to sub-volume transformation is computed based on the entire set of fused structures. The trajectory-based 3D volume reconstruction method described here is evaluated with four consecutive physical sections and using two metrics for morphological continuity.

Intensity Correction of Fluorescent Confocal Laser Scanning Microscope Images by Mean-Weight Filtering

We addresses the problem of intensity correction of fluorescent confocal laser scanning microscope (CLSM) images. CLSM images are frequently used in medical domain for obtaining 3D information about specimen structures by imaging a set of 2D cross sections and performing 3D volume reconstruction afterwards. However, 2D images acquired from fluorescent CLSM images demonstrate significant intensity heterogeneity, for example, due to photo-bleaching and fluorescent attenuation in depth. We developed an intensity heterogeneity correction technique that (a) adjusts intensity heterogeneity of 2D images, (b) preserves fine structural details, and (c) enhances image contrast, by performing spatially adaptive mean-weight filtering. Our solution is obtained by formulating an optimization problem, followed by filter design and automated selection of filtering parameters. The proposed filtering method is experimentally compared with several existing techniques by using four quality metrics, such as contrast, intensity heterogeneity (entropy) in low frequency domain, intensity distortion in high frequency domain, and saturation. Based on our experiments and the four quality metrics, the developed mean-weight filtering outperforms other intensity correction methods by at least a factor of 1.5 when applied to fluorescent CLSM images.

Automated Hand-filled Form Analysis

The project considers an automated information extraction from hand-filled forms, specifically used by the Department of Human Service (DHS), Illinois. The process includes distribution and re-collection of the printed forms for traditional hand-written filing, raw image acquisition of the collected forms, and semi- and fully automated information extraction for future data mining support. The automated data extraction from a hand-filled form is considered as a hard problem unless a standardized form is used, e.g., teleform. For information extraction without modifying the existing form, we propose automated image processing techniques such as robust registration of geometrically distorted input images, segmentation of the input fields, and classification of binary answers with high confidence. The processed data may be stored as a pdf form with annotations or a plain text form associated with raw images.

Data Integration and Information Gathering about Decision Processes using Geospatial Electronic Records

The size, complexity and heterogeneity of geospatial collections pose formidable challenges for the National Archives in terms of high performance data storage, access, integration, analysis and visualization. These challenges become even more eminent when temporal aspects of data processing are involved as in the context of high assurance decision making and high confidence application scenarios. As part of this project, we address the problems of gathering, archiving and analyzing information about decision-making processes using geospatial electronic records (e-records). The ultimate goal of our research is to understand the cost of information archival using the cutting edge technologies, high performance computing and novel computer architectures.

Network for Earthquake Engineering Simulation (NEES)

(Joint work with Department of Civil and Environmental Engineering, University of Illinois at Urbana-Champaign)

We address the problem of multi-instrument analysis of point and raster data. An image (raster data) is one of the most popular data in many applications due to its low cost and flexibility. However, when using image data only, many applications may face other problems, such as camera calibration, limited field of view and resolution and low accuracy comparing with contact sensors. In addition to image data, our research includes point-based sensors that are being used to obtain more accurate measurements in terms of spatial sampling and value precision.

The measurement process with multiple sensors poses several challenges on multi-instrument data analysis, for example, sensor registration, data interpolation, variable transformation, data overlay and spatial measurement evaluation. We will present a method of sensing, data conversion and data verification and describe our preliminary results from the on-going research investigating large material structures. Our work is part of the National Earthquake Engineering Simulation (NEES) project and is conducted in the collaboration with the Civil and Environmental Engineering Department, UIUC.

Autonomous MobileRobot Navigation using Omni-directional Vision

We addresses the problem of systematic exploration of an unfamiliar world environment to build a qualitative map by searching for recognizable targets. Generated maps are essential for further mobile robot control, self-localization, and path planning. While exploring, a map is constructed which contains a set of regions of free-space delineated by recognizable targets (landmarks) and the connectivity (adjacency and overlap) of these regions. Within a region, the robot can freely navigate without collision and reference its position with respect to at least three landmarks. This is achieved by exploiting the visibility constraint -- if a landmark is visible to a robot, there are no obstacles on the line segment between the robot and the landmark. As the robot moves along a trajectory while tracking a landmark, a region of free-space is swept out. By representing a collection of free regions and their region-to-region connectivity as a graph, path planning amounts to graph sarch and execution of the plan by a robot amounts to local movements to enter into the next free region. Both for exploration and for navigation, no metric information about the robot's path nor absolute coordinates of its position or of landmark locations are required or recorded. After exploring, the robot produces a compact map which covers a large area, and provides information for fast self-localization and flexible path planning.