• Ei tuloksia

Multispectral Imaging systems

The multispectral imaging systems that have been developed at UCL consisted of two different cameras to capture in combination with two lighting systems for both reflective and transmissive imaging, through a series of bandpass filters.

The cameras and lens were:

(1) Digital SLR Nikon D200, which captures RGB images in the visible spectrum.

The camera is equipped with a CCD sensor of dimensions 23.6x15.8 mm with pixel dimension of 6.06 microns. It produces an image of 3900x2600 pixels in 3:2 aspect ratio with 16 bits per pixel. The camera was fitted with Micro-Nikkor AF-S 105mm 1:2.8 GED lens for camera spectral sensitivity measurement and

Micro-Nikkor AF 60mm 1:2.8 lens to capture Macbeth and white balance chart.

The RGB images were captured as per channel 16-bit files in RAW format.

Figure 10: DSLR Nikon D200 camera [114].

(2) An IDS UI 5480CP-M-GL camera (hereafter called U I), which captures monochrome images in the wide spectral range. The camera is equipped with a CMOS sensor of dimensions 5.6x4.2 mm with pixel dimension of 2.2 microns. It produces an image of 2560x1920 pixels in 4:3 aspect ratio with 12 bits per pixel.

The M-GL version of the camera has no Bayer mask or infra-red cut-off filter, and each pixel is sensitive to the whole spectral range of silicon 400–1100 nm [115]. It weighs only 61g and can be powered through an Ethernet cable. The camera was fitted with an unbranded 5mm 1:2.8 lens. The monochrome images were captured as 12-bit files in PNG format.

Figure 11: UI 5480CP-M-GL monochrome camera

For the Nikon camera, a set of 16 bandpass optical interference filters was used, spaced at regular intervals across the visible spectrum. For the monochrome camera additional 5 filters were used to cover the near-infra-red range. Each of the first 16 filters had a bandwidth of approximately 20 nm, at wavelength intervals of 20 nm throughout the visible spectrum from 400 to 700 nm inclusive. The last 5 filters had a bandwidth of approximately 50 nm, at intervals of 50 nm in the near-infrared spectrum from 750 to 950 nm inclusive. All were circular glass filters in 52 mm rings for screw-mounting onto the front of the lens of each camera. This arrangement guaranteed that no stray light entered the lens. The spectral transmittance of each filter was measured with an Ocean Optics HR2000+ spectrometer (Figure 12) [60].

Figure 12: Transmittance factors of 21 optical bandpass filters in the visible and NIR spectrum [60]

For both camera systems, the samples were illuminated by four tungsten-halogen lamps on a standard photographic document copystand (Figure. 13). Light sources (4x Photolux 240V 150W tungsten halogen, frosted, ES27 fitting) were mounted in pairs on lateral support arms with elevation angle of lights 30 deg (i.e. 60 deg to normal) and a working distance of 90cm between the target and lens. The spectral power distribution of lights was measured with PhotoResearch PR-650 spectroradiometer.

Exposure time varied for each filter.

Figure 13: Acquisition setup for Nikon D200 camera on a Copystand with four tungsten-halogen lamps for reflective imaging of samples.

The filter is screw-mounted into the front of the lens [60].

In total 21 images were captured for each target by the monochrome camera. A set of Macbeth Colorchecker chart, Macbeth Dark, White card and White card dark images were acquired by this system. The monochrome images were captured as 12-bit files in PNG format.

For the Nikon camera 16 images were captured for each target. A set of Macbeth and White card images were acquired by this system. The RGB images were captured as 16-bit per channel in RAW format that was converted to TIFF file format using DRAW software.

In addition, Eye one X-rite Spectrophotometer (Figure 15 i ) was used for the spectral reflectance measurement of white card and each colour patches in the wavelength range 380-730nm with 10nm bandwidth intervals (Figure 14).

Figure 14: Spectral reflectance data of color patches measured from X-rite Spectrophotometer

Figure 15: Spectral reflectance curve of white card measured in 10 series (i) from Eye one X-rite Spectrophotometer (i).

380 480 580 680

Re*lectance factor

Wavelength (nm)

Series1 Series2 Series3 Series4 Series5 Series6 Series7 Series8 Series9 Series10

3.3.3 University Milano (Italy) & University Jean Monnet (UJM, France): Multispectral Imaging systems

The multispectral imaging systems developed by the two research groups together at UM & UJM adopted two different camera systems:

(1) The Imaging Source DMK41AU02.AS captures monochrome images with 31 narrowband optical filters with spectral resolution of 10nm over the range 400–

700nm. The camera is equipped with a monochrome ½ inch CCD sensor and produces image of 1280 x 960 pixels [117]. The camera was fitted with a Nikon AF Nikkor 24mm f/2.8 lens that provided the image focusing with C-mount adapter.

Light sources (2x ~3200K 500-Watt) were placed at 2X45° geometry with a working distance of 100cm between the target and lens. Exposure time varied for each filter (from 5s to 1/5). The system is controlled by software and was set to capture images in the visible range (400-700 nm) with 10nm interval with a total number of 31 images.

Figure 16: DMK41AU02.AS monochrome camera [116].

(2) Digital DSLR CMOS camera Canon 1000D captures RGB images in the visible spectrum. The camera was coupled with a yellow wide band filter for the second shot which produced 5 images over the range 400–700nm. Light sources (2x ~3200K 500-Watt) were placed at 2X45° geometry with a working distance of 130cm between the target and lens. Exposure time varied for the first shot was 1/160s (unfiltered) and 1/60s for the second one (filtered). A Canon EF-S 60mm f/2.8 Macro lens provided the image focusing.

Figure 17: DSLR Canon 1000D camera [117].

Calibration workflow for both devices was based on a standard linear reflectance calibration on 99% reflectance spectralon plus non-uniformity correction with neutral gray card. For Canon system, the Macbeth image dataset was reconstructed to 31 monochrome images with 859 x 1289 pixels using Matlab reconstruction matrix. Both systems provided Macbeth images as 16-bit files in HDF5 (.h5) format.

3.3.4 University of Basel (UB): Multispectral Imaging system

The multispectral imaging system developed at UB adopted a sequential acquisition of images with a set of color filter array (CFA) being inserted sequentially in front of the objective.

Figure 18: Acquisition setup for Nikon D3 [120].

A set of 13 interference filters (Andover) with 10nm bandwidth was used; their central wavelengths are between 420nm and 660nm every 20nm (Figure 19).

Figure 19: Transmittance factors of 13 interference filters in the visible spectrum[120].

Digital camera with a color filter array (CFA) are designed to directly acquire color information; this kind of imaging device is not ideal for a multispectral purpose;

however, CFA cameras are cheap and easily available in comparison to other dedicated sensors, and their combination with a set of interference filters can be a cheap and decent solution for multispectral imaging.

Figure 20: DSLR Nikon D3 camera [119].

The digital images were acquired with a DSLR camera (Nikon D3) equipped with a 12Mpixel CMOS full-frame sensor. The spatial sampling density was 246ppi; the image focusing was provided by AF-S NIKKOR 50mm f/1.4 G objective lens. Two Broncolor flashlights equipped with umbrella diffusers were used as light sources; they were placed in 2X45° geometry at 50 cm distance from the object plane, and a good uniformity of illuminance was found (Figure 18).

All the images were shot in the same geometrical arrangement and with the same camera parameters (f:8,1/250sec), adjusting the flash power to correctly expose the channel most sensible to the wavelengths selected by the interference filters. In all images a diffuse white reflector (white) and a light trap (black) were included in the framing. The white consisted in the bottom-left patch of the colorchecker (whiteCC), corrected with its tabulated reflectance factor (R%whiteCC(λ)). The images were saved by the camera in the proprietary RAW format using 14bits; they were then externally converted in TIFF/16bit format using the DCRAW software [Coffin 2013], choosing the parameters to linearly translate the raw data. The RGB channels were split and the properly exposed channel was selected (raw_image) [120].

To provide the transmittances, the following equation was used [120]:

Transmittances = raw_image – black / white – black

The dataset collected from Nikon D3 using 13 interference filters were in the spectral range 420-660nm with 20nm bandwidth interval. These dataset were misregistered due to a nudge of the setup during acquisition, so it was aligned using Adobe Photoshop CS6 before having the image cube created and sent.

3.3.5 Science & Technology in Archaeology Research Centre (STARC): Multispectral Imaging system

The multispectral imaging device at STARC is MuSIS HS system originally developed for the study of artworks. (Figure 21)

(a) (b) Figure 21: Study of artwork using MuSIS system as (a) Imaging Colorimeter or (b) Imaging Spectrometer [120]

This device consists of an imaging monochromator and optical tuneable filtering system integrated to the CCD optical detector that is 1/3 inch square pixel progressive scan CCD sensor. In addition, the system is equipped with IR to visible imaging converter. The combination of spectral and color imaging with calibration enables the system to operate as either Imaging Spectrometer or Imaging Colorimeter [120].

Figure: MuSIS camera[120]

Figure 22: Hardware setup for MuSIS HS system [120]

The system has specially developed software, which is employed for the control of camera sensitivity, filter control and image acquisition etc., as well as for the image analysis and data handling. It performs spectral imaging in 34 selectable spectral bands in the range of 360-1550 nm for B & W imaging and color imaging mode.

MuSIS HS displays in real time and captures several spectral images of 1600x1200 pixels.

Figure 23:

MuSIS software for acquisition control [120]

Calibration method was based on pre-determined settings of the camera for Spectral range, light source and imaging geometry. The spectral scanning of the Macbeth target was in the spectral range 420-1000nm with 20nm interval range of monochrome images. Two Halogen light source (EIKO: 12 Volt / 50 watt / 4700K; Q50MR16 CG/47/36, 36 Flood, MR16 Halogen Display Lamp - Natural Daylight), were selected as a lighting condition that was set at 450 imaging geometry. The stored gray value at the end of the photometry process was used to normalize the brightness of the captured spectral images [120].

The STARC data included 4 versions. There is a focus shift for this data as the camera was manually focused between filters. The dataset was in the wavelength range 420-1000nm with 20nm bandwidth interval. Each dataset consisted of 30 waveband images. All four files are for the ColorChecker and involve two calibration iterations.

The camera was calibrated and two sets of data were acquired and then the camera was recalibrated (in the same way as the first calibration) and another two sets of data were acquired. This was done in order to compare the results between calibrations and to be sure the calibration would be accurate.

3.3.6 Technical University of Catalonia (UPC) (Spain):