85% 78%pingtan/papers/cvpr14_material_poster.pdf · jinwei and c. liu, “discriminative...

A New Perspective on Material Classification and Ink Identification Rakesh Shiradkar * Li Shen † George Landon ‡ Sim Heng Ong * and Ping Tan * * National University of Singapore † Institute for Infocomm Research ‡ Eastern Kentucky University Motivation Most previous methods on material classification use a head- on camera. Jinwei and Chao [1] Wang et. al [2] Conventional camera captures only a 1D slice of the 2D BRDF Since, Isotropic BRDFs can be approximated by a 2D BRDF (θ d , θ h ) Rusinkiewicz [3] Our New Perspective A slanted camera captures a larger portion of the 2D BRDF space. 0 o 45 o 90 o θ h θ d [ 0 o 45 o 90 o Experiments on Ink Database 85% 78% Slanted Camera vs. Conventional Camera Application : Ink Identification v = n l h θ h θ d 0 o 45 o 90 o θ h θ d [ 0 o 45 o 90 o θ d , θ h space θ d = θ h v n l h θ h θ d θ d ≠ θ h Reconstructed 3D points 3D surface fit 1-D BRDF slice Multi view images Ink classification Classifier Multi view images Image Registration θ d , θ h space We observe improved classification accuracy ! Results To simplify the setup, we bring the camera and light together Although accuracy is a bit compromised, we capture important discriminative information (retro reflectance, specular highlights ) Conclusions References Sample Image Our Result Ground Truth 1. G. Jinwei and C. Liu, “Discriminative illumination: Per-pixel classification of raw materials based on optimal projections of spectral BRDF,” in Proc. CVPR, 2012. 2.O. Wang, P. Gunawardane, S. Scher, and J. Davis, “Material classification using BRDF slices,” in Proc. CVPR, pp. 2805 –2811, 2009. 3. S. Rusinkiewicz, “A New Change of Variables for Efficient BRDF Representation,” in Eurographics Rendering Workshop, pp. 11 – 22, 1998. 1. A slanted camera increases the sampling region of the 2D BRDF space. 2. This enhances the performance of BRDF-based material classification. 3. The first work to analyse BRDF for ink identification, an important problem in forensics. 4. A simple handheld camera-flashlight device for data capture. This material is based upon work supported by the National Science Foundation under Grant No. IIS-1008285. Ping Tan is partially supported by the ASTAR PSF project R-263-000-698-305.

Upload: others

Post on 02-Oct-2020

2 views

Category:

Documents

0 download

Report

Download

Embed Size (px):

TRANSCRIPT

Page 1: 85% 78%pingtan/Papers/cvpr14_material_poster.pdf · Jinwei and C. Liu, “Discriminative illumination: Per-pixel classification of raw materials based on optimal projections of spectral

A New Perspective on Material Classification and Ink Identification Rakesh Shiradkar* Li Shen† George Landon‡ Sim Heng Ong* and Ping Tan*

*National University of Singapore †Institute for Infocomm Research ‡Eastern Kentucky University

Motivation

Most previous methods on material classification use a head-on camera.

Jinwei and Chao [1]

Wang et. al [2]

Conventional camera captures only a 1D slice of the 2D BRDF

Since, Isotropic BRDFs can be approximated by a 2D BRDF (θd, θh)

Rusinkiewicz [3]

Our New Perspective

A slanted camera captures a larger portion of the 2D BRDF space.

0o 45o 90o θh

θd

[

Experiments on Ink Database

85% 78%

Slanted Camera vs. Conventional Camera

Application : Ink Identification v = n

θh

θd

0o 45o 90o θh

θd

[

θd , θh space

θd = θh

v n

θh θd

θd ≠ θh

Reconstructed 3D points 3D surface fit

1-D BRDF slice

Multi view images

Ink classification

Classifier

Multi view images

Imag

gist

rati

θd , θh space

We observe improved classification accuracy !

Results

To simplify the setup, we bring the camera and light together

Although accuracy is a bit compromised, we capture important discriminative information (retro reflectance, specular highlights )

Conclusions

References

Sample Image Our Result Ground Truth

1. G. Jinwei and C. Liu, “Discriminative illumination: Per-pixel classification of raw

materials based on optimal projections of spectral BRDF,” in Proc. CVPR, 2012.

2. O. Wang, P. Gunawardane, S. Scher, and J. Davis, “Material classification using

BRDF slices,” in Proc. CVPR, pp. 2805 –2811, 2009.

3. S. Rusinkiewicz, “A New Change of Variables for Efficient BRDF Representation,” in

Eurographics Rendering Workshop, pp. 11 – 22, 1998.

1. A slanted camera increases the sampling region of the 2D BRDF space. 2. This enhances the performance of BRDF-based material classification. 3. The first work to analyse BRDF for ink identification, an important problem in forensics. 4. A simple handheld camera-flashlight device for data capture.

This material is based upon work supported by the National Science Foundation under Grant No. IIS-1008285. Ping Tan is partially supported by the ASTAR PSF project R-263-000-698-305.

Maxent Models and Discriminative Estimation

Discriminative and Consistent ... - cv-foundation.org

Localized discriminative Gaussian process latent variable ... · PDF fileLocalized discriminative Gaussian process latent variable model for text ... we use Discriminative Gaussian

An Introduction of Support Vector Machine Jinwei Gu 2008/10/16

Exploiting Lexical Information and Discriminative ...glicom.upf.edu/lambert/papers/08_04_thesis_lambert.defense.pdf · Exploiting Lexical Information and Discriminative Alignment

Discriminative Multiple Target Tracking

Discriminative SNMF EA201603

Discriminative, Unsupervised, Convex Learning

Discriminative Illumination - Jinwei Gu · Discriminative Illumination: Per-pixel Classification of Raw Materials based on Optimal Projections of Spectral BRDFs Jinwei Gu and Chao

Discriminative Adaptive Training and Discriminative Adaptationmi.eng.cam.ac.uk/seminars/speech/lan_seminar1.pdf · Discriminative Adaptive Training and Discriminative Adaptation Lan

Flora Universe by Cai Jinwei

Discriminative Image Segmentation: Applications to