Multispectral imaging - Paul Bourkepaulbourke.net/papers/multispectral/multispectral.pdf · • Note that here we are interested in identification, much of multispectral imaging

Multispectral imaging

Rock art is often very obvious and interesting Other times less so

Concept

• A normal photograph is throwing away a huge amount of information.

• The energy across a range of wavelengths is being (weighted summed) into just 3 numbers, single R,G,B values.

• Can imagine materials that reflect strongly in different wavelengths but appear to be the same colour.

wavelength

intensity

BG

R

350nm 400nm 450nm 500nm 550nm 600nm 650nm

Human visual system,response of cones

CMOS sensor with standardRGB Bayer filters

• Most SLR cameras use a Bayer filter, each RGB filter has it’s own response curve.

• Also applies to the human eye which has cones that respond to different wavelength bands.

Image cube (x,y, )

wavelengthimage plane

R

G

B

x

y

Hand print, West Angeles rock shelter.

x

y

wavelength

image plane

x

y

“Continuous” wavelength multispectral cameras

• Generally a line scan camera + a diffraction grating, or similar wavelength splitting device.

• See also “pushbroom” multispectral cameras.

• Increasingly being used in the mining industry with mineral signature databases.

300 400 500 600 7006505504503500.0

0.2

0.4

0.6

0.8

1.0Power

Low cost alternative

• Capture narrow wavelength ranges.

• For this initial experiment used 8 interference bandpass filters across the visible range.350nm to 700nm.

• Filter banks 50nm apart and 20nm wide.

wavelength

intensity

350nm 400nm 450nm 500nm 550nm 600nm 650nm

~20nm

700nm

50nm

8 interference bandpass filters

20nm

Image cube (x,y, )

wavelengthimage plane

x

y

Example

• Might imagine multiplying 500nm and 550nm and subtracting 650nm.

• Note that here we are interested in identification, much of multispectral imaging is more about quantitative analysis.

400nm 450nm 500nm

550nm 600nm 650nm

Future work

• Demonstrated the potential with two of three images from the West Angeles rock shelters.

• Next test will be 16 filters and a more convenient means of changing filters.

• Develop algorithms to optimally combine slices to identify features.

• Employ higher dynamic range B/W camera. Avoid multiplicative response curves of Bayer filter.