Zhihong Pan

Face recognition in hyperspectral images

Hyperspectral cameras provide useful discriminants for human face recognition that cannot be obtained by other imaging methods. We examine the utility of using near-infrared hyperspectral images for the recognition of faces over a database of 200 subjects. The hyperspectral images were collected using a CCD camera equipped with a liquid crystal tunable filter. Spectral measurements over the near-infrared allow the sensing of subsurface tissue structure, which is significantly different from person to person but relatively stable over time. The local spectral properties of human tissue are nearly invariant to face orientation and expression, which allows hyperspectral discriminants to be used for recognition over a large range of poses and expressions. We describe a face recognition algorithm that exploits spectral measurements for multiple facial tissue types. We demonstrate experimentally that this algorithm can be used to recognize faces over time in the presence of changes in facial pose and expression.

Hyperspectral face recognition under variable outdoor illumination

We examine the performance of illumination-invariant face recognition in outdoor hyperspectral images using a database of 200 subjects. The hyperspectral camera acquires 31 bands over the 700-1000nm spectral range. Faces are represented by local spectral information for several tissue types. Illumination variation is modeled by low-dimensional spectral radiance subspaces. Invariant subspace projection over multiple tissue types is used for recognition. The experiments consider various face orientations and expressions. The analysis includes experiments for images synthesized using face reflectance images of 200 subjects and a database of over 7,000 outdoor illumination spectra. We also consider experiments that use a set of face images that were acquired under outdoor illumination conditions.

Multiband and spectral eigenfaces for face recognition in hyperspectral images

Spectral reflectance properties of local facial regions have been shown to be useful discriminants for face recognition. To evaluate the performance of spectral signature methods versus purely spatial methods, face recognition tests are conducted using the eigenface method for single-band images extracted from the hyperspectral images. This is the first such comparison based on the same dataset. Selected sets of bands as well as PCA transformed bands are also used for face recognition evaluation with individual band processed separately. A new spectral eigenface method which preserves both spatial and spectral features is proposed. All algorithms based on spectral and/or spatial features are evaluated under the same framework and are compared in terms of accuracy and computational efficiency.

Hyperspectral face recognition for homeland security

Hyperspectral sensors provide useful discriminants for human face recognition that cannot be obtained by other imaging methods. Near-infrared spectral measurements allow the sensing of subsurface tissue structure which is significantly different from person to person but relatively stable over time. The spectral properties of human tissue are also nearly invariant to changes in face orientation which bring significant degradation to most other face recognition algorithms. We examine the utility of using near-infrared hyperspectral images for the recognition of human subjects over a database of 200 subjects. The face recognition algorithm exploits spectral measurements for individual facial tissue types and combinations of facial tissue types. We demonstrate experimentally that hyperspectral imaging promises to support face recognition independent of facial expression and orientation.

Hyperspectral face recognition under unknown illumination

We examine the performance of illumination-invariant face recognition in outdoor hyperspectral images using a database of 200 subjects. The hyperspectral camera acquires 31 bands over the 700- to 1000-nm spectral range. Faces are represented by local spectral information for several tissue types. Illumination variation is modeled by low-dimensional spectral radiance subspaces. Weighted invariant subspace projection over multiple tissue types is used for recognition. The experiments consider various face orientations and expressions. The analysis includes experiments for images synthesized from indoor face reflectance images of 200 subjects, using a database of more than 7,000 outdoor illumination spectra. We also examine a set of images of 10 subjects of the 200 that were acquired under outdoor conditions using a calibrated hyperspectral camera.

Comparison of spectral-only and spectral/spatial face recognition for personal identity verification

Face recognition based on spatial features has been widely used for personal identity verification for security-related applications. Recently, near-infrared spectral reflectance properties of local facial regions have been shown to be sufficient discriminants for accurate face recognition. In this paper, we compare the performance of the spectral method with face recognition using the eigenface method on single-band images extracted from the same hyperspectral image set. We also consider methods that use multiple original and PCA-transformed bands. Lastly, an innovative spectral eigenface method which uses both spatial and spectral features is proposed to improve the quality of the spectral features and to reduce the expense of the computation. The algorithms are compared using a consistent framework.

Illumination-invariant face recognition in hyperspectral images

We examine the performance of illumination-invariant face recognition in hyperspectral images on a database of 200 subjects. The images are acquired over the near-infrared spectral range of 0.7-1.0 microns. Each subject is imaged over a range of facial orientations and expressions. Faces are represented by local spectral information for several tissue types. Illumination variation is modeled by low-dimensional linear subspaces of reflected radiance spectra. One hundred outdoor illumination spectra measured at Boulder, Colorado are used to synthesize the radiance spectra for the face tissue types. Weighted invariant subspace projection over multiple tissue types is used for recognition. Illumination-invariant face recognition is tested for various face rotations as well as different facial expressions.

Global spectral irradiance variability and material discrimination at Boulder, Colorado

We analyze 7,258 global spectral irradiance functions over 0.4-2.2 microm that were acquired over a wide range of conditions at Boulder, Colorado, during the summer of 1997. We show that low-dimensional linear models can be used to capture the variability in these spectra over both the visible and the 0.4-2.2 microm spectral ranges. Using a linear model, we compare the Boulder data with the previous study of Judd et al. [J. Opt. Soc. Am. 54, 1031 (1964)] over the visible wavelengths. We also examine the agreement of the Boulder data with a spectral database generated by using the MODTRAN 4.0 radiative transfer code. We use a database of 223 minerals to consider the effect of the spectral variability in the global spectral irradiance functions on hyperspectral material identification. We show that the 223 minerals can be discriminated accurately over the variability in the Boulder data with subspace projection techniques.

Modeling the spectral variability of ground irradiance functions

We analyze a set of 7,258 0.4-2.2 micron ground spectral irradiance functions measured on different days over a wide range of conditions. We show that a low-dimensional linear model can be used to capture the variability in these measurements. Using this linear model, we compare the data with a previous empirical study. We also examine the agreement of the data with spectra generated by MODTRAN 4.0. Using a database of 224 materials, we consider the implications of the observed spectral variability for hyperspectral material discrimination using subspace projection techniques.

Recognizing 3D objects in hyperspectral images under unknown conditions

We present models and algorithms for recognizing 3D objects in airborne 0.4-2.5 micron hyperspectral images acquired under unknown conditions. Objects of interest exhibit complex geometries with surfaces of different materials. The DIRSIG image generation software is used to build spatial/spectral surfaces of different materials. The DIRSIG image generation software is used to build spatial/spectral subspace models for the objects that capture a range of atmospheric and illumination conditions and viewing geometries. Since we consider scales for which multiple materials will mix in a pixel, the object subspace models also account for spectral mixing. An important aspect of the work is the use of methods for partitioning object subspaces to optimize performance. The new algorithms have been evaluated using hyperspectral data that has been synthesized for a range of conditions.