Spectral sensitivity estimation method and system based on response value prediction

Abstract

The invention discloses a spectral sensitivity estimation method and system based on response value prediction, and the method comprises the steps: collecting an existing public spectral sensitivity database, measuring the spectral data of a training sample set and the spectral power distribution of an ambient light source, carrying out the shooting, obtaining the original format response value of a training sample, and carrying out the normalization processing through channels; then, principal component analysis (PCA) is used for calculating principal components of a spectral sensitivity database in different channels, a pseudo-inverse algorithm is used for calculating initial spectral sensitivity, an objective function and constraint conditions of an interior point method are defined, finally, the initial spectral sensitivity is substituted into the interior point method for iterative optimization, and the interior point method is iterated to convergence to complete spectral sensitivity estimation. According to the method, the color difference of the synthetic response value and the spectral sensitivity mu-factor are taken as optimization targets, so that the color difference of the synthetic response value of the spectral sensitivity is effectively reduced.

Description

Technical Field

[0001] The invention belongs to the technical field of computer digital image processing, and in particular relates to a spectral sensitivity estimation method and system based on response value prediction. Background Art

[0002] The spectral sensitivity estimation of digital cameras has a wide range of applications in solving related fields such as computer vision and color science, such as multispectral imaging, illumination estimation, color correction, and digital camera color characterization. Since consumer-grade cameras are not designed for high-precision visual tasks, manufacturers usually do not disclose their spectral sensitivity data. The methods for obtaining the spectral sensitivity of cameras are generally divided into two categories: direct measurement and indirect estimation. The direct measurement method uses a digital camera to capture monochromatic light images of different wavelengths generated by a monochromator, and then calculates its spectral sensitivity based on the camera's response at different wavelengths. Although this method can obtain relatively accurate results, its process is complicated and cumbersome, and requires expensive equipment and a specific measurement environment.

[0003] Therefore, researchers have proposed to use indirect methods to estimate the camera sensitivity function. The main difficulties in estimating camera spectral sensitivity include: the finite dimension of spectral reflectance leads to the ill-posedness of the problem; the influence of noise on the algorithm during camera shooting; the limitation of the camera spectral sensitivity function curve characteristics on the estimation algorithm; the estimated spectral sensitivity curve has large color difference in practical applications, etc. The main idea of ​​most methods is to use the camera to shoot the color card to obtain its corresponding response value, and then reversely calculate the camera's spectral sensitivity function based on the imaging model and pseudo-inverse algorithm. Most of the existing camera spectral sensitivity estimation methods pursue the similarity between the solution curve and the real curve, but ignore the main role of spectral sensitivity in practical applications, resulting in a large color difference between the synthesized response value and the response value obtained by the actual camera shooting. Summary of the invention

[0004] The purpose of the present invention is to solve the problem described in the background technology and to propose a spectral sensitivity estimation method based on response value prediction.

[0005] In view of the problems existing in the above-mentioned existing research, the present invention proposes a method to solve the problem. First, the existing public spectral sensitivity database is collected, the spectral data of the training sample set and the spectral power distribution of the ambient light source are measured, the original format response values ​​of the training samples are photographed to obtain the response values ​​and normalized by channel, and then principal component analysis (PCA) is used to calculate the principal components of the spectral sensitivity database by channel, the initial spectral sensitivity is calculated using the pseudo-inverse algorithm, the objective function and constraints of the interior point method are defined, and finally the initial spectral sensitivity is brought into the interior point method for iterative optimization. The interior point method is iterated until convergence to complete the spectral sensitivity estimation.

[0006] The technical solution of the present invention is a spectral sensitivity estimation method based on response value prediction, which specifically includes the following steps:

[0007] Step 1: collect the public spectral sensitivity database S, which contains the spectral responses of three channels of multiple cameras, and perform normalization processing on each channel;

[0008] Step 2: Use a digital camera to obtain the original format response value of the training sample, and perform channel-by-channel normalization processing on it to obtain the normalized response value X' of the original format response value of the training sample;

[0009] Step 3, using a spectrophotometer and an illuminometer to measure the spectral data R' of the training sample set and the spectral power distribution L' of the ambient light source respectively;

[0010] Step 4: Use principal component analysis (PCA) to calculate the eigenvector Q of each channel of the spectral sensitivity database S by channel j ;

[0011] Step 5: Combine the spectral data R' of the training sample set, the normalized response value X' and the spectral power distribution L' of the ambient light source, and the characteristic vector Q of each channel obtained by principal component analysis. j , use the pseudo-inverse algorithm to calculate the initial spectral sensitivity S initial ;

[0012] Step 6, defining an optimization function with imaging color quality as the goal and non-negative constraints to ensure the inherent properties of spectral sensitivity;

[0013] Step 7: Set the initial spectral sensitivity S initial Use interior point method for iterative optimization;

[0014] Step 8: The interior point method is iterated until convergence to complete the spectral sensitivity estimation.

[0015] Furthermore, in step 1 and step 2, the public spectral sensitivity database and the original format response value are normalized by channel, and the method is shown in equation (1) and equation (2):

[0016]

[0017] In the formula, S ij represents the spectral response value of the j channel of the ith camera in the spectral sensitivity database, min(S ij ) represents the minimum spectral response value in the j channel of the i-th camera, max(S ij ) represents the maximum spectral response value in the j channel of the i-th camera, S i represents the normalized spectral sensitivity of the i-th camera.

[0018]

[0019] In the formula, X j represents the j-channel response value of all training samples, min(X j ) represents the minimum response value among the j-channel response values ​​of all training samples, max(X j ) represents the maximum response value among the j-channel response values ​​of all training samples, and X' represents the response value of all training samples after three-channel normalization.

[0020] Furthermore, in step 4, principal component analysis (PCA) is used to extract the principal components of the spectral sensitivity database by channel, and the method is shown in formula (3):

[0021] S j =Q j d j ,j∈{R,G,B}.(3)

[0022] In the formula, S j is the j-channel spectral response of all cameras in the public spectral sensitivity database, Q j is the j-channel feature vector obtained by PCA, d j is the weight coefficient corresponding to channel j.

[0023] Furthermore, the initial spectral sensitivity S is calculated using the pseudo-inverse algorithm in step 5. initial , and the method is shown in formula (4):

[0024] S initial =Q[(MQ) T (MQ)] -1 (MQ) T X'.(4)

[0025] Where M is the product of the spectral reflectance R' of the training sample and the relative spectral power distribution L' of the light source, Q is the eigenvector obtained by PCA, the superscript 'T' is the transposition symbol, the superscript '-1' indicates the inverse operation, X' is the response value of all training samples after normalization, S initial is the initial spectral sensitivity obtained by the initial estimation of the PCA method.

[0026] Furthermore, in step 6, the objective function and constraint conditions of the interior point method are defined. The objective function mainly includes two parts, where the μ-factor represents the imaging color quality of the camera, and the DE color difference represents the color difference between the current spectral sensitivity synthetic response value and the actual shooting response value. Both parts can effectively guarantee the color quality of the spectral sensitivity synthetic response value. The spectral sensitivity has the inherent property of non-negative, so a non-negative constraint is added. The corresponding methods are shown in equations (5) and (6):

[0027]

[0028] f(s)=μ A (S)+DE (6)

[0029] In the formula, represents finding a value of variable s so that the objective function f(s) reaches the minimum value, S is the spectral sensitivity to be optimized, S ≥ 0 represents that the constraint in the optimization process is a non-negative constraint, and f(S) is the optimized objective function, which is specifically shown in formula (6), where μ A (S) is the μ-factor of the current spectral sensitivity to be optimized, and the specific calculation method is shown in formula (7): where the superscript ‘T’ is the transposition symbol, the superscript ‘-1’ indicates the inverse operation, trace represents the sum of the diagonal of the returned matrix, and A is the CIE1931 standard observer function.

[0030]

[0031] DE is the predicted response value X pre The color difference between the normalized response value X' and the normalized response value X'. The predicted response value is realized by equation (8), where S is the spectral sensitivity to be optimized, and M is the product of the spectral reflectance R' of the training sample and the relative spectral power distribution L' of the light source.

[0032] X pre =MS.(8)

[0033] The color difference calculation method is shown in formula 9, where L truth , a truth , b truth , L pre , a pre , b pre They represent the normalized response value and the chromaticity value of the predicted response value in the Lab space respectively.

[0034]

[0035] To calculate the color difference, the response value needs to be converted from the RGB color space to the XYZ color space and then to the Lab color space. The calculation method is shown in equations (10) and (11).

[0036]

[0037]

[0038] Among them, L, a, and b are the brightness, red, green, and yellow and blue color values ​​of the sample in the CIELab color space; X, Y, and Z are the tristimulus color data of the sample; X n , Y n , Zn are the tristimulus color data of the reference light source; H, H n are the CIEXYZ tristimulus values ​​of the sample and reference light sources, respectively.

[0039] Furthermore, in step 7, the initial spectral sensitivity S initial The interior point method is used for iterative optimization, and the method is shown in formula (12):

[0040]

[0041] Furthermore, in step 8, the initial spectral sensitivity S initial The interior point method is used for iterative optimization, and the spectral sensitivity estimation is completed until convergence to obtain the final spectral sensitivity.

[0042] The present invention also provides a spectral sensitivity estimation system based on response value prediction, comprising:

[0043] one or more processors;

[0044] A storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement a spectral sensitivity estimation method based on response value prediction as described in the above scheme.

[0045] Aiming at the problem of large chromatic aberration of synthetic response values ​​in existing spectral sensitivity estimation methods, the present invention is based on the principal component analysis (PCA) algorithm. Firstly, a public spectral sensitivity database is collected and principal component analysis is performed to extract the corresponding principal component vectors. The initial spectral sensitivity function is calculated using the pseudo-inverse algorithm. Then, the spectral sensitivity is solved by secondary optimization using the interior point method. It is proposed to take the chromatic aberration of synthetic response values ​​and the spectral sensitivity μ-factor as optimization targets, which effectively reduces the chromatic aberration of synthetic response values ​​of spectral sensitivity. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flow chart of an embodiment of the present invention.

[0047] Figure 2 This is a Canon camera simulation experiment estimation curve diagram in an embodiment of the present invention.

[0048] Figure 3 This is a Nikon camera simulation experiment estimation curve diagram in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] When the technical solution of the present invention is specifically implemented, it can be operated by those skilled in the art using computer software technology.

[0050] like Figure 1As shown, the present invention provides a spectral sensitivity estimation method based on response value prediction, which specifically includes the following steps:

[0051] Step 1: collect the public spectral sensitivity database S, which contains the spectral responses of three channels of multiple cameras, and perform normalization processing on each channel;

[0052] The spectral sensitivity channel normalization method in step 1 is shown in formula (1);

[0053]

[0054] In the formula, S ij represents the spectral response value of the j channel of the ith camera in the spectral sensitivity database, min(S ij ) represents the minimum spectral response value in the j channel of the i-th camera, max(S ij ) represents the maximum spectral response value in the j channel of the i-th camera, S i represents the normalized spectral sensitivity of the i-th camera.

[0055] The public database includes the following two spectral sensitivity databases:

[0056] (1) https: / / www.image-engineering.de / library / data-and-tools .

[0057] (2)Tominaga S,Nishi S,Ohtera R.Measurement and estimation of spectralsensitivity functions for mobile phone cameras[J].Sensors,2021,21(15):4985.

[0058] Step 2: Use a digital camera to obtain the original format response value of the training sample, and perform channel-by-channel normalization processing on it to obtain the normalized response value X' of the original format response value of the training sample;

[0059] In step 2, a Nikon D7200 digital camera is used to shoot the ColorCheck 140 color chart, and the original format response value X' of the ColorCheck 140 color chart is extracted from the .NEF format image and normalized by channel. The normalization method is shown in formula (2);

[0060]

[0061] In the formula, X j represents the j-channel response value of all training samples, min(X j) represents the minimum response value among the j-channel response values ​​of all training samples, max(X j ) represents the maximum response value among the j-channel response values ​​of all training samples, and X' represents the response value of all training samples after three-channel normalization.

[0062] Step 3, using a spectrophotometer and an illuminometer to measure the spectral data R' of the training sample set and the spectral power distribution L' of the ambient light source respectively;

[0063] Step 4: Use principal component analysis (PCA) to calculate the eigenvector Q of each channel of the spectral sensitivity database S by channel j ;

[0064] In step 4, principal component analysis (PCA) is used to calculate the spectral sensitivity database feature vector by channel, and the method is shown in formula (3):

[0065] S j =Q j d j ,j∈{R,G,B}.(3)

[0066] In the formula, S j is the j-channel spectral response of all cameras in the public spectral sensitivity database, Q j is the j-channel feature vector obtained by PCA, d j is the weight coefficient corresponding to channel j.

[0067] Step 5: Combine the spectral data R' of the training sample set, the normalized response value X' and the spectral power distribution L' of the ambient light source, and the characteristic vector Q of each channel obtained by principal component analysis. j , use the pseudo-inverse algorithm to calculate the initial spectral sensitivity S initial ;

[0068] In step 5, the initial spectral sensitivity S is calculated using the pseudo-inverse algorithm. initial , and the method is shown in formula (4):

[0069] S initial =Q[(MQ) T (MQ)] -1 (MQ) T X'.(4)

[0070] Where M is the product of the spectral reflectance R' of the training sample and the relative spectral power distribution of the light source L', Q is the eigenvector obtained by PCA, the superscript 'T' is the transposition symbol, the superscript '-1' indicates the inverse operation, X' is the response value of the training sample after normalization, S initial is the initial spectral sensitivity obtained by the initial estimation of the PCA method.

[0071] Step 6, defining an optimization function with imaging color quality as the goal and non-negative constraints to ensure the inherent properties of spectral sensitivity;

[0072] In step 6, the objective function and constraint conditions of the interior point method are defined. The objective function mainly includes two parts, where μ-factor represents the imaging color quality of the camera, and DE color difference represents the color difference between the current spectral sensitivity synthetic response value and the actual shooting response value. Both parts can effectively guarantee the color quality of the spectral sensitivity synthetic response value. Spectral sensitivity has the inherent property of non-negative, so non-negative constraints are added. The corresponding methods are shown in equations (5) and (6):

[0073]

[0074] f(s)=μ A (S)+DE (6)

[0075] In the formula, represents finding a value of variable s so that the objective function f(s) reaches the minimum value, S is the spectral sensitivity to be optimized, S ≥ 0 represents that the constraint in the optimization process is a non-negative constraint, and f(S) is the optimized objective function, which is specifically shown in formula (6), where μ A (S) is the μ-factor of the current spectral sensitivity to be optimized, and the specific calculation method is shown in formula (7): where the superscript ‘T’ is the transposition symbol, the superscript ‘-1’ indicates the inverse operation, trace represents the sum of the diagonal of the returned matrix, and A is the CIE1931 standard observer function.

[0076]

[0077] DE is the predicted response value X pre The color difference between the normalized response value X' and the normalized response value X'. The predicted response value is realized by equation (8), where S is the spectral sensitivity to be optimized, and M is the product of the spectral reflectance R' of the training sample and the relative spectral power distribution L' of the light source.

[0078] X pre =MS.(8)

[0079] The color difference calculation method is shown in formula (9), where L truth , a truth , b truth , L pre , a pre , b pre They represent the normalized response value and the chromaticity value of the predicted response value in the Lab space respectively.

[0080]

[0081] To calculate the color difference, the response value needs to be converted from the RGB color space to the XYZ color space and then to the Lab color space. The calculation method is shown in equations (10) and (11).

[0082]

[0083]

[0084] Among them, L, a, and b are the brightness, red, green, and yellow and blue color values ​​of the sample in the CIELab color space; X, Y, and Z are the tristimulus color data of the sample; X n , Y n , Z n are the tristimulus color data of the reference light source; H, H n are the CIEXYZ tristimulus values ​​of the sample and reference light sources, respectively.

[0085] Step 7: Set the initial spectral sensitivity S initial Use interior point method for iterative optimization;

[0086] In step 7, the initial spectral sensitivity Sinitial is brought into the interior point method for iterative optimization, and the method is shown in formula (12):

[0087]

[0088] The optimization uses the fmincon optimization function that comes with matlab2021b. The interior point method is used for optimization by default, and the number of iterations is set to 30.

[0089] Step 8: The interior point method is iterated until convergence to complete the spectral sensitivity estimation.

[0090] In step 8, the initial spectral sensitivity S initial The interior point method is used for iterative optimization, and the spectral sensitivity estimation is completed until convergence to obtain the final spectral sensitivity.

[0091] On the other hand, an embodiment of the present invention further provides a spectral sensitivity estimation system based on response value prediction, comprising:

[0092] one or more processors;

[0093] A storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement a spectral sensitivity estimation method based on response value prediction as described in the above scheme.

[0094] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0095] The following is a specific example to illustrate the practical application effect of the present invention:

[0096] The embodiment adopts the CIED65 standard light source recommended by the International Commission on Illumination and the internationally used ColorCheck140 color chart to form simulation experimental conditions, and conducts simulation tests on the method of the present invention. Based on the camera imaging model described in the Chinese invention patent "Liang Jinxing, Yuan Li, Hu Xinrong, et al. A spectral estimation method based on adaptive weighted linear regression, publication number CN111750992A, publication date 2020-10-09", and using the spectral sensitivities of two cameras disclosed in Image Engineering[1] as the true spectral sensitivities of the test camera, the original format response values ​​of all training samples of the ColorCheck140 color chart are calculated using the above model and data for simulation experiments.

[0097] We used four evaluation indicators to evaluate the quality of the estimated curve, namely, spectral error (SE), Vora value, CIEDE2000 color difference, and root mean square error RMSE. The camera spectral sensitivity obtained by monochromator calibration is taken as the camera's true spectral sensitivity S. The spectral sensitivity estimated by the method is

[0098] The calculation formula of spectral error (SE) is shown in formula (13). By calculating the curve S and The Frobenius norm between them is used to evaluate the similarity between two curves.

[0099]

[0100] The Vora value is also one of the indicators used to measure the similarity between two spectral sensitivities. Unlike the spectral error (SE), Vora can calculate the similarity between two vectors across channels, and then make an overall comparison of the responses of the three channels, which can effectively reflect the overall performance of the estimation algorithm.

[0101]

[0102] The closer the Vora value is to 1, the more similar the estimated curve is to the true spectral sensitivity.

[0103] CIEDE2000 color difference reflects the difference in color perception of the human eye and can be used to evaluate the color quality of the spectral sensitivity composite response value. The calculation uses S and The average color difference ΔE between the predicted response values.

[0104] The root mean square error RMSE is the error between the RGB predicted by the spectral sensitivity and the RGB captured by the camera, C i and They represent the predicted value and the real value captured by the camera respectively, and n is the number of color samples.

[0105]

[0106] In the simulation experiment, we compared the commonly used spectral sensitivity estimation methods, including: classical PCA, weighted PCA, quadratic programming (QP) and T regularization. The experimental results are shown in Table 1. Among them, RMSE rgb Represents the average root mean square error of the three channels, and the estimated curve is as follows Figure 2 and Figure 3 As shown:

[0107] Table 1 Simulation results

[0108]

[0109] For the PCA method, please refer to: Jiang J, Liu D, Gu J, et al. What is the space of spectral sensitivity functions for digital color cameras? [C] / / 2013IEEE Worksho p on Applications of Computer Vision (WACV). IEEE, 2013: 168-179.

[0110] Weight_PCA please see: Fan H,Xu L,Luo M R.Optimized principal compon entanalysis for camera spectral sensitivity estimation[J].JOSA A,2023,40(8):1515-1526.

[0111] For QP methods, please see: Barnard K, Funt B.Camera characterization for color research[J].Color Research&Application:Endorsed by Inter-Society Color Counci l,The Color Group(Great Britain),Canadian Society for Color,Color Scienc eAssociation of Japan,Dutch Society for the Study of Color,The Swedish C olourCentre Foundation,Colour Society of Australia,Centre de la Couleur,2002,27(3):152-163.

[0112] For the T regularization method, please refer to: Dyas B. Robust color sensor response characterization [C] / / Color and Imaging Conference. Society of Imaging Science and Technology, 2000, 8: 144-148.

[0113] The experimental results show that the method of the present invention has more excellent performance in synthesizing response values. Compared with other methods, the method proposed in the present invention has a smaller color difference in the synthesized response value on the basis of ensuring the accuracy of spectral sensitivity curve estimation, and has better effect in practical applications.

[0114] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

Claims

1. A spectral sensitivity estimation method based on response value prediction, characterized in that: The steps include: Step 1: collect the public spectral sensitivity database S, which contains the spectral responses of three channels of multiple cameras, and perform normalization processing on each channel; Step 2: Use a digital camera to obtain the original format response value of the training sample, and perform channel-by-channel normalization processing on it to obtain the normalized response value X' of the original format response value of the training sample; Step 3, respectively measuring the spectral data R' of the training sample set and the spectral power distribution L' of the ambient light source; Step 4: Use principal component analysis to calculate the characteristic vector Q of each channel of the spectral sensitivity database S j , where j represents the jth channel; Step 5: Combine the spectral data R' of the training sample set, the normalized response value X' and the spectral power distribution L' of the ambient light source, and the characteristic vector Q of each channel obtained by principal component analysis. j , use the pseudo-inverse algorithm to calculate the initial spectral sensitivity S initial ; Step 6, defining an optimization function with the goal of optimizing the imaging color quality and a non-negative constraint condition to ensure the inherent properties of spectral sensitivity; Step 7: Set the initial spectral sensitivity S initial Use interior point method for iterative optimization; Step 8: The interior point method is iterated until convergence to complete the spectral sensitivity estimation.

2. A spectral sensitivity estimation method based on response value prediction as claimed in claim 1, characterized in that: In step 1, the public spectral sensitivity database is normalized by channel, and the method is shown in formula (1): In the formula, S ij represents the spectral response value of the j channel of the ith camera in the spectral sensitivity database, min(S ij ) represents the minimum spectral response value in the j channel of the i-th camera, max(S ij ) represents the maximum spectral response value in the j channel of the i-th camera, S i represents the normalized spectral sensitivity of the i-th camera, where R, G, and B represent the red, green, and blue channels, respectively.

3. A spectral sensitivity estimation method based on response value prediction as claimed in claim 1, characterized in that: In step 2, the original format response value is normalized by channel, and the method is shown in formula (2): In the formula, X j represents the j-channel response value of all training samples, min(X j ) represents the minimum response value among the j-channel response values ​​of all training samples, max(X j ) represents the maximum response value among the j-channel response values ​​of all training samples, X' represents the response value of all training samples after three-channel normalization, and R, G, and B represent the red, green, and blue channels respectively.

4. A spectral sensitivity estimation method based on response value prediction as claimed in claim 1, characterized in that: In step 4, principal component analysis (PCA) is used to extract the principal components of the spectral sensitivity database by channel, and the method is shown in formula (3): S j =Q j d j ,j∈{R,G,B} (3) In the formula, S j is the j-channel spectral response of all cameras in the public spectral sensitivity database, Q j is the j-channel feature vector obtained by PCA, d j is the corresponding j channel weight coefficient, R, G, B represent the red, green, and blue channels respectively.

5. The spectral sensitivity estimation method based on response value prediction according to claim 1, characterized in that: In step 5, the initial spectral sensitivity S is calculated using the pseudo-inverse algorithm. initial , and the method is shown in formula (4): S initial =Q[(MQ) T (MQ)] -1 (MQ) T X'(4) In the formula, M is the product of the spectral data R' of the training sample set and the relative spectral power distribution L' of the light source, Q is the eigenvector obtained by PCA, the superscript 'T' is the transposition symbol, the superscript '-1' indicates the inverse operation, X' is the response value of all training samples after normalization, S initial is the initial spectral sensitivity obtained by the initial estimation of the PCA method.

6. A spectral sensitivity estimation method based on response value prediction as claimed in claim 1, characterized in that: The objective function in step 6 contains two parts: the imaging color quality of the camera and the color difference between the current spectral sensitivity synthetic response value and the actual shooting response value. Both parts can effectively guarantee the color quality of the spectral sensitivity synthetic response value. The spectral sensitivity has the inherent property of non-negative, so a non-negative constraint is added; specifically, as shown in equations (5) and (6): f(s) = μ A (S)+DE (6) In the formula, represents finding a value of variable s so that the objective function f(s) reaches the minimum value, S is the spectral sensitivity to be optimized, S ≥ 0 represents that the constraint in the optimization process is a non-negative constraint, and f(S) is the optimized objective function, which is specifically shown in formula (6), where μ A (S) is the μ-factor of the current spectral sensitivity to be optimized, and the specific calculation method is shown in formula (7): where the superscript ‘T’ is the transposition symbol, the superscript ‘-1’ indicates the inverse operation, trace represents the sum of the diagonal of the returned matrix, and A is the CIE1931 standard observer function; DE is the predicted response value X pre The color difference between and the normalized response value X'.

7. A spectral sensitivity estimation method based on response value prediction as claimed in claim 6, characterized in that: The predicted response value is realized by formula (8), where S is the spectral sensitivity to be optimized, and M is the product of the spectral data R' of the training sample set and the relative spectral power distribution L' of the light source; X pre =MS(8) The color difference calculation method is shown in formula (9), where L truth , a truth , b truth , L pre , a pre , b pre Respectively represent the normalized response value and the chromaticity value of the predicted response value in the Lab space; Among them, to calculate the color difference, the response value needs to be converted from the RGB color space to the XYZ color space, and then converted to the Lab color space. The calculation method is shown in equations (10) and (11); Among them, L, a, and b are the brightness, red, green, and yellow and blue color values ​​of the sample in the CIELab color space; X, Y, and Z are the tristimulus color data of the sample; X n , Y n , Z n are the tristimulus color data of the reference light source; H, H n are the CIEXYZ tristimulus values ​​of the sample and reference light sources, respectively.

8. A spectral sensitivity estimation method based on response value prediction as claimed in claim 6, characterized in that: In step 7, the initial spectral sensitivity S initial The interior point method is used for iterative optimization, and the method is shown in formula (12):

9. The spectral sensitivity estimation method based on response value prediction according to claim 1, characterized in that: In step 3, a spectrophotometer and an illuminance meter are used to measure the spectral data R' of the training sample set and the spectral power distribution L' of the ambient light source respectively.

10. A spectral sensitivity estimation system based on response value prediction, characterized in that: include: one or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, enables the one or more processors to implement a spectral sensitivity estimation method based on response value prediction as described in any one of claims 1 to 9.

Images