METHOD FOR DETERMINING A CORRECTION VALUE FUNCTION AND METHOD FOR GENERATING A FREQUENCY-CORRECTED HYPERSPECTRAL IMAGE

Abstract

A method for determining a correcting quantity function k.sub.F(x, y) for calibrating an FTIR measurement arrangement with an IR detector. The IR detector includes a plurality of sensor elements, which are each located at a position (x, y), and the method includes: (a) recording interferograms IFG.sub.Rxy of a reference sample using the sensor elements of the IR detector, (b) calculating spectra R.sub.xy of the reference sample by Fourier transforming the interferograms of the reference sample for at least four sensor elements, (c) calculating correcting quantities k.sub.xy by comparing each spectrum R.sub.xy of the reference sample calculated in step b) with a reference data set of the reference sample, and (d) determining the correcting quantity function k.sub.F(x, y) using the correcting quantities k.sub.xy calculated in step c). This permits frequency shifts that occur in FTIR spectrometers with extensive detectors to be effectively corrected regardless of the position of the sensor element.

Claims

1. A method for determining a correcting quantity function k.sub.F(x, y) for calibrating a Fourier Transform infrared (FTIR) measurement arrangement with an infrared (IR) detector, wherein the IR detector comprises a plurality of sensor elements, which are each located at a respective position (x, y), wherein the method comprises: a) recording interferograms IFG.sub.Rxy of a reference sample with the sensor elements of the IR detector; b) calculating spectra R.sub.xy of the reference sample by Fourier transforming the interferograms of the reference sample for at least four of the sensor elements; c) calculating correcting quantities k.sub.xy by comparing each spectrum of the spectra R.sub.xy of the reference sample calculated in said step b) with a reference data set of the reference sample; d) determining the correcting quantity function k.sub.F(x, y) based on the correcting quantities k.sub.xy calculated in said step c).

2. The method as claimed in claim 1, wherein the reference data set comprises a target position ν.sub.L of a selected absorption peak P of the reference sample and said calculating of the correcting quantities k.sub.xy in said step c) is implemented by comparing the target position ν.sub.L with respective actual positions ν.sub.xy of the absorption peak P in the spectra R.sub.xy of the reference sample.

3. The method as claimed in claim 2, wherein the correcting quantities k.sub.xy are determined by subtracting the target position ν.sub.L from the respective actual positions ν.sub.xy or by dividing the respective actual positions ν.sub.xy by the target position ν.sub.L:
k.sub.xy=ν.sub.xy−ν.sub.L or k.sub.xy=ν.sub.xy/ν.sub.L

4. The method as claimed in claim 1, wherein the reference data set comprises a simulated spectrum S.sub.sim with a plurality of absorption peaks of the reference sample and said calculating of the correcting quantities k.sub.xy in said step c) is implemented by comparing the spectra R.sub.xy of the reference sample calculated in said step b) with the simulated spectrum S.sub.sim.

5. The method as claimed in claim 4, wherein the correcting quantities k.sub.xy are determined by maximizing a correlation, by varying the correcting quantities k.sub.xy, between the simulated spectrum S.sub.sim(ν) and the spectra R.sub.xy(ν−k.sub.xy) shifted by k.sub.xy or between the simulated spectrum S.sub.sim(ν) and the spectra R.sub.xy(ν/k.sub.xy) stretched or compressed by 1/k.sub.xy.

6. The method as claimed in claim 1, wherein the correcting quantity function k.sub.F(x,y) is given by $k_F\left(x,y\right)=k_c\cos \left(\arctan \left(\frac{\sqrt{{\left(c_y-y\right)}^2+{\left(c_x-x\right)}^2}}{f_{\mathrm{eff}}}\right)\right)$ where c.sub.y, c.sub.x, f.sub.eff and k.sub.c are parameters for matching the correcting quantity function k.sub.F(x,y) to the calculated correcting quantities k.sub.xy.

7. The method as claimed in claim 1, wherein the correcting quantity function k.sub.F(x,y) is given by
k.sub.F(x,y)=a*(x.sup.2+y.sup.2)+b*x+c*y+d where a, b, c and d are parameters for matching the correcting quantity function k.sub.F(x,y) to the calculated correcting quantities k.sub.xy.

8. The method as claimed in claim 6, wherein the correcting quantity function k.sub.F(x,y) is matched to the calculated correcting quantities k.sub.xy by minimizing an error function:
Σ.sub.xy(k.sub.F(x,y)−k.sub.xy).sup.2.

9. The method as claimed in claim 7, wherein the correcting quantity function k.sub.F(x,y) is matched to the calculated correcting quantities k.sub.xy by minimizing an error function:
Σ.sub.xy(k.sub.F(x,y)−k.sub.xy).sup.2.

10. The method as claimed in claim 6, wherein the parameters required for the matching are determined by setting up equations with the correcting quantities k.sub.xy calculated in said step c) for at least the four sensor elements to produce a system of equations and by solving the system of equations through curve fitting.

11. The method as claimed in claim 7, wherein the parameters required for the matching are determined by setting up equations with the correcting quantities k.sub.xy calculated in said step c) for at least the four sensor elements to produce a system of equations and by solving the system of equations through curve fitting

12. A method for generating a frequency-corrected hyperspectral image of a sample with an FTIR measurement arrangement including an IR detector having a plurality of sensor elements, comprising, for each of the sensor elements which are respectively located at a position (x,y) of the IR detector: recording an interferogram IFG.sub.Pxy with an equidistant sampling grid a.sub.xy with the sensor element; Fourier transforming the interferogram, to determine a spectrum S.sub.xy(ν) with a frequency axis; wherein the spectrum S.sub.xy(ν) for each of the sensor elements is corrected with the correcting quantity function k.sub.F(x,y) determined according to claim 1.

13. The method as claimed in claim 12, wherein the correcting quantity function k.sub.F(x,y) is determined by stretching or compressing the spectra R.sub.xy of the reference sample calculated in said step b) or by dividing the target position ν.sub.L and the actual position ν.sub.xy of the selected absorption peak P of the reference sample, and wherein each spectral point (ν.sub.n, I.sub.n) of the spectrum S.sub.xy(ν) of the sample is corrected to (ν.sub.n/k.sub.F(x,y), I.sub.n).

14. The method as claimed in claim 12, wherein the correcting quantity function k.sub.F(x,y) is determined by shifting the spectra R.sub.xy of the reference sample calculated in said step b) or by subtracting the target position ν.sub.L and the actual position ν.sub.xy, and wherein each spectral point (ν.sub.n, I.sub.n) of the spectrum S.sub.xy(ν) is corrected to (ν.sub.n−k.sub.F(x,y), I.sub.n).

15. The method as claimed in claim 12, wherein the correcting quantity function k.sub.F(x,y) is determined by stretching or compressing the spectra R.sub.xy of the reference sample calculated in said step b) or by dividing the target position ν.sub.L and the actual position ν.sub.xy, wherein the interferogram IFG.sub.Pxy is recorded with a sampling grid of a.sub.xy=a.sub.0/k.sub.F(x, y) and wherein the spectra of the sample are subsequently generated by a Fourier transform of the corrected interferograms, where a.sub.0 provides a value for the sampling grid when calculating the values for the frequency axis.

16. The method as claimed in claim 12, wherein the correcting quantity function k.sub.F(x,y) is determined by stretching or compressing the spectra R.sub.xy of the reference sample calculated in said step b) or by dividing the target position ν.sub.L and the actual position ν.sub.xy, wherein the interferogram IFG.sub.Pxy is recorded with a sampling grid of a.sub.xy=a.sub.0 and wherein the spectra of the sample are subsequently generated by a Fourier transform of the interferograms, where a.sub.0*k.sub.F(x,y) provides a value for the sampling grid when calculating the values for the frequency axis.

17. The method as claimed in claim 1, wherein the FTIR measurement arrangement is an IR microscope.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0045] FIG. 1 shows the setup of an FTIR microscope for reflection-type FTIR measurements.

[0046] FIG. 2 shows the course of IR beams which emerge from the interferometer at different angles and strike an IR detector.

[0047] FIG. 3 shows the basic method steps of the method according to the invention for determining the correcting quantity function k.sub.F(x.sub.y).

[0048] FIG. 4 shows, in spatially resolved fashion, the measured band position of an absorption peak of a reference sample selected by way of example, for the various sensor element positions (x,y) of the detector.

[0049] FIG. 5 shows, in spatially resolved fashion, the calculated correcting quantities k.sub.xy for the various sensor element positions (x,y) of the detector.

[0050] FIG. 6 shows, in spatially resolved fashion, a correcting quantity function k.sub.F(x,y) which was determined in step d) with the aid of the correcting quantities k.sub.xy calculated in step c) and an analytical model for the various sensor element positions (x,y) of the detector.

[0051] FIG. 7 shows, in spatially resolved fashion, the difference between the correcting quantity function k.sub.F(x,y) of FIG. 6 and the correcting quantities k.sub.xy, calculated in step c), of FIG. 5 for the various sensor element positions (x,y) of the detector.

[0052] FIG. 8 shows the curve of the correcting quantity function k.sub.F(x,y) of FIG. 7 as a function of the pixel number of the IR detector.

[0053] FIG. 9 shows the curve of the values of the correcting quantity k.sub.xy of FIG. 5 as a function of the pixel number of the IR detector.

[0054] FIG. 10 shows a diagram which represents the wavenumber accuracy of the selected absorption peak as a function of the pixel number of the IR detector, both with and without correction.

[0055] FIG. 11 shows a spectrum of a reference sample (air with water vapor and CO.sub.2) measured with a sensor element.

[0056] FIG. 12 shows a portion (absorption lines of water vapor) of the spectrum R.sub.xy(ν) illustrated in FIG. 11 following a logarithmization, baseline correction and normalization.

[0057] FIG. 13 shows a simulated absorption spectrum S.sub.sim(ν) for water vapor.

[0058] FIG. 14 shows a diagram in which the spectra R.sub.xy(ν) and S.sub.sim(ν) of FIGS. 12 and 13 are plotted together.

[0059] FIG. 15 shows the correlation of the spectrum R.sub.xy(ν/ν.sub.L) of FIG. 12, modified by a correcting quantity k.sub.xy, with the simulated spectrum S.sub.sim(ν) of FIG. 13 as a function of the correcting quantity k.sub.xy.

[0060] FIG. 16 shows the comparison of the simulated spectrum S.sub.sim(ν) with the measured spectrum modified with k.sub.xy=0.9973099.

[0061] FIG. 17 shows the basic method steps of the method according to the invention for generating a frequency-corrected hyperspectral image.

[0062] FIG. 18 shows the method steps of variants I and II of the method according to the invention for generating a frequency-corrected hyperspectral image.

[0063] FIG. 19 shows the method steps of a variant III of the method according to the invention for generating a frequency-corrected hyperspectral image.

[0064] FIG. 20 shows the method steps of a variant IV of the method according to the invention for generating a frequency-corrected hyperspectral image.

DETAILED DESCRIPTION

[0065] FIG. 1 shows an FTIR microscope in a reflection arrangement. The infrared light of an IR source 1 is collected by a mirror 2, collimated and steered into a (modified) Michelson interferometer 3. Here, the light strikes a beam splitter 3a, which transmits some of the radiation and passes the latter to a fixed mirror 3b and which reflects another part of the radiation and steers the latter to a movable mirror 3c. The light reflected at the mirrors 3b and 3c then is superimposed again at the beam splitter 3a and leaves the interferometer 3. The infrared light modulated by the interferometer 3 leaves the interferometer 3 and is guided into microscope optics 4. There, it is steered via various mirrors 4a to a beam splitter or half-mirror 4b, from where it is guided into an objective 4c (condenser), which illuminates the microscopic sample situated at the sample position 5. The sample at the sample position 5 interacts with the modulated infrared radiation and reflects some of the radiation. The reflected radiation is subsequently captured by the objective 4c and focused on the infrared detector 6. On its path to the infrared detector 6, the radiation passes the half-mirror or beam splitter 4b.

[0066] FTIR microscopy can also be implemented in transmission. A transmission FTIR microscope (not shown) comprises a further objective which is used to steer the light from the interferometer 3 onto the sample while the objective 4c is used to focus the light transmitted by the sample onto the infrared detector.

[0067] The objective 4c generates an image of the sample plane 5 on the sensor of the infrared detector 6. If the infrared detector consists of a plurality of sensitive elements which are arranged in a row or in an array, it is possible to simultaneously examine a plurality of regions of the sample with spatial resolution. Each element of the detector 6 can record an interferogram, from which it is then possible to calculate an infrared spectrum. Consequently, each pixel of the detector 6 registers a spatially resolved spectrum of the sample. In practice, such an area detector 6 can have various embodiments. In addition to detectors in which small detector elements are arranged in rows or in an array, use is also made of so-called focal plane arrays, in which the infrared-sensitive pixels are read in a manner similar to a CCD camera.

[0068] The microscope optics 4 firstly ensure that the collimated radiation from the interferometer 3 illuminates the sample and secondly ensure that, with the aid of the modulated radiation an image of the sample arises on the sensor elements of the detector.

[0069] FIG. 2 schematically shows the microscope optics 4 with the effective focal length f.sub.eff, which focus collimated radiation from the interferometer 3 on the area detector 6 with sensor elements 7. In this case, the central sensor element (on the optical axis) is reached by the radiation from the interferometer 3 which extends in collimated fashion and parallel to the optical axis. By contrast, sensor elements (pixels) which are at a distance d from the center of the area detector 6, through which the optical axis extends, see collimated radiation from the interferometer 3 which extends within the interferometer 3 with an inclination to the optical axis at an angle α. In this case, the following relationship applies:

tan α=d/f.sub.eff.

[0070] Since every sensor element 7 (pixel) of the detector 6 records radiation at a different angle α, the spectra recorded by the various sensor elements are compressed to different extents depending on the position of the sensor element 7 on the detector 6; absorption bands in the spectra have differently pronounced shifts.

[0071] According to the present invention, the spectra recorded by the various sensor elements 7 or parts of the spectra of a reference sample recorded by the various sensor elements 7 are evaluated on an individual basis for each sensor element and a separate correction is undertaken for each sensor element 7. The steps of the method according to the invention required to this end are illustrated in FIG. 3: initially, interferograms IFG.sub.Rxy are recorded for a reference sample (in this case: water vapor) with the sensor elements of the IR detector. Spectra R.sub.xy are calculated by Fourier transforming the interferograms. The spectra can be calculated for all sensor elements, but at least for four sensor elements. Correcting quantities k.sub.xy are determined by comparing the spectra R.sub.xy of the reference sample calculated in step b) with a reference data set of the reference sample. By way of example, the reference data set can be obtained from specialist literature or through a simulation. The calculated correcting quantities k.sub.xy are used to determine a correcting quantity function k.sub.F(x, y), which specifies the values of the correcting quantity k.sub.xy depending on the pixel position (x,y). There is no need to use the entire measured spectrum within the comparison according to the invention of the calculated spectra with the reference data set; instead, parts of the calculated spectra, for example a certain spectral range or else an individual absorption peak, can be used for the comparison. Two variants with which a corresponding correcting quantity function k.sub.F(x,y) can be determined are shown below.

Variant 1: Determining the Correcting Quantity Function Using a Selected Absorption Peak

[0072] FIG. 4 shows the result of a measurement of the relative position of a selected absorption peak (in this case: absorption band of water vapor with a target position at 1576.130 cm.sup.−1) using an FTIR microscope, wherein use was made of a detector with a 32×32 pixel detector array (FPA). The relative position of the absorption band was evaluated for each spectrum and hence for each sensor element of the detector. FIG. 4 shows the position (actual position) of this water vapor band as grayscale values, encoded as a function of the detector row and the detector column. It is evident that the relative position of the selected band is at a maximum at approximately row 30 and column 7. From there, concentric rings are formed around this maximum with decreasing values of the band position (absorption peak is shifted to smaller wavenumbers). This is compatible with the theory since the angles α increase and the spectra are increasingly compressed with increasing distance of the pixels from the optical axis, which does not strike the sensor exactly in the center in this example.

[0073] The correcting quantities k.sub.xy for the respective sensor elements are calculated by virtue of a comparison value for the selected absorption peak, for example the target position of the selected absorption peak known from the literature, being compared (by subtraction or division) to the actual position of the corresponding absorption peak determined from the measured spectrum.

[0074] FIG. 5 illustrates the result of such a comparison (correcting quantities k.sub.xy as a function of the detector row and the detector column), in which the measured peak positions were divided by the comparison value ν.sub.L (k.sub.xy=ν.sub.xy/ν.sub.L).

[0075] In FIG. 9 the correcting quantity k.sub.xy is presented as a function of the pixel position/pixel number. It is evident that the correcting quantity k.sub.xy determined from the reference sample is noisy. This is due to the fact that the individual spectra from which the band positions of water vapor were determined also have a certain amount of noise. This noise influences the accuracy of the determination of the band position. Disadvantageously, the noise in the correcting quantity k.sub.xy is transferred to all subsequent measurements that are corrected with this correcting quantity k.sub.xy. Better results can be obtained if further insight about the cause of the compression of the frequency axis is also included. This can be implemented by the use of an analytical model, as described below: it was already determined that radiation running at an angle α through the interferometer experiences smaller optical path length differences A between the two interferometer arms than radiation parallel to the optical axis:

Δ=2L cos α  (1)

[0076] As a result of the reduced modulation frequency by the interferometer, which has been reduced by a factor of cos α, the measured frequencies ν′ of the selected absorption peak likewise deviate from the true frequencies ν by the factor cos α.

ν′=ν cos α  (2)

[0077] Each sensor element with the coordinates x and y on the detector 6 records radiation that runs through the interferometer at an angle α(x,y), with

tan α(x,y)=sqrt((x−c.sub.x).sup.2+(y−c.sub.y).sup.2)/f.sub.eff  (3)

[0078] In this case, the optical axis intersects the sensor of the detector 6 at the coordinates (c.sub.x, c.sub.y).

[0079] If Equations 2 and 3 are combined, it is possible to describe the measured relative band position ν′=ν.sub.xy as a function of the position of the sensor element (pixel) on the detector 6. Here, c.sub.x and c.sub.y are the coordinates at which the optical axis intersects the detector, ν.sub.c is the measured relative band position at this position and f.sub.eff is the effective focal length of the microscope optics.

[00002]$\begin{array}{cc}{v}_{\mathrm{xy}}=v_c.Math.\cos a=v_c.Math.\cos \left(\arctan \left(\frac{\sqrt{{\left(c_y-y\right)}^2+{\left(c_x-x\right)}^2}}{f_{\mathrm{eff}}}\right)\right)& \left(4\right)\end{array}$

[0080] The following correcting quantity arises:

[00003]$k=\frac{v_{\mathrm{xy}}}{v_L}=\frac{v_c\cos \left(\arctan \left(\frac{\sqrt{{\left(c_y-y\right)}^2+{\left(c_x-x\right)}^2}}{f_{\mathrm{eff}}}\right)\right)}{v_L}$$k_F\left(x,y\right)=k_c\cos \left(\arctan \left(\frac{\sqrt{{\left(c_y-y\right)}^2+{\left(c_x-x\right)}^2}}{f_{\mathrm{eff}}}\right)\right)$

[0081] The following applies to small angles: arctan x≈x and cos x≈1−x.sup.2/2. Thus, it is necessary to find a quadratic function of the form

[00004]$\begin{array}{cc}k\left(x,y\right)=& \frac{v_c}{v_L}.Math.(1-a.Math.\left({\left(c_y-y\right)}^2+{\left(c_x-x\right)}^2\right)=\\ & k_c.Math.(1-a.Math.\left({\left(c_y-y\right)}^2+{\left(c_x-x\right)}^2\right)\\ =& -k_{c}a.Math.x^2+2k_c{\mathrm{ac}}_x.Math.x-k_{c}a.Math.y^2+\\ & 2k_c{\mathrm{ac}}_y.Math.y+k_c-k_c{\mathrm{ac}}_y^2-k_c{\mathrm{ac}}_x^2\end{array}$$k\left(x,y\right)=a^{\prime }.Math.\left(x^2+y^2\right)+b^{\prime }.Math.x+c^{\prime }.Math.y+d^{\prime }$

which the measurement data describe to the best possible extent. It is possible to set up an equation of the aforementioned type for each of the n pixels with the coordinates (x, y).sub.n and the correcting quantity k.sub.n. All n equations can then be represented in matrix form:

[00005]$A.Math.\stackrel{.fwdarw.}{x}=\stackrel{.fwdarw.}{k}.Math.\left(\begin{array}{cccc}{x}_1^2+y_1^2& x_1& y_1& 1\\ x_2^2+y_2^2& x_2& y_2& 1\\ \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}\\ x_n^2+y_n^2& x_n& y_n& 1\end{array}\right).Math.\left(\begin{array}{c}{a}^{\prime }\\ b^{\prime }\\ c^{\prime }\\ d^{\prime }\end{array}\right)=\left(\begin{array}{c}{k}_1\\ k_2\\ \mathrm{.Math.}\\ k_n\end{array}\right)$

[0082] This system of equations has n equations (e.g., 32×32) and four unknowns. Thus, it is overdetermined. The best optimized solution (least-squares fit) is:

[00006]$\left(\begin{array}{c}{a}^{\prime }\\ b^{\prime }\\ c^{\prime }\\ d^{\prime }\end{array}\right)={\left(A^T.Math.A\right)}^{-1}.Math.A^T.Math.\left(\begin{array}{c}{k}_1\\ k_2\\ \mathrm{.Math.}\\ k_n\end{array}\right)$

Thus, to obtain a solution, all that is needed is knowledge about matrix multiplication, the formation of a transposed matrix and the formation of an inverse matrix. By way of example, the inverse matrix can be formed using Cramer's rule with determinants. From the results a′, b′, c′, d′, it is possible to calculate the desired parameters c.sub.x, c.sub.y, k.sub.c, and a.

[00007]$c_x=-\frac{b^{\prime }}{2a^{\prime }}$$c_y=-\frac{d^{\prime }}{2a^{\prime }}$$k_c=d^{\prime }-a^{\prime }.Math.c_y^2-a^{\prime }.Math.c_y^2$$a=-\frac{a^{\prime }}{k_c}$

[0083] FIGS. 6 and 8 show, by way of example, the correcting quantity function k.sub.F(x,y), calculated via the above-described model, for a peak position P of a reference absorption peak at ν.sub.L=1576.130 cm.sup.−1. Solving the system of equations arising from the analytical model yields:

c.sub.x=5.12346

c.sub.y=31.9599

k.sub.c=0.9999918157

a=4.20110015E−8

This yields a noise-free correcting quantity function k.sub.F(x,y).

[0084] The difference between the correcting quantity function k.sub.F(x,y) calculated with the analytical model and the correcting quantities k.sub.xy calculated from the measured data is illustrated in FIG. 7.

[0085] FIG. 10 shows a comparison of the wavenumber accuracy or frequency accuracy for a measurement using an imaging FTIR microscope at 1576.130 cm.sup.−1 without frequency correction and with the described frequency correction. FIG. 10 plots the deviation of the relative band position from the literature value (1576.130 cm.sup.−1) as a function of the pixel number (number of the sensor element). It is quite evident that the wavenumber accuracy with the correction is an order of magnitude better than without correction.

Variant 2: Determining the Correcting Quantity Function Using a Selected Region of a Spectrum

[0086] Instead of using a selected peak for the purposes of calculating the correcting quantity function, it is also possible to use the complete spectra or a selected frequency range of the spectra of the reference sample measured by the individual sensor elements. Such a spectrum is shown in FIG. 11 for a reference sample of air with water vapor and CO.sub.2. FIG. 12 shows a portion of the spectrum illustrated in FIG. 11, specifically absorption lines of water vapor. The spectrum shown in FIG. 12 was subject to logarithmization, baseline correction and normalization, with it also being possible to dispense with normalization in principle.

[0087] A simulated absorption spectrum for water vapor as shown in FIG. 13 is used as a reference data set for determining the correcting quantities k.sub.xy. The simulated spectrum shown in FIG. 13 was simulated with the aid of HITRAN (high-resolution transmission molecular absorption database). In FIG. 14, the two spectra (the selected portion of the measured spectrum and the simulated spectrum) are plotted together for comparison purposes. It is evident that the spectra are shifted in relation to one another.

[0088] There now is an iterative correction of the frequency axis of the measured spectrum by virtue of each frequency value ν.sub.n of a spectral point (ν.sub.n, I.sub.n) being divided by a correcting quantity k.sub.xy of the sensor element at the position (x,y) such that the spectral point is altered to (ν.sub.n/k.sub.xy, I.sub.n).

[0089] To this end, there is an iterative variation of the correcting quantity k.sub.xy and correlations of the spectra modified by k.sub.xy with the simulated spectrum are calculated for each sensor element. FIG. 15 shows the corresponding correlation of the spectra shown in FIG. 14, as a function of the correcting quantity k.sub.xy. In the example shown, a maximum correlation arises for k.sub.xy=0.9973099.

[0090] FIG. 16 shows the comparison of the simulated spectrum with the measured spectrum modified by the determined correction value k.sub.xy=0.9973099. It is evident that (in contrast to FIG. 14) the peaks of the simulated spectrum coincide with the peaks of the corrected measured spectrum.

[0091] According to the invention, the correcting quantity function determined is used to correct the spectra of a sample measured by various sensor elements of an IR detector of an FTIR measurement arrangement in order to obtain a frequency-corrected hyperspectral image of the sample (FIG. 17). To determine a hyperspectral image, an interferogram IFG.sub.Pxy with an equidistant sampling grid a.sub.xy is initially recorded with the sensor element for each sensor element of the IR detector. By Fourier transforming the interferogram IFG.sub.Pxy, a spectrum S.sub.xy(ν) with a frequency axis is determined. According to the invention, the spectrum S.sub.xy(ν) of each sensor element is corrected by a correcting quantity function k.sub.F(x,y).

[0092] In this case, the correction can be implemented in the calculated spectrum, i.e., after the Fourier transform (variants I and II, illustrated in FIG. 18), or during the recording and the subsequent Fourier transform (variant III, illustrated in FIG. 19, and variant IV, illustrated in FIG. 20):

[0093] The individual spectra of each sensor element can be corrected in the frequency axis by virtue of the frequency axis of each spectrum being stretched by the corresponding correcting quantity, i.e., each frequency or wavenumber is multiplied by an appropriate factor (variant I). This initially ensures that the frequency grid in the spectra of the individual sensor elements is no longer the same. However, the spectra can be returned to the same frequency grid by a possible interpolation.

[0094] In variant II, the spectra are shifted in relation to the frequency axis by the corresponding correcting quantity k.sub.xy.

[0095] In variant III, the correcting quantities k.sub.xy are already taken into account when recording the interferograms by virtue of the interferograms being sampled with a grid a.sub.xy=a.sub.0/k.sub.F(x, y) that has been stretched in relation to the sampling grid a.sub.0 theoretically required for the measurement with a sensor element situated on the optical axis for those sensor elements where a compression of the frequency axis is expected. By way of a Fourier transform, the intensity values I.sub.n of the spectra S.sub.xy are calculated. Subsequently, the associated frequencies are determined for the calculated intensity values I.sub.n: ν.sub.n=n/(N*a.sub.0). Here, a.sub.0 is used as a value for the sampling grid. This procedure ensures that the spectra of all sensor elements are frequency-corrected and also have the same frequency grid. The equidistant frequency grid with Δν=1/(N*a.sub.0) extends over a range of 0 to (N/2−1)*Δν for all spectra. Here, N denotes the number of recorded points in each interferogram [4].

[0096] In the variant IV, the interferograms are recorded with the sampling grid a.sub.0. The correction is implemented by virtue of the frequency axis being calculated on an individual basis for each spectrum S.sub.xy such that an equidistant frequency grid arises for each spectrum from 0 to (N/2−1)*Δν.sub.xy, with Δν.sub.xy=1/(N*a.sub.0*k.sub.F(x, y)).

[0097] In all variants, a hyperspectral image is obtained, in which the influence of the positioning of the various sensor elements in relation to the optical axis of the FTIR measurement arrangement is taken into account.

CITATIONS

[0098] [1] Robert John Bell, [0099] Introductory Fourier Transform Spectroscopy, [0100] Academic Press, 1972 [0101] [2] E. V. Lowenstein [0102] Fourier Spectroscopy: An Introduction, Aspen Int. Conf. on Fourier Spectroscopy, [0103] 1970, p. 3, AFCRL-71-0019, 5 Jan. 1971, Spec. Rep. No. 114 [0104] [3] Peter R. Griffith, James A. de Haseth [0105] Fourier transform infrared Spectrometry [0106] Vol. 83 in Chemical Analysis [0107] pp. 32-39, [0108] [4] Werner Herres and Joern Gronzolz, Understanding FTIR Data Processing

LIST OF REFERENCE SIGNS

[0109] 1 IR source [0110] 2 Mirror [0111] 3 Interferometer [0112] 3a Beam splitter [0113] 3b, 3c Mirror [0114] 4 Microscope optics [0115] 4a Mirror [0116] 4b Half-mirror [0117] 5 Sample position [0118] 6 IR detector [0119] 7 Sensor elements of the IR detector EXPRESSIONS [0120] (x,y) Position of a sensor element of the IR detector [0121] k.sub.xy Correcting quantity for the sensor element at the position (x,y) [0122] k.sub.F(x, y) Correcting quantity function [0123] ν.sub.xy Actual position of a selected absorption peak P in the spectrum [0124] ν.sub.L Target position of a selected absorption peak P in the spectrum [0125] R.sub.xy Spectrum of the reference sample, measured by the sensor element at the position (x,y) [0126] S.sub.sim Simulated spectrum of the reference sample [0127] S.sub.xy(ν) Spectrum of the sample, measured by the sensor element at the position (x,y) [0128] IFG.sub.Pxy Interferogram of the sample with an equidistant sampling grid a.sub.xy, measured by the sensor element at the position (x,y) [0129] a.sub.xy Sampling grid for measuring the interferogram of the sample with the sensor element at the position (x,y) [0130] a.sub.0 Base sampling grid; preferably chosen such that the entire spectrum is located in the spectral range from 0 to ν.sub.max for axiparallel beams [0131] ν.sub.max Maximum frequency which can be recorded with a certain sampling grid

[00008]$v_{\max }=\frac{N}{2}\Delta v=\frac{1}{2a}$ [0132] (ν.sub.n, I.sub.n) Spectral point within a spectrum S.sub.xy(ν) of the sample [0133] In Intensity of the n-th spectral point in the spectrum S.sub.xy(ν) of the sample (at the frequency position ν.sub.n) [0134] N Number of recorded points in the interferogram