A5035964079- Advanced Vision and Imaging
- Retinal Imaging and Analysis
- Robotics and Sensor-Based Localization
- Medical Image Segmentation Techniques
- Advanced Image and Video Retrieval Techniques
- Electoral Systems and Political Participation
- Image and Object Detection Techniques
- Identification and Quantification in Food
- Species Distribution and Climate Change
- Glaucoma and retinal disorders
- Optical measurement and interference techniques
- Digital Imaging for Blood Diseases
- Video Surveillance and Tracking Methods
- Image Processing Techniques and Applications
- Retinal Diseases and Treatments
- Wildlife Ecology and Conservation
- Marine animal studies overview
- Internet Traffic Analysis and Secure E-voting
- Social Media and Politics
- Cell Image Analysis Techniques
- Retinal and Optic Conditions
- Remote Sensing and LiDAR Applications
- Media Influence and Politics
- Judicial and Constitutional Studies
- Environmental DNA in Biodiversity Studies
Rensselaer Polytechnic Institute
2013-2025
The Ohio State University
2023-2024
Massachusetts Institute of Technology
2014-2023
The London College
2023
Princeton University
2023
Mpala Research Center and Wildlife Foundation
2023
University of Wisconsin–Madison
1988-2021
Moscow Institute of Thermal Technology
2013
Troy University
2011
Williams & Associates
2010
Estimation techniques in computer vision applications must estimate accurate model parameters despite small-scale noise the data, occasional large-scale measurement errors (outliers), and measurements from multiple populations same data set. Increasingly, robust estimation techniques, some borrowed statistics literature others described literature, have been used solving these parameter problems. Ideally, should effectively ignore outliers other populations, treating them as outliers, when...
Motivated by the goals of improving detection low-contrast and narrow vessels eliminating false detections at nonvascular structures, a new technique is presented for extracting in retinal images. The core likelihood ratio test that combines matched-filter responses, confidence measures vessel boundary measures. Matched filter responses are derived scale-space to extract widely varying widths. A measure defined as projection vector formed from normalized pixel neighborhood onto ideal...
Motivated by the problem of retinal image registration, this paper introduces and analyzes a new registration algorithm called Dual-Bootstrap Iterative Closest Point (Dual-Bootstrap ICP). The approach is to start from one or more initial, low-order estimates that are only accurate in small regions, bootstrap regions. In each region, iteratively: 1) refines transformation estimate using constraints within region; 2) expands 3) tests see if higher order model can be used, stopping when region...
This paper describes a robust hierarchical algorithm for fully-automatic registration of pair images the curved human retina photographed by fundus microscope. Accurate is essential mosaic synthesis, change detection, and design computer-aided instrumentation. Central to 12-parameter interimage transformation derived modeling as rigid quadratic surface with unknown parameters. The parameters are estimated matching vascular landmarks recursively tracing blood vessel structure. parameter...
Kernel-based objective functions optimized using the mean shift algorithm have been demonstrated as an effective means of tracking in video sequences. The resulting algorithms combine robustness and invariance properties afforded by traditional density-based measures image similarity, while connecting these techniques to continuous optimization algorithms. This paper demonstrates a connection between kernel-based more template methods. here is well known equivalence function SSD-like measure...
Our goal is an automated 2D-image-pair registration algorithm capable of aligning images taken a wide variety natural and man-made scenes as well many medical images. The should handle low overlap, substantial orientation scale differences, large illumination variations, physical changes in the scene. An important component this ability to automatically reject pairs that have no overlap or too differences be aligned well. We propose complete including techniques for initialization,...
MINPRAN is a new robust estimator capable of finding good fits in data sets containing more than 50% outliers. Unlike other techniques that handle large outlier percentages, does not rely on known error bound for the data. Instead, it assumes bad are randomly distributed within dynamic range sensor. Based this, uses random sampling to search fit and inliers least likely have occurred randomly. It runs time O(N/sup 2/+SN log N), where S number samples N points. We demonstrate analytically...
A fully automated approach is presented for robust detection and classification of changes in longitudinal time-series color retinal fundus images diabetic retinopathy. The method to: 1) spatial variations illumination resulting from instrument limitations both within, between patient visits; 2) imaging artifacts such as dust particles; 3) outliers the training data; 4) segmentation alignment errors. Robustness to variation achieved by a novel iterative algorithm estimate reflectance retina...
We present HotSpotter, a fast, accurate algorithm for identifying individual animals against labeled database. It is not species specific and has been applied to Grevy's plains zebras, giraffes, leopards, lionfish. describe two approaches, both based on extracting matching keypoints or "hotspots". The first tests each new query image sequentially database image, generating score in isolation, ranking the results. second, building recent techniques instance recognition, matches using fast...
We present a novel dataset for animal behavior recognition collected in-situ using video from drones flown over the Mpala Research Centre in Kenya. Videos DJI Mavic 2S January 2023 were acquired at 5.4K resolution accordance with IACUC protocols, and processed to detect track each frames. An image subregion centered on was extracted combined sequence form "mini-scene". Be-haviors then manually labeled frame of mini-scene by team annotators overseen an expert behavioral ecologist. The...
A model-based algorithm, termed exclusion region and position refinement (ERPR), is presented for improving the accuracy repeatability of estimating locations where vascular structures branch cross over, in context human retinal images. The goal two fold. First, accurate morphometry branching crossover points (landmarks) neuronal/vascular structure important to several areas biology medicine. Second, these are valuable as landmarks image registration, so improved their signatures leads more...
This paper proposes a 5-component detection pipeline for use in computer vision-based animal recognition system. The end result of our proposed is collection novel annotations interest (AoI) with species and view-point labels. These AoIs, example, could be fed as the focused input data into an appearance-based identification goal method to increase reliability automation censusing studies provide better ecological information conservationists. Our able achieve localization mAP 81.67%,...
This paper is the result of a nationwide study polling place dynamics in 2016 presidential election. Research teams, recruited from local colleges and universities located twenty-eight election jurisdictions across United States, observed timed voters as they entered queue at their respective places then voted. We report results about four specific operations practices: length check-in line, number leaving line once have joined it, time for voter to check vote (i.e., verify voter’s...
Abstract Determining which species are at greatest risk, where they most vulnerable, and what the trajectories of their communities populations is critical for conservation management. Globally distributed, wide-ranging whales dolphins present a particular challenge in data collection because no single research team can record over biologically meaningful areas. Flukebook.org an open-source web platform that addresses these gaps by providing researchers with latest computational tools. It...
When fitting models to data containing multiple structures, such as when surface patches taken from a neighborhood that includes range discontinuity, robust estimators must tolerate both gross outliers and pseudo outliers. Pseudo are the structure of interest, but inliers different structure. They differ because their coherence. Such occurs frequently in computer vision problems, including motion estimation, model fitting, analysis. The focus this paper is problem surfaces near...
An algorithm for constructing image mosaics from multiple, uncalibrated, weak-perspective views of the human retina is presented and analyzed. It builds on an registering pairs retinal images using a noninvertible, 12-parameter, quadratic transformation model hierarchical, robust estimation. The major innovation linear, feature-based, noniterative method jointly estimating consistent transformations all onto mosaic "anchor image." Constraints this estimation are derived pairwise registration...
In this paper, we investigate the effect of substantial inter-image intensity changes and in modality on performance keypoint detection, description, matching algorithms context image registration. doing so, modify widely-used descriptors such as SIFT shape contexts, attempting to capture insight that some structural information is indeed preserved between images despite dramatic appearance changes. These extensions include (a) pairing opposite-direction gradients formation orientation...
Abstract Access to large image volumes through camera traps and crowdsourcing provides novel possibilities for animal monitoring conservation. It calls automatic methods analysis, in particular, when re-identifying individual animals from the images. Most existing re-identification rely on either hand-crafted local features or end-to-end learning of fur pattern similarity. The former does not need labeled training data, while latter, although very data-hungry typically outperforms enough...
We present a simple usage of pre-trained Vision Transformers (ViTs) for fine-grained analysis, aiming to identify and localize the traits that distinguish visually similar categories, such as different bird species or dog breeds. Pre-trained ViTs DINO have shown remarkable capabilities extract localized, informative features. However, using saliency maps like Grad-CAM can hardly point out traits: they often locate whole object by blurred, coarse heatmap, not traits. propose novel approach...
Despite many successful applications of robust statistics, they have yet to be completely adapted computer vision problems. Range reconstruction, particularly in unstructured environments, requires a estimator that not only tolerates large outlier percentage but also several discontinuities, extracting multiple surfaces an image region. Observing random outliers and/or points from across discontinuities increase hypothesized fit's scale estimate (standard deviation the noise), our new...
Abstract Image‐based re‐identification of animal individuals allows gathering information such as population size and migration patterns the animals over time. This, together with large image volumes collected using camera traps crowdsourcing, opens novel possibilities to study populations. For many species, can be done by analysing permanent fur, feather, or skin that are unique each individual. In this paper, authors pattern feature aggregation based consider two ways improving accuracy:...
Abstract Drones have become invaluable tools for studying animal behaviour in the wild, enabling researchers to collect aerial video data of group‐living animals. However, manually piloting drones track groups consistently is challenging due complex factors such as terrain, vegetation, group spread and movement patterns. The variability manual can result unusable downstream behavioural analysis, making it difficult standardized datasets collective behaviour. To address these challenges, we...