---

We are developing and demonstrating new algorithms for the illumination, atmospheric and temperature invariant recognition of materials in hyperspectral images. The algorithms are based on the use of invariant representations computed from image regions. These invariant representations are derived from physical models for image formation and are independent of viewpoint and the illumination, atmospheric, and thermal environments. We have shown that invariants can be computed that capture arbitrary combinations of spectral and spatial information allowing spectral/spatial tradeoffs to be optimized according to the characteristics of a particular recognition problem. Since the algorithms are derived from physical models, constraints on the physical environment can be incorporated to improve performance. Since invariants can be computed efficiently from hyperspectral imagery without requiring any form of segmentation, an end-to-end recognition system has been constructed based on invariant computation and indexing. Extensive experiments are being conducted that demonstrate the effectiveness of the approach.

Scene 1: HYDICE (210 bands, 0.4µm-2.5µm) image taken at Aberdeen Proving Grounds 24 August 1995. Pixels are approximately 1 m². A zoom of a sunlit green fabric panel is shown in the lower right. The HYDICE spectral radiance for the panel is shown in the lower left.

Scene 2: HYDICE (210 bands, 0.4µm-2.5µm) image taken at Aberdeen Proving Grounds 26 August 1995. Pixels are approximately 1 m². A zoom of a concealed green fabric panel is shown in the lower right. This is the same panel shown in Scene 1. The HYDICE spectral radiance for the panel is shown in the lower left.

A comparison of the normalized HYDICE spectra for the same green fabric panel in the two scenes. The red curve corresponds to Scene 1 and the white curve corresponds to Scene 2.

Recognition results obtained when directly comparing the normalized spectrum for the green fabric from Scene 1 against Scene 2. Many false matches (shown in yellow) are obtained without matching the green fabric panel target (inside white box). This demonstrates the need for the invariant approach developed in [5].

Recognition results obtained using invariant algorithm on Sunlit panels scene. All eight green material panels are correctly identified (red pixels) with few false matches (yellow pixels).

Recognition results obtained using invariant algorithm on Shaded panels scene. Six of six green fabric panels are correctly identified (red pixels) with few false matches (yellow pixels).

Recognition results obtained using invariant algorithm on Concealed panels scene. Five of six target green panels are correctly identified (red pixels, missed panel in blue) with few false matches (yellow pixels).

Recognition results for buildings, roads, and vegetation obtained using invariant algorithm on Fort Hood scene. Results are in strong agreement with material locations. This data was supplied by the Digital Mapping Laboratory, Carnegie-Mellon University.

[1] G. Healey

``Image Understanding Research at UC Irvine: Automatic Recognition in Multispectral Imagery,'' Proceedings of the Image Understanding Workshop, May 1997, 1063-1067.

[2] D. Slater and G. Healey

``ATR in Multispectral Imagery using Physics-based Invariant Representations,'' ATRWG System and Technology Symposium, Redstone Arsenal, AL, October, 1997.

[3] D. Slater and G. Healey

``A Method for Material Classification in AVIRIS Data with Unknown Atmospheric and Geometric Parameters,'' JPL Airborne Earth Science Workshop, January 1998.

[4] D. Slater and G. Healey

``Exploiting an Atmospheric Model for Automated Invariant Material Classification in Hyperspectral Imagery,'' SPIE Conference on Algorithms for Multispectral and Hyperspectral Imagery IV, Orlando, April 1998.

[5] G. Healey and L. Benites

``Linear Models for Spectral Reflectance Functions over the Mid-Wave and Long-Wave Infrared,'' Journal of the Optical Society of America A, 15(8), 2216-2227, August 1998.

[6] D. Slater and G. Healey

``Analyzing the Spectral Dimensionality of Outdoor Visible and Near-infrared Illumination Functions,'' Journal of the Optical Society of America A, 15(11), 2913-2920, November 1998.

[7] G. Healey and D. Slater

``Image Understanding Research at UC Irvine: Automated Invariant Recognition in Hyperspectral Imagery,'' Proceedings of the DARPA Image Understanding Workshop, 631-639, 1998.

[8] B. Thai and G. Healey

``Invariant Subpixel Material Identification in Hyperspectral Imagery,''Proceedings of the DARPA Image Understanding Workshop, 809-814, 1998.

[9] D. Slater and G. Healey

``Reflectance Estimation and Material Identification for 3D Objects in Outdoor Hyperspectral Images,''Proceedings of the DARPA Image Understanding Workshop, 815-820, 1998.

[10] B. Thai, G. Healey, and D. Slater

`` Invariant Subpixel Material Identification In AVIRIS Imagery,'' JPL AVIRIS Workshop, February 1999.

[11] B. Thai and G. Healey

``Invariant ATR for Subpixel Targets in Hyperspectral Imagery,'' ATRWG System and Technology Symposium, Monterey, March, 1999.

[12] D. Slater and G. Healey

``Material Mapping for 3D Objects in Hyperspectral Images,'' SPIE International Symposium on Aerospace/Defense Sensing Simulation and Controls, Orlando, April 1999.

[13] B. Thai and G. Healey

``Invariant Subpixel Target Identification in Hyperspectral Imagery,'' SPIE International Symposium on Aerospace/Defense Sensing Simulation and Controls, Orlando, April 1999.

[14] G. Healey and D. Slater

``Invariant Recognition in Hyperspectral Images,'' Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, Vol. I, 438-443, June 1999.

[15] D. Slater and G. Healey

``A Spectral Change Space Representation for Invariant Material Tracking in Hyperspectral Images,'' SPIE International Symposium on Optical Science, Engineering, and Instrumentation (SPIE Annual Meeting), Denver, 18-23 July 1999.

[16] P. Suen and G. Healey

``Modeling and Recognizing Hyperspectral Textures Under Unknown Conditions,'' SPIE International Symposium on Optical Science, Engineering, and Instrumentation (SPIE Annual Meeting), Denver, 18-23 July 1999.

[17] G. Healey and D. Slater

``Models and Methods for Automated Material Identification in Hyperspectral Imagery Acquired Under Unknown Illumination and Atmospheric Conditions,'' IEEE Transactions on Geoscience and Remote Sensing, 37(6) 2706-2717, November 1999.

[18] P. Suen, G. Healey and D. Slater

``Material Identification over Variation of Scene Conditions and Viewing Geometry,'' SPIE Volume 4049, Algorithms for Multispectral, Hyperspectral, and Ultraspectral Imagery VI, Orlando, Florida, April 2000.

[19] Z. Pan, G. Healey and D. Slater

``Modeling the Spectral Variability of Ground Irradiance Functions,'' SPIE Volume 4049, Algorithms for Multispectral, Hyperspectral, and Ultraspectral Imagery VI, Orlando, Florida, April 2000.

Back to the ICVL homepage