WO2015071786A1

WO2015071786A1 - Method for imaging of spectral reflectance at several wavelengths

Info

Publication number: WO2015071786A1
Application number: PCT/IB2014/065036
Authority: WO
Inventors: Janis Spigulis
Original assignee: Institute of Solid State Physics University of Latvia
Current assignee: Institute of Solid State Physics University of Latvia
Priority date: 2013-11-12
Filing date: 2014-10-03
Publication date: 2015-05-21
Anticipated expiration: 2016-05-12
Also published as: LV14782A; LV14782B

Abstract

The invention relates to spectral imaging, in particular to imaging of spectral reflectance distribution at several fixed wavelengths using a single RGB image data set. In the proposed method for obtaining several spectral reflectance images, they are extracted from a digital RGB image data set at the wavelengths corresponding to those used at polychromatic illumination of the target, providing linearity of the photo-response, using the reference reflector and taking into account the overlapping of the RGB photo-detector spectral sensitivity bands, where the spectral reflectance is calculated for each image pixel or group of pixels using a calibrated grey reference reflector with reflection coefficients at all exploited wavelengths less than those of the white reference and by value is close to the largest reflection coefficient of the displayed object area. In case of different intensities of the illumination spectral lines, corrections for spectral reflectance values are proposed.

Description

Method for imaging of spectral reflectance at several wavelengths Tehnical field

The invention relates to spectral imaging, in particular to imaging of spectral reflectance distribution at a number of fixed wavelengths using a single RGB image data array.

Background Art

Spectral reflectance (http://rsgislearn.blogspot.coin/2007/04/spectral-reflectance.html) (X) is the ratio Ι(λ)/Ι₀(λ), where Ι(λ) is the object surface area's reflected intensity at single fixed wavelength λ and Ι₀(λ) - the reflected intensity from fully reflective reflector (the so-called white reference) under the same illumination conditions at wavelength λ.

Digital image sensors (two-dimensional photo matrices) supplied with filters permeable in specific spectral regions often are used for spectral imaging, e.g. by means of a rotating disk with a set of different colour filters (E.C. Ruvolo et al., Proc. SPIE, Vol. 7548, 75480A, 2010); acousto-optic or liquid crystal filters which can be spectrally scanned with electrical signals (http://www.usgs. gov/science/science.php?term=765) are used, as well. Spectral filters of three colours (blue, green and red) are integrated in digital RGB image sensors, where the said filters in various combinations with external spectral filters provide increased number of spectral images (US 7612822 B2, US 2009290124 (Al), JP 2008136251 (A)). Another way to obtain a set of spectral images is using of light sources that emit in a number of different spectral bands (e.g. different colour LED - WO 2008093988 (Al)) with sequential object illumination, taking a single image at each spectral band.

There are known methods for obtaining a number of spectral images from a single digital RGB image data set. For example, numerical values of signals recorded in red (R), green (G) and blue (B) spectral band for each image point or pixel are defined - Ri, Gi and Bi ( i - the pixel number) and they are used for spectral imaging in the red, green and blue portions of spectrum, and/or multi- spectral analysis is performed by dividing such sub- images, subtracting one from the other, etc. (D. Kapsokalyvas et al., Proc. SPIE, Vol. 7548, 754808, 2010). By fixing a definite discrimination level of the sensor output signals above which any wavelength in a sensor spectral sensitivity range corresponds to only one or two of the R, G, B signal values, further logical analysis allows obtaining up to six narrower spectral intervals from the RGB data set (J. Spigulis et al., Proc. SPIE, Vol. 7557, 75570M, 2010). Variable discrimination level significantly widens the number of spectral intervals obtained from a single digital RGB data set (WO 2012/002787 A 1, 2012).

This and other technical solutions may be implemented in a single snapshot mode when RGB image is captured at poly-chromatic illumination of the object, i.e. it is illuminated simultaneously by a number of spectral bands being as narrow as possible. The polychromatic illumination has particular advantages - it makes possible to reconstruct the spectral images immediately, as compared, for example, with the sequential illumination by a number of broadband LED. A possibility for mapping the skin oxy-haemoglobin relative concentration distribution at bi-chromatic laser illumination by using data of a single RGB image cube has been demonstrated (J. Spigulis et al., Proc. SPIE, Vol. 7557, 75570M, 2010). The method, however, does not allow determining the specific spectral reflectance values for each image pixel at the two fixed wavelengths.

Since the three spectral sensitivity curves of the RGB images sensor partially overlap, none of the colour channel signals can describe adequately the perceived intensity of monochromatic radiation. It is possible to correct the effect of inter-channel crosstalk at a fixed wavelength, if the RGB channel spectral sensitivity curves are known and linearity of photo-response is ensured (J. Spigulis et al., Proc. SPIE, v.8216, 82160L, 2012).

A method and device for obtaining a number of spectral images from a single RGB data set according to the number of illumination spectral lines was proposed (WO 2013135311 Al). This solution allows to define the numerical values of (X) for each object image point (pixel) at a number of fixed wavelengths and provides parametric imaging of the (X) values at each of the selected wavelengths with the „crosstalk" correction. It is achieved by placing a fully reflective element (white reference) in the displayed object area, and its reflected signals are compared with those from different areas of the object.

The described solution has also some weaknesses:

- The limit of photo-resonse linearity is defined by the signal reflected from the white reference, which in real situations can be considerably more intense than the signal reflected from the object surface (such as a human skin). Therefore, the ratio of signal/noise (S/N) and the accuracy of the k(¾) values to be determined (i - number of the spectral lines used for illumination) at particular image pixels of the object are relatively low. For example, if each pixel of the detector can distinguish 256 intensity levels, the average noise level is 20 and the signal level reflected from the white reference area is 250, and the signal levels reflected from the object surface are in the range between 40 and 80, then the S/N value for the reference signals is 15 (relative error 8%), and that for the object signals is in the range between 2 and 4 ( relative error 25...50%), which results in the relative error of the k( i) values to be determined in the range between 33% and 58%;

- Inter-channel crosstalk corrections at fixed wavelength bands are calculated assuming that all the spectral lines of the illumination are equally intensive. Using actually available light sources, spectral line intensities can differ, thus lowering the accuracy of the k( i) values to be determined. The application of equalizing filters in such situations is quite complicated and also expensive solution.

Disclosure of the Invention

The aim of the invention is to increase the imaging quality of the spectral reflectance (X) and the accuracy of the determined (X) values by using a single RGB image data set at the conditions of polychromatic illumination by several spectral lines.

To increase the detection accuracy of the signals reflected from the object surface, it is proposed to replace the white reference with a calibrated grey reference, which reflects all the spectral lines of the used illumination wavelengths and its reflection coefficient is close to that of a reflective surface of the object - e.g. 10% above the highest value of the signal reflected by the object. It is possible to calibrate the reflection coefficient of the grey reference k' = VIo (I - reflected intensity from the grey reference, Io - reflected intensity from the white reference) at fixed wavelengths with accuracy less than 1%. For the calculation of spectral reflectance values, formulas of WO 2013135311 Alcan be used, wherein signals of the white reference are replaced by signals of the grey reference; the adjusted spectral reflectance values are obtained by multiplication of the found ki, k₂, k_n values by k' . The obtained positive effect - reduced relative error of the reflected intensity measurements - is illustrated by a numerical example when the signal level reflected from the grey reference is 250, the signal levels reflected from the object are in the range between 150 and 225 and the image sensor noise level is 20. In this case the corresponding S/N values for the reference and for the object is 12.5 and between 7.5 and 11.3, with the corresponding relative errors 8% and 9-13%, respectively. The relative error of k(¾) value to be determined is only between 18 and 23%, i.e. more than 2 times lower than using the white reference (see the example described above). The application of the grey reference also allows increasing significantly the level of illumination of the object, providing linearity of photoresponse, which results in improved quality (contrast, resolution) of the acquired spectral image. In brief, the replacement of white reference with the grey one significantly increases the accuracy of k( i) values to be determined, as well as the quality of the spectral reflectance images. During the capture of spectral reflectance image, intensities of the spectral lines involved in polychromatic illumination may differ. In such cases adjustments to the image processing calculation formulas are necessary, because the signals received from all the image pixels depend not only on the spectral sensitivity S of images sensor at the selected wavelengths, but also on the relative intensities of the respective spectral lines of the illumination.

In case of bi-chromatic illumination it is proposed to introduce a correction coefficient c₁₂ = Ii/I₂ (the ratio of illumination intensities at wavelengths i and λ₂) and to multiply with it the relative photosensitivity SR, SG and/or SB, as defined in WO 2013135311 Al . By replacing the corresponding values of SR, SG and/or SB with products of CI₂*SR, CI₂*SG and/or CI₂*SB in formulas for calculating of spectral reflectance, the different intensities of spectral lines at bi-chromatic illumination are taken into account.

The correction due to intensity differences at tri-chromatic illumination has to be carried out similarly: it is proposed to introduce correction coefficients c₁₂ = Ii/I₂, c₁₃ = Ii/I₃ and c₂₃ = I₂/I₃ (the ratios of illumination intensities at wavelengths λι, λ₂ and λ_3ι respectively). In the next step, sensor relative photo- sensitivities of the images at the respective wavelengths are multiplied with the correction coefficients in all three RGB sensitivity bands, and the S(R ) S(R ) relative photo-sensitivities used in WO 2013135311 Al - S„„ = — , S„,, = — ,

^RU S(R₂ ) ^RU S(R3) s =¾ s - ^{S(Gl )} s - ^{S(Gl )} s - ^S(Gl) s s - ^{S(Bl )} ^R23 5(R₃ ) ' ⁰¹² 5(G₂ ) ' ^G13 5(G₃ ) ' ^G23 5(G₃) ' ⁵¹² S(B₂ ) ' ⁵¹³ S(B₃ )

S(B )

and S_B23 = — are replaced with these products. In particular, the used adjusted values

S(B₃ )

with which S_Ri₂, S_G12, S_B12, S_R13, S_G13, S_B13, S_R23, S_G23 and S_B23 values are replaceable in the known solution, are as follows: ci₂*S_Ri₂, c_i2*S_Gi₂, ci₂*S_Bi₂, ci₃*S_Ri₃, ci₃*S_Gi₃, ci₃*S_Bi₃, C₂₃*S_R23, C₂₃*S_G23 and c₂₃*S_B23.

In general, by separating n spectral reflectance images from a single RGB data set at illumination by n different spectral lines, adjustment is to be carried out in the system matrix defined in WO 2013135311 Al, which consists of R, G and B channel spectral sensitivities at all wavelength bands, respectively λι, λ₂, ..., λ_η: A In

this situation, correction by means of ratio of two spectral line intensities is not rational due to large number of such correction coefficients . It is proposed first to define the relative intensities of each illumination spectral line with respect to the most intense of them, i.e. positive values / , I₂ , I_n ', which do not exceed 1. Next, all members of matrix column 1 are multiplied with / , all members of matrix column 2 - with I₂ , etc.; all members of matrix column n - with /„'. Such corrected A-matrix is used to define the specified spectral reflectance values for each image pixel or pixels group at unlimited number n of the illumination spectral lines.

Claims

Claims:

1. A method for imaging of spectral reflectance at several wavelengths, where spectral reflectance images are obtained from a single digital RGB image data set at a number of wavelengths, corresponding to the number of spectral lines of polychromatic illumination, providing linearity of photo-response, using a reference reflector and taking into account the overlapping of the RGB sensitivity spectral bands, wherein the spectral reflectance is calculated for each image pixel or group of pixels using a calibrated grey reference reflector, reflectance coefficient of which at all exploited wavelengths is less than that of the white reference and by value is close to the largest reflection coefficient of the displayed object area.

2. The method according to claim 1, wherein due to differences of spectral line intensities at bi-chromatic illumination the spectral reflectance values are determined by introducing a correction coefficient c₁₂ = Ii/I_2, being the ratio of illumination intensities at the wavelengths λι and λ₂, respectively, and the relative photosensitivity values S_R, S_G and/or S_B at the selected wavelengths in the known calculation formula are replaced with the products of CI₂*S_R, CI₂*S_G and/or CI₂*S_B-

3. The method according to claim 1, wherein due to differences of spectral line intensities at tri-chromatic illumination, the spectral reflectance value errors are adjusted by introducing correction coefficients ci₂ = Ii/I₂, ci₃ = Ii/I₃ and c₂₃ = I₂/I₃, being the ratio of illumination intensities at wavelengths λι, λ₂ and λ₃, respectively, and they are multiplied with the relative photo- sensitivities of the image sensor at the respective wavelength pairs, and the relative photosensitivity values in the known formulas for spectral reflectance calculation are replaced with these products.

4. The method according to claim 1, wherein due to intensity differences of n spectral lines at poly-chromatic illumination, the spectral reflectance value errors are adjusted, first, by defining the intensity ratio of each illumination spectral line to the most intense of them, i.e. positive values / , I2 , I_n ', not exceeding 1, and then by multiplication of all the members of column 1 with / , all the members of column 2 with I2 etc., including multiplication of all the members of the matrix column n with /„' (n>2) in system matrix , which consists of R, G and B channel spectral sensitivities at the

wavelength bands λι, λ₂, ..., λ_η.