Connections
Play some custom games of NYT Connections.
128kb
"Sliding Grids". 128x128 GIF, 128 frames. Created with p5.js.
Submission for the 128kb exhibition by Tim Rodenbröker.
Try it out below!
Dissertation Project
Robust Photometric Stereo for Specular and Shadowed Objects Under Noise
The report for the project can be downloaded here.
The accompanying code is available here.
Photometric Stereo is a technique concerned with extracting the geometry (normals) of a surface given a series of photographs captured under different lighting directions (Woodham, 1980). Once a normal field is reconstructed, it can be integrated to give a surface.
Input [1/20]
Output
Input images on the left are taken from Xiong et al., 2012 (click image to cycle through). Corrupted versions were generated as part of the project. The reconstructed surface is shown on the right (drag to rotate).
The basic photometric stereo method assumes that surfaces are perfectly Lambertian, and as such suffers in the presence of image noise and specular (shiny) surfaces. The project aims to derive, implement and evaluate a robust algorithm on these trickier cases.
The codebase contains three main features:
- A pipeline for "corrupting" images with image noise and specular highlights.
- A robust Sparse Bayesian Learning solver for Photometric Stereo (Ikehata et al., 2012).
- A normal smoothing algorithm derived from Jones, Durand and Zweiker (2004).
