Columbia Object Image Library (COIL-20)

Page 1

Sameer A. Nene and Shree K. Nayar and Hiroshi Murase

Department of Computer Science

Columbia University

New York, N.Y. 10027

[email protected]

Technical Report No. CUCS-006-96

Page 2

Abstract

Columbia Object Image Library (COIL-20) is a database of gray-scale images of 20

objects. The objects were placed on a motorized turntable against a black background.

The turntable was rotated through 360 degrees to vary object pose with respect to a xed

camera. Images of the objects were taken at pose intervals of 5 degrees. This corresponds

to 72 images per object. The database has two sets of images. The rst set contains 720

unprocessed images of 10 objects. The second contains 1,440 size normalized images of 20

objects. COIL-20 is available online via ftp.

Page 3

1 Introduction

We have constructed a database of 1,440 grayscale images of 20 objects (72 images per

object). The objects have a wide variety of complex geometric and re ectance characteristics

(see gure 1(a)). The database, called Columbia Object Image Library (COIL-20), was used

in a real-time 20 object recognition system Murase and Nayar-1995]. Figure 1(b) shows

an object from the database being placed in front of the system sensor. In gure 1(c), the

system displays the recognized object and it's pose in the upper right corner. The recognition

system used the parametric eigenspace technique Murase and Nayar-1995] for visual learning

and recognition. For related publications, see Nayar and Poggio-1996] Nayar et al.-1996a]

Nayar et al.-1996b] Nayar et al.-1996c] Nene and Nayar-1994]. COIL-20 is available by

(logged) ftp for research purposes (see Section 3).

2 Database Acquisition

The experimental setup used for image acquistion is shown in gure 2. A CCD camera

(SONY XC77) with a 25mm lens was xed to a rigid stand about 1 feet from it's base. A

motorized turntable was placed about 2 feet from the base of the stand. The camera was

tilted down at about 25 degrees to point towards the turntable. This way most objects

appeared at the center of the image when placed at the center of the turntable. To avoid

strong shadows, only ambient ( uorescent) room lighting was used. A black background was

provided by covering the turntable and visible background surfaces with black cloth.

Each object was placed in a stable con guration at approximately the center of the

turntable. The turntable was then rotated through 360 degrees and 72 images were taken

per object; one at every 5 degrees of rotation. The images were digitized using an Analogics

grayscale frame grabber. Two sets of images were stored in the database. The rst set

contains a total of 720 unprocessed 640x480 grayscale images of 10 selected objects from

gure 1. The second set contains 1,440 grayscale images of all the 20 objects and have

been size normalized as follows. The object is clipped out from the black background using

a rectangular bounding box. The bounding box is resized to 128x128 using interpolation-

decimation lters to minimize aliasing Oppenheim and Schafer-1989]. When resizing, aspect

ratio is preserved. Size normalization was necessary for the 20 object recognition system

mentioned above Murase and Nayar-1995].

In addition to size normalization, every image was histogram stretched, i.e. the intensity

of the brightest pixel was made 255 and intensities of the other pixels were scaled accordingly.

The images were saved as 8-bit PGM (portable graymap) images. Note that PGM images

can be viewed with xv. A sample lename of a database image is \obj7 10.pgm". The pre x

obj7 identi es the object. The numeric value 10 following the double underscore separator

identi es the pose. The su x .pgm indicates the le type. The database is available as

two compressed tar les of sizes 12Mb and 6Mb for the unprocessed and processed images

respectively. A color image database COIL-100, similar to COIL-20, is also available. See

Nene et al.-1996].

Page 4

(a)

(b)

(c)

Figure 1: The Columbia Object Image Library (COIL-20) contains 1,440 images of 20 ob-

jects. (a) The objects have a wide variety of complex geometric, appearance and re ectance

characteristics. A real-time 20 object recognition system was constructed using COIL-20.

(b) An object from the database is shown to the system for recognition. (c) The system rec-

ognizes the object in less than one second. The recognized object and it's pose are displayed

in the upper right hand corner.

Page 5

Figure 2: The objects were placed at the center of a motorized turntable. The turntable was

rotated through 360 degrees. An image was acquired with a xed camera at every 5 degrees

of rotation.

3 Access Instructions

COIL-20 is available over the Internet by ftp. All accesses are logged to help us know who is

using the database. The following is a sequence of commands required to download COIL-20:

$ ftp zen.cs.columbia.edu

Name: coil-20

Password: Coil-20

ftp> cd coil-20

ftp> bin

ftp> get coil-20-unproc.tar.gz

ftp> get coil-20-proc.tar.gz

ftp> quit

In case of any problem or questions, the reader is advised to send mail to [email protected]

or [email protected].

Acknowledgements

This database was collected at the Center for Research on Intelligent Systems at the Depart-

ment of Computer Science, Columbia University. It was supported by DOD/ONR MURI

Page 6

Grant N00014-95-1-0601 and a NSF National Young Investigator Award.

References

Murase and Nayar, 1995] H. Murase and S. K. Nayar. Visual Learning and Recognition

of 3D Objects from Appearance. International Journal of Computer Vision, 14(1):5{24,

January 1995.

Nayar and Poggio, 1996] S. K. Nayar and T. Poggio. Early Visual Learning. In S. K. Nayar

and T. Poggio, editors, Early Visual Learning. Oxford University Press, March 1996.

Nayar et al., 1996a] S. K. Nayar, H. Murase and S. A. Nene. Parametric Appearance Repre-

sentation. In S. K. Nayar and T. Poggio, editors, Early Visual Learning. Oxford University

Press, March 1996.

Nayar et al., 1996b] S. K. Nayar, S. A. Nene and H. Murase. Real-Time 100 Object Recog-

nition System. In Proceedings of ARPA Image Understanding Workshop, Palm Springs,

February 1996.

Nayar et al., 1996c] S. K. Nayar, S. A. Nene and H. Murase. Real-Time 100 Object Recog-

nition System. In Proceedings of IEEE International Conference on Robotics and Automa-

tion, Minneapolis, April 1996.

Nene and Nayar, 1994] S. A. Nene and S. K. Nayar. SLAM: A Software Library for Ap-

pearance Matching. In Proceedings of ARPA Image Understanding Workshop, Monterey,

November 1994. Also Technical Report CUCS-019-94.

Nene et al., 1996] S. A. Nene, S. K. Nayar and H. Murase. Columbia Object Image Library:

COIL-100. Technical Report CUCS-006-96, Department of Computer Science, Columbia

University, February 1996.

Oppenheim and Schafer, 1989] A. V. Oppenheim and R. W. Schafer. Discrete-Time Signal

Processing, chapter 3, pages 111{130. Prentice Hall, 1989.