studying objects with mirror and highly specular reflectance


                  
Our understanding of Lambertian (diffuse reflectance) objects is rich. Lambertian objects have very specific properties that make a lot of computer vision problems possible. This is, in part, due to the observability of geometric features from images of such objects. However, the same cannot be said for mirrors. Charting the range of reflectances, at one end, while we have immense knowledge of Lambertian objects including computational tools to handle them, we have limited tools and understanding of how to handle specular (part diffuse and part mirror) and even less so for objects that exhibit pure mirror reflectance. This line of research seeks to develop theory and practical methods and tools for handling mirrors, with the eventual goal of being able to handle objects with arbitrary reflectance properties.

people

papers

Image Invariants for Smooth Reflective Surfaces
Aswin C. Sankaranarayanan, Ashok Veeraraghavan, Oncel Tuzel, and Amit Agrawal
ECCV 2010
Specular Surface Reconstruction from Sparse Reflection Correspondences
Aswin C. Sankaranarayanan, Ashok Veeraraghavan, Oncel Tuzel, and Amit Agrawal
CVPR 2010

Talk Slides

Aswin's UMD Talk on "Vision on Mirrors" [PDF]

code and dataset

Code for the ECCV 2010 paper on estimating the parabolic curves of a mirror: [synthetic dataset and code] [real dataset and code]

Code for the CVPR 2010 paper on reconstructing the shape of a mirror from multiple images: [synthetic dataset and code][real dataset and code]

real data and code coming soon ...