Computer Graphics (CG)

Geometry Analysis and Computer Vision (Python and C++)

Geometry Processing and Analysis for Archaeology

Research Problem: Quickly identifying archaeological features in the form of curvelinear structures

Research Task: (1) Develop an efficient and accurate pipeline archaeology researchers, who are unfamiliar with coding, can easily use (2) Evaluate the accuracy of this tool using Ground-Penetrating-Radar (GPR) data, which exposes similar curvelinear structures underground

Actions: (3000+ lines of code)

  1. Created mesh from very-high-resolution LiDAR point cloud data from sites in Uzbekistan (Screened Poisson Reconstruction)
  2. Extracted curvelinear features from this mesh using the CrestCODE algorithm
  3. Extracted curvelinear features from GPR image using the Ridge-Detection algorithm
  4. Deformed the mesh as well as the curvelinear features to fit the distortion of the GPR image (due to reprojecting to a 2D plane) using the Smooth-Excess-Area algorithm
  5. Developed a metric to compare the curvelinear features from the deformed mesh and the GPR image
  6. Calculated summary statistics to identify the accuracy of our method
  7. Presented our results in the 2022 Undergraduate Research Symposium
  8. Our project was featured in the WashU's Technical Exchange Planning Meeting with the National Geospatial-Intelligence Agency (NGA)
  9. Discussed with archaeologists to confirm and update expectations for research outcome
  10. Produced well-documented, modularized code, which I circulated with new PhD students joining this project
  11. Reported findings in weekly project meetings
  12. Created an easy-to-use interface for all this automated pipeline

Results:

  1. Developed an efficient, user-friendly pipeline to automatically idenfity archaeological features in large scale
  2. Spearheaded and laid foundation for the collaboration between the two labs involved in this project
  3. Drafting: Reported our findings in a publication

Extracting Insect Wings Venations and their Closed Regions from Images

Research Problem: (1) Automatically identify the venation of insect wings and mutually-exclusive regions they enclose (2) record the extracted curvelinear features in the form of clean SVG images

Research Task: (1) Develop a pipeline to extract curvelinear features from images of insect wings (2) Develop a method to identify the small enclosed regions in the wings (3) Convert the extracted features into SVG format

Actions: (1500+ lines of code)

  1. Studied Seth Donohogue and Christopher Rycroft's work on this topic
  2. Reached out to them to better understand their methods (segmentation and optimization) and strategies
  3. Researched similar methods that extracts curvelinear features from 2D images
  4. Implemented a combination of thresholding, denoising, and skeletonization to quickly extract curvelinear structures
  5. Stored the extracted features in the forms of nodes and edges in an undirected graph
  6. Applied the minimum-cycle-basis algorithm to identify mutually-exclusive cycles in this graph, which corresponds to small enclosed areas on the wings of insects like dragonfly or grasshopper
  7. Implementated a workflow to convert the extracted features to SVG at high and low resolutions
  8. Developed a program to create 3D models from the wings that area ready for 3D-printing

Results:

  1. Developed an automatic and efficient pipeline to identify curvelinear features in insect wings
  2. Developed a follow-up workflow to extract the small enclosed areas from a network of those features
  3. Created an easy, user-friendly interface for artists and researchers who aren't familiar with coding