JShakyaPsych2012Project

Back to Psych221-Projects-2012

Background

Hyperspectral images contain spectral response from each pixel in the scene and hence contain more information than standard RGB image. Based on exact shape of the spectrum of each pixel, it is possible to identify the material composition and distribution in the reflecting surface. If it is a painted surface, it is possible to identify the exact color with same spectral response. In recent years, there has been many advancement in fluroscent microscopy, which uses light spectrum to identify molecules in biological samples. Using hyperspectral camera, we can identify fluorophores and hence spatial distribution of certain molecules. Different illuminants used in cameras have different spectrum and hence using hyperspectral camera, we can identify the illuminant and correct for proper illumination.

Motivation

The RBG image captured by digital cameras contain 3 colors per pixel. However each pixel in fact has its own charateristic spectrum based on reflectance of that pixel. In RGB cameras, the spectrum is reduced into 3 values (R,G and B) by placing 3 RGB filter in front of 3 CMOS sensors. These filters have their own transmission spectrum so that each filter passes only a limited band of spectrum through it. The 3 number for R,G and B are obtained by multiplying the image spectrum by the corresponding filters. In contrast to RGB camera, a hyperspectral camera, in fact captures the entire spectrum. For doing that the hyperspectral camera receives a line image, passes through a prism which decomposes spectral composition and then various CMOS sensors are different wavelengths capture intensity at each wavelength. In such hyperspectral camera the full 2D view is captured by rotating the camera to acquire successive line images. However this takes time and also requires physical motion in the camera. Often due to scanning nature, it requires scene to be rather static so that it doesn't change while it is scanning. In this project we are proposing a Camera System which can capture hyperspectral images the same way as digital camera in a single shot.

Related Work

To be written.

Methods

Interferometric Color Filter

Human visual system can only detect intensity and frequency (hence wavelength). However light has various other properties such as polarization, phase etc which our eyes cannot see. However through interference phenomenon we can convert phase difference of two light sources into intensity which we can see. There has been many recent usage of interference techniques in imaging systems. One of such application of interference phenomenon is to create color filters. When light reflected from two reflective surfaces (mirrors) seperated by distances in the order of wavelength, based on the distance and wavelength of light it is possible to have either constructive or distructive interference. Such a device is called Fabrey-Perot Interferometer. It consists of two semi-transparent mirrors seperated by transparent medium. When light enters the first mirror, part of it is reflected and part of it is transmitted. A portion of transmitted light gets reflected by second mirror, a portion of which gets transmitted out of first mirror and rest goes through mutiple reflections inside the medium. When the phase difference between light reflected from first mirror and second mirror is pi, the two waves distructively interference and the device becomes minimally reflective. Under same condition, the waves which are transmitted through the device contructively interfere and hence most of the light energy is transmitted through the device. The extent to which a particular wavelength is transmitted depends on the reflectances of two mirrors and their seperation.

MEMS Tunable Interferometric Filters

The interferometric filter describe above can be built as a MEMS device, where two semi-reflective layers on glass substrate are seperated by air. By using electrostatic drive, it is then possible to change the seperation between two mirrors by applying certain voltage. The seperation is a direct function of applied DC voltage, when other factors such as restoring spring force is constant.

• Figure 1 - Filter Transmission Spectra for N number of Filters
• 
  Number of Filters = 2 to 16

Interferometric Hyperspectral Image Sensor

Using above described Tunable color filters, it is possible to take multiple images of a scene applying different filter each time. If the filters' spectral resolution is higher than highest frequency content (in wavelength space) in the scene spectrum, it is possible to represent the scene using images captured with different filters. The number of filters required and their sharpness will depend on nature of the scene. Since most of the natural scene don't have rapid variations in spectrum, it is possible to capture natural hyperspectral scenes using less number of filters than that will be required for spectrum with sharp rapid variations. For example capturing absorption or emission spectra of a many species of molecule will require large number of filters to preserve the shape of the spectrum.

Conclusions

<TBD>

References

<TBD>

Appendix I - Code and Data

<TBD>