Jacobus J Barnard

Jacobus J Barnard

Professor, Computer Science
Associate Director, Faculty Affairs-SISTA
Professor, Electrical and Computer Engineering
Professor, Cognitive Science - GIDP
Professor, Genetics - GIDP
Professor, Statistics-GIDP
Professor, BIO5 Institute
Member of the General Faculty
Member of the Graduate Faculty
Primary Department
Department Affiliations
Contact
(520) 621-4632

Research Interest

Kobus Barnard, PhD, is an associate professor in the recently formed University of Arizona School of Information: Science, Technology, and Arts (SISTA), created to foster computational approaches across disciplines in both research and education. He also has University of Arizona appointments with Computer Science, ECE, Statistics, Cognitive Sciences, and BIO5. He leads the Interdisciplinary Visual Intelligence Lab (IVILAB) currently housed in SISTA. Research in the IVILAB revolves around building top-down statistical models that link theory and semantics to data. Such models support going from data to knowledge using Bayesian inference. Much of this work is in the context of inferring semantics and geometric form from image and video. For example, in collaboration with multiple researchers, the IVILAB has applied this approach to problems in computer vision (e.g., tracking people in 3D from video, understanding 3D scenes from images, and learning models of object structure) and biological image understanding (e.g., tracking pollen tubes growing in vitro, inferring the morphology of neurons grown in culture, extracting 3D structure of filamentous fungi from the genus Alternaria from brightfield microscopy image stacks, and extracting 3D structure of Arabidopsis plants). An additional IVILAB research project, Semantically Linked Instructional Content (SLIC) is on improving access to educational video through searching and browsing.Dr. Barnard holds an NSF CAREER grant, and has received support from three additional NSF grants, the DARPA Mind’s eye program, ONR, the Arizona Biomedical Research Commission (ABRC), and a BIO5 seed grant. He was supported by NSERC (Canada) during graduate and post-graduate studies (NSERC A, B and PDF). His work on computational color constancy was awarded the Governor General’s gold medal for the best dissertation across disciplines at SFU. He has published over 80 papers, including one awarded best paper on cognitive computer vision in 2002.

Publications

Morris, S., Gimblett, R., & Barnard, K. (2005). Probabilistic travel modeling using GPS. MODSIM05 - International Congress on Modelling and Simulation: Advances and Applications for Management and Decision Making, Proceedings, 149-155.

Abstract:

Recreation simulation modeling, when combined with intelligent monitoring, is becoming a valuable tool for natural resource managers. The goal of recreation simulation is to accurately model recreational use, both current and future. Models are applied to gain a thorough understanding of the characteristics of recreation. Indicator variables such as visitor experience, carrying capacity and impact on resources can be computed. If the model is valid it can be used to predict future use as well as to investigate the effect of new scenarios and management decisions. Recent research has focused on agent-based modeling techniques. Recreators are represented by autonomous, intelligent agents that travel across the landscape. A central issue is the model used for agent travel decisions. Current techniques range from replicating trips exactly to making local, intersection level decisions based on probability. But little attention has been paid to justifying these models. In this work we examine a range of probabilistic models. The models differ in the length of the Markov chain used to compute agent decisions. The length of the chain ranges from zero (local decisions only) to infinity (exact trip replication). We test the length of the chain on held out data for validation. We show that the choice of model strongly influences the validity and results of the simulation. To test these models we present a framework for automatically constructing agent-based models from an input set of GPS tracklogs. The GPS tracklogs are collected by volunteers as they recreate in natural areas. Traditionally, data on where recreators travel is collected in the form of trip diaries, filled out on paper by visitors or by interview. Other demographic and attitudinal data is also collected along with the actual route traveled. Although the additional information is valuable, the data must be collected and entered by hand. Paper diaries also place a significant time burden on visitors, reducing the compliance rate as well as skewing the results (ensuring only visitors with excess time participate). Using GPS devices to record visitor trips helps alleviate these problems. The framework for processing GPS trips and automatically building a model presented in this work significantly reduces the time required to build a model, lowers the cost and widens the applicability of recreation simulation modeling to new areas. GPS devices automatically record their data, requiring only that visitors turn the unit on and carry it with a marginal view of the sky. GPS use is also becoming more widespread among recreators. As more recreators use GPS to record their trips, data useful to modeling is becoming increasingly available. The steps in GPS driven model generation are as follows. First, the set of GPS tracklogs is combined to form the underlying travel network along which agents will travel. Each GPS tracklog is then traced along the network in order to determine what choices were made as the recreator traveled across the network. This produces a list of trip itineraries. Model parameters (probability tables) can then be computed from the trips. The length of the Markov chain used in the probability tables is a parameter to the model. The optimal value is found by testing the likelihood of heldout data for different chain lengths. This step is done automatically. Once the optimal length of the chain is chosen the model is complete and agent-based simulation can proceed. The entire framework for automatically producing GPS driven agent-based models is implemented in our TopoFusion GPS mapping software. We present results from two collections of GPS tracklogs from different trail systems. The first is from Tucson Mountain Park and is the result of a volunteer collection effort by the authors. A trails master plan is underway at the park, with input from our model. The second is a collection of tracks from mountain bike rides in the Finger Rock Wash area, collected by the author. Testing by held-out data on both GPS datasets indicates that current modeling methods are insufficient to model recreator travel decisions. The middle ground (neither exact replication nor local decisions) consistently performs better.

Peralta, R. T., Rebguns, A., Fasel, I. R., & Barnard, K. (2013). Learning a policy for gesture-based active multi-touch authentication. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 8030 LNCS, 59-68.

Abstract:

Multi-touch tablets can offer a large, collaborative space where several users can work on a task at the same time. However, the lack of privacy in these situations makes standard password-based authentication easily compromised. This work presents a new gesture-based authentication system based on users' unique signature of touch motion when drawing a combination of one-stroke gestures following two different policies, one fixed for all users and the other selected by a model of control to maximize the expected long-term information gain. The system is able to achieve high user recognition accuracy with relatively few gestures, demonstrating that human touch patterns have a distinctive "signature" that can be used as a powerful biometric measure for user recognition and personalization. © 2013 Springer-Verlag Berlin Heidelberg.

Schlecht, J., Barnard, K., & Pryor, B. (2007). Statistical inference of biological structure and point spread functions in 3D microscopy. Proceedings - Third International Symposium on 3D Data Processing, Visualization, and Transmission, 3DPVT 2006, 373-380.

Abstract:

We present a novel method for detecting and quantifying 3D structure in stacks of microscopic images captured at incremental focal lengths. We express the image data as stochastically generated by an underlying model for biological specimen and the effects of the imaging system. The method simultaneously fits a model for proposed structure and the imaging system's parameters, which include a model of the point spread function. We demonstrate our approach by detecting spores in image stacks of Alternaria, a microscopic genus of fungus. The spores are modeled as opaque ellipsoids and fit to the data using statistical inference. Since the number of spores in the data is not known, model selection is incorporated into the fitting process. Thus, we develop a reversible jump Markov chain Monte Carlo sampler to explore the parameter space. Our results show that simultaneous statistical inference of specimen and imaging models is useful for quantifying biological structures in 3D microscopic images. In addition, we show that inferring a model of the imaging system improves the overall fit of the specimen model to the data. © 2006 IEEE.

Predoehl, A., Morris, S., & Barnard, K. (2013). A statistical model for recreational trails in aerial images. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 337-344.

Abstract:

We present a statistical model of aerial images of recreational trails, and a method to infer trail routes in such images. We learn a set of textons describing the images, and use them to divide the image into super-pixels represented by their text on. We then learn, for each text on, the frequency of generating on-trail and off-trail pixels, and the direction of trail through on-trail pixels. From these, we derive an image likelihood function. We combine that with a prior model of trail length and smoothness, yielding a posterior distribution for trails, given an image. We search for good values of this posterior using a novel stochastic variation of Dijkstra's algorithm. Our experiments, on trail images and ground truth collected in the western continental USA, show substantial improvement over those of the previous best trail-finding method. © 2013 IEEE.

Gabbur, P., Hua, H., & Barnard, K. (2010). A fast connected components labeling algorithm and its application to real-time pupil detection. Machine Vision and Applications, 21(5), 779-787.

Abstract:

We describe a fast connected components labeling algorithm using a region coloring approach. It computes region attributes such as size, moments, and bounding boxes in a single pass through the image. Working in the context of real-time pupil detection for an eye tracking system, we compare the time performance of our algorithm with a contour tracing-based labeling approach and a region coloring method developed for a hardware eye detection system. We find that region attribute extraction performance exceeds that of these comparison methods. Further, labeling each pixel, which requires a second pass through the image, has comparable performance. © Springer-Verlag 2009.