Light source spectrum estimation system. light source spectrum estimation method and program

Abstract

To provide a light source spectrum estimation system capable of estimating the light source spectrum of any captured image.SOLUTION: There is provided a light source spectrum estimation system that comprises an achromatic area extraction part which estimates an achromatic area of a subject included in an image and extracts pixels of the achromatic area, and a light source spectrum estimation part which inputs the pixel values of the pixels of the achromatic area into a trained model and estimates the spectrum of the light source in the image.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a light source spectrum estimation system, a light source spectrum estimation method, and a program. [Background technology]

[0002] Conventionally, there is a technology for converting an image captured by a digital camera into an image under a light source different from that at the time of capture. This conversion uses information on the spectrum of the light source. Patent Document 1 discloses a technology for estimating the spectrum of the light source (illumination light) at the location where image data was captured from image data captured of a standard plate (patch) on which multiple color samples with different spectral reflectances are arranged on the same plane.

[0003] Methods for detecting achromatic areas from an image include GrayEdge (see Non-Patent Document 1), GrayPixel (see Non-Patent Document 2), MeanShiftedGrayPixel (see Non-Patent Document 3), and GraynessIndex (see Non-Patent Document 4). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 11-287712 [Non-patent literature]

[0005] [Non-Patent Document 1] Joost van de Weijer, Theo Gevers, Arjan Gijseni, “Edge-Based Color Constancy”, IEEE Transactions on Image Processing, 2007, 16 (9), pp.2207-2214. ff10.1109 / TIP.2007.901808ff. ffinria-00548686 [Non-Patent Document 2] Kai-Fu Yang Shao-Bing Gao Yong-Jie Li, “Efficient Illuminant Estimation for Color Constancy Using Gray Pixels”, [online], [Retrieved June 6, 2023], Internet <https: / / openaccess.thecvf.com / content_cvpr_2015 / papers / Yang_Efficient_Illuminant_Estimation_2015_CVPR_paper.pdf> [Non-Patent Document 3] Yanlin Qian, Said Pertuz, Jarno Nikkanen, Joni-Kristian Kamarainen and Jiri Matas, “Revisiting Gray Pixel for Statistical Illumination Estimation”, [online], [Retrieved June 6, 2020], Internet,<https: / / www.scitepress.org / Papers / 2019 / 74069 / 74069.pdf> [Non-Patent Document 4] Yanlin Qian, Joni-Kristian Kamarainen, Jarno Nikkanen, Jiri Matas, “On Finding Gray Pixels”, [online], [Retrieved June 6, 2020], Internet, <https: / / openaccess.thecvf.com / content_CVPR_2019 / papers / Qian_On_Finding_Gray_Pixels_CVPR_2019_paper.pdf> Summary of the Invention [Problem to be solved by the invention]

[0006] However, the technique described in Patent Document 1 has a problem in that an image must be captured by including an object with a known spectral reflectance, such as a patch, in the angle of view.

[0007] The present invention has been made in consideration of the above circumstances, and provides a light source spectrum estimation system, a light source spectrum estimation method, and a program capable of estimating the light source spectrum for any captured image without using an object with known spectral reflectance, such as a patch. [Means for solving the problem]

[0008] The present invention has been made to solve the above-mentioned problems, and one aspect of the present invention is a light source spectrum estimation system comprising an achromatic region extraction unit that estimates an achromatic region of a subject included in an image and extracts pixels of the achromatic region, and a light source spectrum estimation unit that inputs pixel values ​​of the pixels of the achromatic region into a trained model and estimates the spectrum of a light source in the image.

[0009] Another aspect of the present invention is the above-mentioned light source spectrum estimation system, which includes a learning unit that generates the trained model by machine learning the relationship between the light source spectrum and pixel values ​​obtained by multiplying the spectral sensitivity of the camera and the light source spectrum by the spectral reflectance of an achromatic color.

[0010] Another aspect of the present invention is the above-mentioned light source spectrum estimation system, wherein the light source spectrum includes a spectrum obtained by adding perturbation to a predetermined light source spectrum.

[0011] Another aspect of the present invention is the light source spectrum estimation system described above, wherein the achromatic region extraction unit extracts a predetermined number of pixels from the achromatic region, and the light source spectrum estimation unit inputs pixel values ​​of the predetermined number of pixels to the trained model in a predetermined order based on the pixel values ​​of the predetermined number of pixels, thereby estimating the spectrum of the light source in the image.

[0012] Another aspect of the present invention is the light source spectrum estimation system described above, wherein the achromatic region extraction unit extracts the predetermined number of pixels by arranging pixels of the achromatic region in an order based on pixel values ​​and extracting pixels at predetermined intervals from the order based on pixel values.

[0013] Another aspect of the present invention is the above-mentioned light source spectrum estimation system, wherein the achromatic region extraction unit clusters pixels of the achromatic region into the predetermined number of clusters based on pixel values, and sets pixels consisting of representative values ​​of each of the clusters to the predetermined number of pixels.

[0014] Another aspect of the present invention is the above-mentioned light source spectrum estimation system, wherein the achromatic region extraction unit forms the pixels of the achromatic region into a histogram consisting of the predetermined number of bins, calculates an average value of pixel values ​​for each of the predetermined number of bins, and sets the pixel consisting of the average value as the predetermined number of pixels.

[0015] Another aspect of the present invention is a method for estimating a light source spectrum, comprising the steps of estimating an achromatic region of a subject included in an image and extracting pixels of the achromatic region, and inputting pixel values ​​of the pixels of the achromatic region into a trained model to estimate the spectrum of a light source in the image.

[0016] Another aspect of the present invention is a program for causing a computer to function as an achromatic region extraction unit that estimates achromatic regions of a subject included in an image and extracts pixels of the achromatic region, and as a light source spectrum estimation unit that inputs pixel values ​​of the pixels of the achromatic region into a trained model and estimates the spectrum of a light source in the image. Effect of the Invention

[0017] According to the present invention, the light source spectrum estimation system, light source spectrum estimation, and program can estimate the light source spectrum for any captured image. [Brief description of the drawings]

[0018] [Figure 1] FIG. 1 is a schematic block diagram showing a configuration of a light source spectrum estimation system 100 according to an embodiment of the present invention. [Diagram 2] 4 is a flowchart illustrating the operation of the light source spectrum estimation device 10 in the embodiment. [Diagram 3] FIG. 13 is a diagram showing an example of an image G1 in the same embodiment. [Figure 4] 13 is a diagram showing an example of an image in which the grayness value of each pixel of image G1 is pseudo-colorized and output in the embodiment. FIG. [Diagram 5] 11 is a diagram showing an example of an image from which an achromatic color area is extracted in the embodiment. FIG. [Figure 6] 13 is a graph showing an example of an estimated light source spectrum S in the embodiment. [Figure 7] 10 is a flowchart illustrating the operation of the trained model generation device 20 in the embodiment. [Figure 8] 13 is an example of achromatic spectral reflectances stored in the achromatic spectral reflectance DB 24 in the embodiment. [Figure 9] 13 is another example of achromatic spectral reflectances stored in the achromatic spectral reflectance DB 24 in the embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0019] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. Fig. 1 is a schematic block diagram showing the configuration of a light source spectrum estimation system 100 according to an embodiment of the present invention. When an image G1 captured by a digital camera is input, the light source spectrum estimation system 100 estimates the spectrum S of the light source when the image G1 was captured.

[0020] The light source spectrum estimation system 100 includes a light source spectrum estimation device 10 and a trained model generation device 20. Each of the light source spectrum estimation device 10 and the trained model generation device 20 may be realized by one or more computers reading and executing a program. Furthermore, the light source spectrum estimation device 10 and the trained model generation device 20 may be realized by a single computer, or may be connected to each other so as to be able to communicate with each other via a network or the like.

[0021] The light source spectrum estimation device 10 includes an achromatic region extraction unit 11 and a light source spectrum estimation unit 12. The achromatic region extraction unit 11 estimates the achromatic region of the subject included in the input image G1, and extracts pixels of the achromatic region. A known method such as GrayEdge, GrayPixel, MeanShiftedGrayPixel, and GraynessIndex can be used to estimate the achromatic region. The light source spectrum estimation unit 12 inputs the pixel values ​​of the pixels extracted by the achromatic region extraction unit 11 to a trained model M, and estimates the spectrum S of the light source using the trained model M. The trained model M is a model that has been machine-learned to learn the relationship between the pixel values ​​of the subject having an achromatic spectral reflectance and the light source spectrum.

[0022] The trained model generating device 20 includes a learning unit 21, a camera spectral sensitivity DB (Data Base) 22, a light source spectrum DB 23, and an achromatic color spectral reflectance DB 24. The learning unit 21 performs machine learning of the relationship between the light source spectrum and a pixel value calculated by multiplying the camera spectral sensitivity and the light source spectrum by the achromatic color spectral reflectance. For the machine learning, a convolutional neural network (CNN) or multiple regression analysis may be used.

[0023] The camera spectral sensitivity DB 22 stores the spectral sensitivity of each of the multiple cameras. The spectral sensitivity is a value obtained by dividing each pixel value output by the camera by the amount of light in each band that constitutes the spectrum. For example, if there are three pixel values, R (red), G (green), and B (blue), and the spectrum consists of band 1 to band λ, the spectral sensitivity of one camera consists of 3×λ elements, such as the sensitivity of R band 1, the sensitivity of R band 2, ..., the sensitivity of R band λ, the sensitivity of G band 1, ..., the sensitivity of G band λ, the sensitivity of B band 1, ..., the sensitivity of B band λ.

[0024] The light source spectrum DB 23 stores the light source spectrum of each of a plurality of light sources. The number of bands constituting the light source spectrum may be the same as the number of bands λ used in the spectral sensitivity. The achromatic color spectral reflectance DB 24 stores the spectral reflectance of each of a plurality of achromatic objects. The number of bands constituting the spectral reflectance may be the same as the number of bands λ used in the spectral sensitivity.

[0025] FIG. 2 is a flow chart for explaining the operation of the light source spectrum estimation device 10 in this embodiment. First, the achromatic region extraction unit 11 estimates the achromatic region of the image G1 (step Sa1). Next, the achromatic region extraction unit 11 extracts a predetermined number of pixels from the achromatic region (step Sa2). Here, the predetermined number may be 1 or an integer of 2 or more. When the predetermined number is 1, any method may be used to determine the one pixel to be extracted. For example, a pixel randomly selected from the achromatic region may be the pixel to be extracted, or the average value of each of the pixel values ​​R, G, and B in the achromatic region may be the pixel value of the pixel to be extracted. When the predetermined number is an integer of 2 or more, any method may be used to determine the pixel to be extracted. For example, a first method using an order based on pixel values, a second method using clustering, a third method using a histogram, or the like may be used.

[0026] In the first method, the achromatic region extraction unit 11 arranges pixels in the achromatic region in an order based on pixel values, and extracts pixels from the order at a predetermined interval. Here, the order may be an ascending or descending order of the norm or total value of pixel values, or an ascending or descending order of any of R, G, and B. The predetermined interval may be Np / (Ne+1), where Np is the number of pixels in the achromatic region and Ne is the predetermined number. When extracting pixels at a predetermined interval, the first pixel extracted may be offset from the beginning of the order. This offset may be, for example, half the predetermined interval.

[0027] In the second method, the achromatic region extraction unit 11 clusters the pixels in the achromatic region into a predetermined number of clusters based on pixel values, and extracts pixels having representative values ​​of each cluster as the predetermined number of pixels. The representative value of a cluster may be, for example, the average value of the pixels constituting the cluster.

[0028] In the third method, the achromatic region extraction unit 11 organizes the pixels in the achromatic region into a histogram consisting of a predetermined number of bins, calculates the average pixel value for each of the predetermined number of bins, and sets the pixel consisting of this average value as the predetermined number of pixels to be extracted.

[0029] Following step Sa2, the light source spectrum estimation unit 12 inputs the pixel values ​​of the predetermined number of pixels extracted in step Sa2 to the trained model M (step Sa3). The light source spectrum estimation unit 12 inputs the pixel values ​​of the predetermined number of pixels to the trained model M in a predetermined order based on the pixel values ​​of the predetermined number of pixels. For example, the light source spectrum estimation unit 12 arranges the predetermined number of pixels in ascending order of the norm or total value of the pixel values, and inputs the pixel values ​​according to the order. The predetermined order may be ascending or descending, but the same order as that used when generating the trained model M is used.

[0030] Next, the light source spectrum estimating unit 12 outputs the light source spectrum S estimated by the trained model M (step Sa4).

[0031] 3 to 5 show an example of an image G1 in the case where MeanShiftedGrayPixel is used to estimate an achromatic region in the achromatic region extraction unit 11, an example of an image output by converting the Grayness value (Grayness Function in the paper) into a grayscale, and an example of an image from which an achromatic region has been extracted. FIG. 3 shows an example of image G1. FIG. 4 shows an image output by the achromatic region extraction unit 11 after calculating a Grayness value for the image in FIG. 3 and converting the Grayness value into a grayscale. In FIG. 4, it is shown that a darker colored portion is more likely to be achromatic. FIG. 5 shows an example of an image from which an achromatic region has been extracted by the achromatic region extraction unit 11 based on the calculation result of the Grayness value in FIG. 4. In the example of FIG. 5, the white portion is the portion extracted as the achromatic region.

[0032] FIG. 6 is an example of the estimated light source spectrum S in this embodiment. In FIG. 6, the horizontal axis represents wavelength, and the vertical axis represents intensity. Note that the estimated light source spectrum S in this embodiment does not estimate the brightness of the light source when the image G1 was captured. Therefore, a process of scaling the estimated light source spectrum S may be performed as post-processing. As a method of scaling, for example, a method of measuring the illuminance of the shooting environment separately using a light meter when the image is captured, and scaling the light source spectrum S using the measured illuminance may be used.

[0033] FIG. 7 is a flowchart for explaining the operation of the trained model generating device 20 in this embodiment. First, the learning unit 21 calculates pixel values ​​from the camera's spectral sensitivity, light source spectrum, and spectral reflectance (step Sb1). The camera's spectral sensitivity, light source spectrum, and spectral reflectance are read out from the camera spectral sensitivity DB 22, light source spectrum DB 23, and achromatic spectral reflectance DB 24, respectively. If the number of pixels (predetermined number) extracted by the achromatic region extraction unit 11 is Np, Np achromatic spectral reflectances are prepared for one combination of light source spectrum and camera spectral sensitivity, and Np sets of pixel values ​​are calculated. Here, one set of pixel values ​​is a pixel value consisting of R, G, and B.

[0034] The spectral sensitivity of the i-th band of the camera is Sr i , Sg i , Sb i and the intensity of the i-th band in the light source spectrum is L i and the value of the i-th band of the j-th achromatic spectral reflectance is Rf i,j Then, the j-th set pixel value R j , G j , B j are calculated using the following formulas (1), (2), and (3), respectively.

[0035]

number

number

number

[0036] Np such sets of pixel values ​​are calculated for multiple combinations of light source spectra and camera spectral sensitivities.

[0037] Next, the learning unit 21 performs machine learning on the relationship between the pixel values ​​of the Np set calculated in step Sb1 and the light source spectrum used when the pixel values ​​of the Np set were calculated (step Sb2). At this time, the order of the pixel values ​​is determined according to the predetermined order of the pixel values ​​to be input to the trained model M in the light source spectrum estimation unit 12.

[0038] Next, the learning unit 21 outputs the trained model obtained by the machine learning in step Sb2 (step Sb3). The trained model is made up of a set of parameter values ​​that determine the trained model. For example, when a neural network is used for the machine learning, the trained model is made up of weight values ​​of the neural network, etc.

[0039] Fig. 8 is an example of achromatic spectral reflectance stored in the achromatic spectral reflectance DB 24 in this embodiment. Fig. 8 shows an example in which the number of pixels (predetermined number) extracted by the achromatic region extraction unit 11, i.e., Np, is 6. In Fig. 8, the horizontal axis represents wavelength, the vertical axis represents reflectance, and each of six graphs L21 to L26 represents achromatic spectral reflectance.

[0040] The achromatic spectral reflectance in FIG. 8 is a measurement result obtained by measuring objects of various achromatic luminances, but a constant reflectance may be used regardless of wavelength. FIG. 9 is another example of the achromatic spectral reflectance stored in the achromatic spectral reflectance DB 24 in this embodiment. FIG. 9 is an example in which the number of pixels (predetermined number) extracted by the achromatic region extraction unit 11, that is, Np, is 10. In FIG. 9, the horizontal axis is wavelength and the vertical axis is reflectance. In the example of FIG. 9, each of the graphs L31 to L40 of the achromatic spectral reflectance is constant over all wavelengths. The reflectance of graph L31 is 1.0, the reflectance of graph L32 is 0.9, the reflectance of graph L33 is 0.8, and so on, with the reflectance being changed in increments of 0.1.

[0041] In addition, in the above-described embodiment, the pixel values ​​are R (red), G (green), and B (blue), but they may also be C (cyan), M (magenta), and Y (yellow), or Y (luminance), U (blue chrominance), V (red chrominance), or others.

[0042] Although the learning unit 21 calculates the pixel values ​​used in the machine learning from the light source spectrum, the achromatic spectral reflectance, and the spectral sensitivity of the camera, a measured value may be used as the pixel value. In that case, the trained model generating device 20 does not need to include the camera spectral sensitivity DB 22 and the achromatic spectral reflectance DB 24.

[0043] The learning unit 21 may also use the light source spectrum stored in the light source spectrum DB 23 with random noise or known noise as a perturbation for machine learning. The learning unit 21 may also use the light source spectrum stored in the light source spectrum DB 23 with a perturbation based on a priori knowledge of a given light source for machine learning to generate a new light source spectrum. The learning unit 21 may use the light source spectrum obtained by linearly adding the principal component vectors of various light sources based on the principal component analysis while giving random coefficients, for example, or may use the light source spectrum obtained by perturbing the spectrum based on the emission characteristics of the light emitters of each light source or the excitation characteristics of the phosphors, and give the spectrum fluctuation based on the emission characteristics of the light emitters of each light source as a perturbation and give the spectrum to the learning as a new light source spectrum.

[0044] As described above, the light source spectrum estimation system 100 includes an achromatic region extraction unit 11 that estimates an achromatic region of a subject included in an image and extracts pixels of the achromatic region, and a light source spectrum estimation unit 12 that inputs pixel values ​​of the pixels of the achromatic region into a trained model to estimate the spectrum of the light source in the image. This makes it possible to estimate the light source spectrum of any captured image without using an image of a specific color such as a patch.

[0045] The present invention may be embodied as follows. (1) One embodiment is a light source spectrum estimation system including an achromatic region extraction unit that estimates an achromatic region of a subject included in an image and extracts pixels of the achromatic region, and a light source spectrum estimation unit that inputs pixel values ​​of the pixels of the achromatic region into a trained model to estimate the spectrum of a light source in the image.

[0046] (2) In another embodiment, the light source spectrum estimation system according to (1) includes a learning unit that generates the trained model by machine learning the relationship between the light source spectrum and pixel values ​​obtained by multiplying the spectral sensitivity of a camera and the light source spectrum by the spectral reflectance of an achromatic color.

[0047] (3) In another embodiment, the light source spectrum estimation system is described in (2), in which the light source spectrum includes a spectrum obtained by adding perturbation to a predetermined light source spectrum.

[0048] (4) Also, another embodiment is the light source spectrum estimation system described in (1), in which the achromatic region extraction unit extracts a predetermined number of pixels from the achromatic region, and the light source spectrum estimation unit inputs pixel values ​​of the predetermined number of pixels to the trained model in a predetermined order based on the pixel values ​​of the predetermined number of pixels, to estimate the spectrum of the light source in the image.

[0049] (5) Also, another embodiment is the light source spectrum estimation system described in (4), in which the achromatic region extraction unit extracts the predetermined number of pixels by arranging pixels of the achromatic region in an order based on pixel values ​​and extracting pixels at predetermined intervals from the order based on pixel values.

[0050] (6) Also, another embodiment is the light source spectrum estimation system described in (4), wherein the achromatic region extraction unit clusters pixels of the achromatic region into the predetermined number of clusters based on pixel values, and sets pixels consisting of representative values ​​of each of the clusters to the predetermined number of pixels.

[0051] (7) Also, another embodiment is the light source spectrum estimation system described in (4), in which the achromatic region extraction unit creates a histogram of the pixels of the achromatic region consisting of the predetermined number of bins, calculates an average value of pixel values ​​for each of the predetermined number of bins, and sets the pixel consisting of the average value to the predetermined number of pixels.

[0052] (8) Another embodiment is a method for estimating a light source spectrum, comprising the steps of estimating an achromatic region of a subject included in an image and extracting pixels of the achromatic region, and inputting pixel values ​​of the pixels of the achromatic region into a trained model to estimate the spectrum of a light source in the image.

[0053] (9) Yet another embodiment is a program for causing a computer to function as an achromatic region extraction unit that estimates achromatic regions of a subject included in an image and extracts pixels of the achromatic region, and as a light source spectrum estimation unit that inputs pixel values ​​of the pixels of the achromatic region into a trained model to estimate the spectrum of a light source in the image.

[0054] 1 may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read into a computer system and executed to realize the light source spectrum estimation device 10 and the trained model generation device 20. Note that the term "computer system" as used herein includes hardware such as an OS and peripheral devices.

[0055] In addition, if a WWW system is used, the "computer system" also includes the homepage providing environment (or display environment). In addition, "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs, and CD-ROMs, and storage devices such as hard disks built into computer systems. Furthermore, "computer-readable recording medium" also includes those that dynamically hold a program for a short period of time, such as a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line, and those that hold a program for a certain period of time, such as volatile memory inside a computer system that serves as a server or client in such cases. Furthermore, the above program may be one that realizes part of the above-mentioned functions, or may be one that can realize the above-mentioned functions in combination with a program already recorded in the computer system.

[0056] Although an embodiment of the present invention has been described in detail above with reference to the drawings, the specific configuration is not limited to this embodiment, and design changes and the like that do not deviate from the gist of the present invention are also included. [Explanation of symbols]

[0057] 10. Light source spectrum estimation device 11 Achromatic color area extraction part 12 Light source spectrum estimation unit 20 Trained model generator 21 Learning Department 22 Camera Spectral Sensitivity DB 23 Light Source Spectrum DB 24 Achromatic spectral reflectance DB

Claims

1. an achromatic region extraction unit that estimates an achromatic region of a subject included in an image and extracts pixels of the achromatic region; a light source spectrum estimation unit that inputs pixel values ​​of the pixels in the achromatic region into a trained model to estimate a spectrum of a light source in the image; A light source spectrum estimation system comprising:

2. 2. The light source spectrum estimation system according to claim 1, further comprising a learning unit that generates the trained model by machine learning a relationship between a pixel value obtained by multiplying the spectral sensitivity of a camera and the light source spectrum by a spectral reflectance of an achromatic color and the light source spectrum.

3. The light source spectrum estimation system of claim 2 , wherein the light source spectrum includes a spectrum obtained by perturbing a predetermined light source spectrum.

4. The achromatic region extraction unit extracts a predetermined number of pixels from the achromatic region, The light source spectrum estimation unit inputs pixel values ​​of the predetermined number of pixels to the trained model in a predetermined order based on the pixel values ​​of the predetermined number of pixels to estimate a spectrum of a light source in the image. The system of claim 1 .

5. 5. The light source spectrum estimation system according to claim 4, wherein the achromatic region extraction unit extracts the predetermined number of pixels by arranging pixels of the achromatic region in an order based on pixel values ​​and extracting pixels at predetermined intervals from the order based on pixel values.

6. 5. The light source spectrum estimation system according to claim 4, wherein the achromatic region extraction unit clusters pixels of the achromatic region into the predetermined number of clusters based on pixel values, and sets pixels consisting of representative values ​​of each of the clusters as the predetermined number of pixels.

7. 5. The light source spectrum estimation system according to claim 4, wherein the achromatic region extraction unit creates a histogram of pixels in the achromatic region consisting of the predetermined number of bins, calculates an average value of pixel values ​​for each of the predetermined number of bins, and sets pixels consisting of the average value as the predetermined number of pixels.

8. A step of estimating an achromatic region of an object included in an image and extracting pixels of the achromatic region; inputting pixel values ​​of the pixels in the achromatic region into a trained model to estimate the spectrum of an illuminant in the image; The light source spectrum estimation method according to claim 1,

9. Computer, an achromatic region extraction unit that estimates an achromatic region of a subject included in an image and extracts pixels of the achromatic region; a light source spectrum estimation unit that inputs pixel values ​​of the pixels in the achromatic region into a trained model to estimate a spectrum of a light source in the image; A program to function as a

Images