Image Processing
Problem Statement
To segment the Human Embryo kidney cells from Fluoroscent microscpic images.
​
Approach
​
The right panel contains the original microscopic images and the left panel shows the segmentation algorithm correctly marking all the cells in red. Stpes involved: Contrast enhancement and boundary contour segmentations. Used Matlab for this project.
​
More details can be found in this Github repository.
​
Problem Statement
Peer Engagement as a Common Resource: Managing Interaction Patterns in Institutions
Experimenting with background subtraction methods on video footage of Drexel University library to analyze the influx and outflux of students and to study their interactions over time.
​
This work has been published in the Society for College and University Planning.
Winner of Perry Chapman Prize (2015) - “Peer Engagement As A Common Resource” Johnson.M, Nitecki D., John Koo. M, Nathani R, Swaminathan S. Society of College and University Planning (SCUP), 2015. ISBN 978-1-937724-52-8.
​
Approach
​
While face-to-face collaboration has been theorized to be a key element in intellectual development and cognition, no formal method of quantitative measurement has been applied to understand collective face-to-face learning in academic institutions or how patterns of interaction and individual reflection may reveal information exchange among students within educational institutions. To address this gap, this study introduces a novel tool and framework to promote the systematic study of peer collaboration for general use in education.
Results of this applied research will be useful to architects, interior designers, librarians, educators, and researchers interested in obtaining empirical evidence and applying it to the design of learning environments and the assessment of how well spaces intentionally relate to learning. This research project introduces a common means for researchers in space design, education, and information science to develop principles and best practices to improve return on investment in the design of informal learning environments.
​
​
Publication
Problem Statement
To identify typical mitotic shape patterns in phase contrast microscopic images of stem cells and to differentiate between mitotic and non-mitotic cell images.
​
Approach
​
I developed an unsupervised machine learning algorithm for classifying stem cells into 2 classes: Non-Mitotic(single cells) and Mitotic(cells that are in the process of division). Dataset contains 24 images: 12 of Non-mitotic and 12 of Mitotic. Steps: Compress each image using 2D wavelets. Calculate the normalized compression distance between each image pair. Convert this affinity matrix into spectral representation using symmetric divisive normalization. Find the 2 largest eigenvectors of the resulting matrix and use K-means clustering to classify into 2 groups. Used Matlab.
​
More details can be found in this Github repository.
​