My research focuses on computer vision and graphics. I am particularly interested in image-based reconstruction (such as photometric stereo), BRDF estimation, and inverse lighting.

Photometric Stereo With Non-Parametric and Spatially-Varying Reflectance
Toward Reconstructing Surfaces With Arbitrary Isotropic Reflectance : A Stratified Photometric Stereo Approach
Resolving the Generalized Bas-Relief Ambiguity by Entropy Minimization
Planar Light Probe
Super-Resolution
Reflectance Estimation Under Natural Illumination
Lighting Recovery with Dense Stereo
Pedestrian Detection
Activecampus.ucsd.edu
The Universal Planar Manipulator
Course Projects

Photometric Stereo With Non-Parametric and Spatially-Varying Reflectance

We present a method for simultaneously recovering shape and spatially varying reflectance of a surface from photometric stereo images. The distinguishing feature of our approach is its generality; it does not rely on a specific parametric reflectance model and is therefore purely ``data-driven''. This is achieved by employing novel bi-variate approximations of isotropic reflectance functions. By combining this new approximation with recent developments in photometric stereo, we are able to simultaneously estimate an independent surface normal at each point, a global set of non-parametric ``basis material'' BRDFs, and per-point material weights. Our experimental results validate the approach and demonstrate the utility of bi-variate reflectance functions for general non-parametric appearance capture.

(Left) Example input images (GOURD: 1 of 102 images, APPLE: 1 of 112 images). All input images acquired from a single viewpoint. (Right) Synthetic images rendered rendered from novel viewpoints / illumination using shape and reflectance acquired by our algorithm.

N. Alldrin, T. Zickler, and D. Kriegman, "Photometric Stereo With Non-Parametric and Spatially-Varying Reflectance", 2008 Conf. on Comp. Vision and Pattern Recognition (CVPR), Anchorage, AK, June 2008. [pdf]

Toward Reconstructing Surfaces With Arbitrary Isotropic Reflectance : A Stratified Photometric Stereo Approach

We consider the problem of reconstructing the shape of a surface with an arbitrary, spatially varying isotropic bidirectional reflectance distribution function (BRDF), and introduce a novel, stratified photometric stereo method. By using a particular configuration of lights, it is possible to use symmetry in the image measurements resulting from BRDF isotropy to estimate at each point a plane containing the surface normal. For differentiable surfaces, this allows us to recover the isocontours of the depth map, but not the actual depth associated with each contour. The isocontour structure provides topological information about the surface (critical points, Reeb graph, etc.). By using additional cues in the image data or imposing additional constraints on the surface (e.g., shadows, specular highlights, Helmholtz Reciprocity, uniform BRDF), the unknown height of each isocontour can be estimated and the metric structure is resolved. We validate this technique on real and synthetic data by successfully recovering the isocontours of the depth map from images.

(Left) 6 of 32 total input images of a helmet. (Center) Recovered surface gradient directions. (Right) Recovered iso-depth contours.

N. Alldrin and D. Kriegman, "Toward Reconstructing Surfaces With Arbitrary Isotropic Reflectance : A Stratified Photometric Stereo Approach", 2007 International Conf. on Computer Vision (ICCV), Rio de Janeiro, Brazil, October 2007. [pdf]

Resolving the Generalized Bas-Relief Ambiguity by Entropy Minimization

It is well known in the photometric stereo literature that uncalibrated photometric stereo, where light source strength and direction are unknown, can recover the surface geometry of a Lambertian object up to a 3-parameter linear transform known as the generalized bas-relief (GBR) ambiguity. Many techniques have been proposed for resolving the GBR ambiguity, typically by exploiting prior knowledge of the light sources, the object geometry, or non-Lambertian effects such as specularities. A less celebrated consequence of the GBR transformation is that the albedo at each surface point is transformed along with the geometry. Thus, it should be possible to resolve the GBR ambiguity by exploiting priors on the albedo distribution. To the best of our knowledge, the only time the albedo distribution has been used to resolve the GBR is in the case of uniform albedo. We propose a new prior on the albedo distribution : that the entropy of the distribution should be low. This prior is justified by the fact that many objects in the real-world are composed of a small finite set of albedo values.

(Left) Illustration of the effect of the GBR transformation on the albedo distribution. (Right) Results obtained from the proposed method (see paper for more details).

N. Alldrin, S. Mallick, and D. Kriegman, "Resolving the Generalized Bas-Relief Ambiguity by Entropy Minimization", 2007 Conf. on Comp. Vision and Pattern Recognition (CVPR), Minneapolis, MN, June 2007. [pdf]

Planar Light Probe

The planar light probe is a novel technique for measuring lighting that exploits the interaction of light with a set of custom BRDFs. Because of the way our probe is constructed, we are able to recover lighting in terms of a set of globally supported basis functions, which has certain advantages over existing methods for measuring lighting. Our probe consists of a sheet of glass with paper on the underside. Printed patterns are placed on both sides of the glass to create custom BRDFs that enable recovery of the lighting over the upper-hemisphere of the probe.

(Left) The light probe viewed with a point light source at various positions. Note the different responses from each BRDF in the probe. (Right) A closeup of the probe's surface.

N. Alldrin and D. Kriegman, "A Planar Light Probe", 2006 Conf. on Comp. Vision and Pattern Recognition (CVPR), New York, NY, June 2006. [pdf]

Super-Resolution

I wrote a report on super-resolution for my research exam (a CSE department requirement). Since it might be useful to others, I have made the report available online.


NASA frequently uses super-resolution techniques to recover information from multiple degraded images.

Super-Resolution - Research Exam (UCSD Computer Science & Engineering Dept.) [ps][pdf]

Reflectance Estimation Under Natural Illumination

Knowledge of material reflectance properties is of central concern in computer graphics. Traditionally, BRDFs have been estimated with low dimensional parametric models while more recently data-driven techniques have become more common. While some progress has been made to ease the pain of direct BRDF measurements, the process is still far from ideal, typically requiring carefully calibrated aparatus and controlled lighting. The goal of this project is to ease this burden by enabling BRDF measurements under natural illumination using spherical harmonics.

Reflectance Estimation Under Natural Illumination [ps][pdf]

Lighting Recovery with Dense Stereo

By representing the lighting with a low-order spherical harmonics expansion and using aggressive outlier detection we attempt to recover low-frequency lighting using geometry provided by dense stereo reconstruction. The algorithm works as follows : Given a stereo image pair of a scene we recover geometry using standard stereo algorithms. Linear constraints on the spherical harmonic lighting coefficients are then formed from pixel intensities, recovered surface normals, and an assumed BRDF. Since many of the constraints are noisy or incorrect, RANSAC is used to remove outliers while solving the system.

Pedestrian Detection

I implemented a version of the Adaboost pedestrian detector popularized by Viola and Jones. Code, training data, and videos are available on the web page below.

Detecting Pedestrians - CSE252c (Computer Vision) [ps][pdf][web site]

Activecampus.ucsd.edu

One summer I worked with Bill Griswold on the ActiveCampus.ucsd.edu project, exploring ideas in ubiquitous computing. From the website:

"The ActiveCampus project aims to provide location-based services for educational networks and understand how such systems are used. activeclass enables collaboration between students and professors by serving as a visual moderator for classroom interaction. ActiveCampus Explorer uses a person's context, like location, to help engage them in campus life."

The Universal Planar Manipulator

While an undergraduate at UC Berkeley, I worked on the universal planar manipulator (UPM). The UPM is able to independently move objects on it's surface by forming localized "jets" that only affect a small part of the surface area. Here is a page I created documenting my work and here is the main page for the project.

D. Reznik, J. Canny, N. Alldrin, "Leaving on a Plane Jet", 2001 Int. Conf. on Intell. Robots & Systems (IROS), Maui, Hawaii, October 2001. [pdf]

Course Projects

Failure Prediction in Hardware Systems

Failure Prediction in Hardware Systems - CSE221 (Operating Systems) [ps][pdf]

TCP Nicer

TCP Nicer : Support for Hierarchical Background Transfers - CSE222 (Networks) [ps][pdf]

Neural Networks

Clustering With EM and K-Means - CSE253 (Neural Networks) [ps][pdf]

A Three-Unit Network is All You Need to Discover Females - CSE253 (Neural Networks) [ps][pdf]

Classifying Facial Expression with Radial Basis Function Networks, using Gradient Descent and K-means - CSE253 (Neural Networks) [ps][pdf]

Consistent Hashing

Consistent Hashing - CSE202 (Algorithms) [ps][pdf]