I have been involved in a number of projects during my graduate studies. The bulk of my work has been at the intersection of computer vision and medical imaging, but I have also applied the former to document image analysis. I am always happy to answer any questions about any of the projects listed below.
Tree-like branching structures are ubiquitous in nature. We are working on a novel framework and methodology for determining the topology and geometry of three-dimensional branching structures, such as retinal vessels, plant roots, and lightning, from two-dimensional
images. The projection of a tree onto an image surface introduces branch occlusions
and crossings. This superposition of branches introduces false branch-points, which make a tree's topology ambiguous. We have developed a simple, but powerful Bayesian tree growth model that allows us to determine the likelihood of a possible 3D source tree given a 2D projection, as well as a set of graph-theoretic operations that allow us to enumerate all possible source trees. We are currently developing algorithms to efficiently explore this space, both deterministically and stochastically.
We developed a novel methodology for extracting the vascular network in the human retina using Dijkstra's shortest-path algorithm. Our method preserves vessel thickness, requires no manual intervention, and follows vessel branching naturally and efficiently. To test our method, we constructed a retinal video indirect ophthalmoscopy (VIO) image database from pediatric patients and compared the segmentations achieved by our method and state-of-the-art approaches to a human-drawn gold standard. Our experimental results show that our algorithm outperforms prior state-of-the-art methods, for both single VIO frames and automatically generated, large field-of-view enhanced mosaics. The journal paper for this work is available here. We have made the corresponding dataset and source code freely available online here.
Indirect ophthalmoscopy (IO) is the standard of care for evaluation of the neonatal retina. When recorded on video from a head-mounted camera, IO images have low quality and narrow Field of View (FOV). We developed an image fusion methodology for converting a video IO recording into a single, high quality, wide-FOV mosaic that seamlessly blends the best frames in the video. To this end, we developed fast and robust algorithms for automatic evaluation of video quality, artifact detection and removal, vessel mapping, registration, and multi-frame image fusion. Our experiments showed the effectiveness of our mosaicing methodology. The journal paper for this work is available here.
Many types of degradation can render ancient manuscripts very hard to read. In bleed-through, the text from the reverse, or verso, side of a page seeps through into the front, or recto. We developed a hysteresis thresholding approach to greatly reduce bleed-through. Thresholding alone cannot properly separate ink and bleed-through because the ranges of intensities for the two classes overlap. Hysteresis thresholding overcomes this limitation via the two steps of thresholding and ink regrowth. In order to provide quantitative measures of the effectiveness of this approach, we constructed a novel dataset which features bleed-through and has available ground truth. We evaluated our method and a number of previously proposed approaches on ink pixel precision and recall. Hysteresis thresholding significantly improves over existing methods. The conference paper for this work is available here.
Glaucoma is the
second leading cause of blindness in the world affecting over 70 million people worldwide, including
around 3 million people in the US alone. In my Research Initiation Project, we developed a method for determining the most significant layers of the retina given spectral domain optical coherence tomography volumetric data using active contours or snakes. Furthermore, we then developed an algorithm for determining the location and diameter of the optic nerve head (ONH) by fitting a three-dimensional, cylindrical snake. The 3D cylinder was more robust at finding the ONH than single slice, 2D methods.
|
