||The Vision, Dynamics and Learning Lab is a research lab in the Department of Biomedical Engineering at Johns Hopkins University. We are a member
of the Center for Imaging Science (CIS) and of the Whitaker Institute of Biomedical Engineering.
Our research spans a wide range of areas in biomedical imaging, computer vision, dynamics and controls, machine learning and robotics.
In particular, we are interested in inference problems involving geometry, dynamics, photometry and statistics, such as (1) inferring
models from images (image/video segmentation and structure from motion), static data (generalized PCA) or dynamic data (identification of hybrid systems), and (2) using such models to accomplish a complex
mission (land a helicopter, pursue a team of evaders, follow a formation). Please feel free to contact any member of this lab if you have any questions or comments!
Subspace clustering is an open research area that deals with clustering high-dimensional data into lower dimensional subspaces. These types of problems are highly applicable in the age of Big Data and specifically useful in the realm of computer vision. In particular, our research is in developing advanced algorithms that utilize sparse representations, generalized PCA, and manifold learning applied to problems such as motion segmentation.
As humans we have developed a rich visual understanding of our world. We can instantly recognize what is in an image, where it is located and what activities are being done and other conceptual relationships. Teaching a computer to automatically do these tasks is a difficult problem. We develop advanced algorithms that combine category-level (top-down) and pixel-level (bottom-up) information to simultaneously categorize and segment objects in a 2D images and activities in time series data using graph theory and sparse representation theory.more >>
The analysis of biomedical images is an important field for automating tasks such as disease classification, lesion segmentation, and registration. In our lab we use methodologies rooted in computer vision, machine learning and optimization to analyze diffusion magnetic resonance images (dMRI) of the brain as well as cardiac stem cell signals. For dMRI we are developing advanced algorithms for fiber orientation estimation, feature extraction, registration, disease classification and sparse representation. For cardiac stems cells we use advanced registration to classify mature stems cells.
The task of pattern recognition in high-dimensional time-series data can be modeled using linear dynamical systems. But in order to develop statistical analyses on these time-series data, we must study the geometry of the spaces of linear dynamical systems to define distances between systems.
Our research also includes the identification and observability of hybrid systems. more >>
You can also check out the Vision Lab app iMixPics
, developed by members of our lab, which allows users to overlay and combine multiple photos using interactive image segmentation techniques. This app is now available free at the iTunes App Store for iPhone, iPad, and iPod touch, under Johns Hopkins Mobile medicine.