A method for reconstructing spectrum of a filter based on human eye vision

Abstract

The application discloses a kind of based on human eye visual system screening optical filter's spectral reconstruction method and application, steps are as follows: using LMS cone response function to the original filter is weighted, using weighted area screening method, select and the LMS cone response function of human eye visual system most matched three filter as initial filter;Determine initial filter, and the remaining filter is sequentially exhaustively combined with three initial filters;The filter group generated is substituted into spectral reconstruction model, and the spectral recovery and chroma error of each filter group are obtained;According to self-defined recovery error set sorting, select the optimal filter group under L weighting, M weighting and S weighting, select the optimal group through self-defined error score ranking.When the number of channels increases, the respective optimal filter group under L weighting, M weighting and S weighting is exhaustively combined with the remaining filter, and the above steps are repeated according to the number of channels until the number of filters in the selected filter group meets the number of channels.

Description

Technical Field

[0001] This invention belongs to the field of spectral reconstruction technology, specifically relating to a spectral reconstruction method based on human visual screening of filters. Background Technology

[0002] Over the past few decades, multispectral imaging technology has garnered widespread attention because it solves the metamerism problem inherent in traditional three-color digital cameras, enabling the accurate recording of spectral information from object surfaces. This technology is increasingly being applied in fields such as museums, art galleries, and computer graphics. One of the most crucial components of a multispectral acquisition system is the set of filters that allows acquisition across different bands of the visible light spectrum. Selecting a specific set of filters from a given set significantly impacts the accuracy of spectral reconstruction. While using more filters generally improves the accuracy of spectral recovery, it also correspondingly increases operational complexity, image acquisition time, and data volume. Therefore, researchers have conducted extensive studies on how to optimize filter selection.

[0003] Filter group optimization has been studied to some extent in the field of remote sensing, but many problems remain to be solved. Some scholars have theoretically designed filters with optimal spectral transmittance based on specific optimization criteria. However, the design process is very complex due to the combined effects of the optical path, light source, and sensor spectral characteristics. Furthermore, actual filters cannot guarantee that the theoretically designed optimal filter will have the exact same spectral transmittance. Another option is to select the best filter from the existing filters. When the total number of filters is small, exhaustive search is practical. However, as the total number of filters to be selected increases, the computational complexity of exhaustive search increases dramatically, making it unsuitable.

[0004] With extensive research, people no longer simply rely on empirical methods to select filter compositions. Filter Vector Analysis (FVAM) is a commonly used filter selection method. Hardeberg first used the Maximum Linear Independence (MLI) method to select spectral sample sets, and Li subsequently applied it to filter selection. The selection principle of the MLI method is: the spectral transmittance matrix of the selected filter group has the minimum condition number. The MaxOrthogonal Spectral Transmittance Vector Maximization method selects the filter with the largest spectral transmittance vector norm as the preferred filter, then uses the filters to form a spectral transmittance space, and selects the filter group with the largest orthogonality in the spectral transmittance space. The Linear Distance Maximization Method (LDMM) uses the linear distance between the spectral transmittance vectors of the filters as the sole criterion for selecting a filter group. FVAM directly selects filters to form a filter group based on the mathematical properties of the filter transmittance curves, which is simple, time-saving, and the stability of the selection results is better than empirical methods.

[0005] The above method does not consider other parameters in the multispectral imaging system, such as the spectral distribution (SPD) of the light source, the spectral sensitivity of the camera, and the characteristics of the imaging scene. This means that although the filter selected by the filter vector analysis method can ensure the effectiveness of the first channel response of the multispectral camera, it is difficult to meet the optimization requirements of the entire system as the number of channels increases. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, the present invention provides a spectral reconstruction method with high reconstruction accuracy, simple calculation, best matching with the human visual system, and filter selection that takes into account all factors of the multispectral imaging system as a whole.

[0007] This invention is achieved through the following technical solution:

[0008] A spectral reconstruction method based on human visual screening of filters includes the following steps:

[0009] Step 1, Determine the initial filters: Use the LMS cone response function to weight the original filters, calculate the area reduction rate of the filters before and after weighting, and select the initial filters based on the area reduction rate. Select the three filters that best match the LMS cone response function of the human visual system as the initial filters. These filters are determined based on the shape and mathematical characteristics of the spectral transmittance curve of the weighted filters.

[0010] Step 2, exhaustive combination of filters: After determining the initial filters, exhaustively combine the remaining filters with the three initial filters one by one to form a multi-channel filter group.

[0011] Step 3, Spectral Reconstruction: Substitute each filter group into the multispectral imaging system to perform spectral reconstruction, obtaining the spectral recovery error and chromaticity error of that filter group. Root mean square error (RMSE), goodness of fit (GFC), peak signal-to-noise ratio (PSNR), spectral angle (SAM), and chromatic difference (∆E) are selected as reference indicators.

[0012] Step 4, Custom Error Set Sorting: Normalize the calculated custom error reference index and then select the optimal filter group under L-weighted, M-weighted and S-weighted methods according to the set sorting. Select the optimal filter group from the three optimal filter groups through the custom error set ranking.

[0013] As the number of channels increases, the optimal filter groups under L-weighted, M-weighted, and S-weighted methods are combined with the remaining filters in an exhaustive manner, and the above steps are repeated according to the number of channels until the number of filters in the selected filter group meets the number of channels.

[0014] Preferably, the specific steps for determining the initial filter in step 1 are as follows:

[0015] (1) The original filter is weighted using the LMS cone response function. The weighting process is shown in formula (1):

[0016] (1)

[0017] In the formula b The original transmittance function of the filter; i Indicates the first i One filter; V Represent the response functions of the three cones, L, M, and S; n The first element in the LMS cone response function represents the... n The response function of a cone. B This represents the filter spectral transmittance function weighted by the LMS cone response function;

[0018] (2) Weight the filters based on the band distance between the peak value of the filter and the peak value of the LMS curve. First, determine the band corresponding to the peak of each of the three cone response functions (L, M, S), and then determine the band corresponding to the peak of each filter. Calculate the distance between the band corresponding to the peak of each filter and the band corresponding to the peak of the three cone response functions, and weight the filters based on this distance. The process is shown in formulas (2) and (3).

[0019] (2)

[0020] (3)

[0021] In the formula, b maxi Indicates the peak band of the filter response value; Vmax Represents the peak band of the LMS cone response function; c n This indicates the band distance between the filter peak and the LMS curve peak. Cn This represents the filter transmittance function weighted by wavelength range. Multiplying the two weighted filter transmittance functions yields the final weighted filter spectral transmittance function. Dn , as in equation (4);

[0022] (4)

[0023] (3) Using the spectral band as the horizontal axis and the filter spectral transmittance as the vertical axis, calculate the area before and after weighting the filter spectral transmittance curves. Then subtract the original area from the weighted area to obtain the area difference. Divide the area difference by the original area; the filter with the smallest area reduction rate is the preferred filter we need. The selection process is as described in formula (5);

[0024] (5)

[0025] S Represents the area of ​​the unweighted filter; S w The area represents the weighted filter. Using formula (5), the three preferred filters that best match the L, M, and S cone cell response functions of the human eye are obtained.

[0026] Preferably, the specific steps for exhaustively combining filters in step 2 are as follows:

[0027] Will b 1 , b 2 and b 3 These are defined as the three preferred filters that best match the human eye after weighting. The remaining filters are then exhaustively combined with these three filters, as shown in formula (6).

[0028]

[0029] (6)

[0030]

[0031] In the formula K 1 , K 2 and K 3 They represent respectively by b 1 , b 2 and b 3 The remaining filter group consists of [the following].

[0032] Preferably, the specific steps of spectral reconstruction in step 3 are as follows:

[0033] (1) During the spectral reconstruction process after obtaining the filter set, the similarity between the training samples and the test samples will also affect the final recovery accuracy. Therefore, the Euclidean distance between the response values ​​of the training samples and the test samples is used as the weighting function to optimize the spectral reconstruction process, as shown in formula (7);

[0034] (7)

[0035] In the formula, It is the response value of the test sample; It is the response value of the best local training sample; subscript j It is the training sample of the firstj One sample; s j Representing the j The Euclidean distance between training samples and test samples.

[0036] (2) Arrange the training samples and test samples in ascending order of distance, select the first N (1≤N≤j) training samples as local optimal training samples, and calculate the inverse distance weighted (IDW) coefficient for each selected local optimal training sample, as shown in formula (8).

[0037] (8)

[0038] subscript in formula k It is the first of the locally optimal training samples k The sample; is the first sample. k The Euclidean distance between a local optimum training sample and a test sample; e It is a very small additional value to avoid division by zero, using e =0.001. Weighted matrix. W Defined as formula (9).

[0039] (9)

[0040] (3) After optimizing the sample, the pseudo-inverse method is used for spectral reconstruction.

[0041] (10)

[0042] (11)

[0043] The superscript "-1" indicates a matrix inverse; R Train It is the spectral reflectance of the selected local training sample; P Train It is the normalized response value of the training samples; P Test This represents the normalized response value of the test sample; R It is the reconstructed spectral reflectance.

[0044] Preferably, the specific steps for custom error set sorting in step 4 are as follows:

[0045] (1) Based on the reconstructed spectral reflectance, the spectral reconstruction error and chromaticity error are calculated. We select root mean square error (RMSE), goodness of fit (GFC), peak signal-to-noise ratio (PSNR), spectral angle (SAM), and color difference (∆E) as custom error calculation indicators. The calculation of the above custom error calculation indicators is shown in formulas (12)-(16).

[0046] (12)

[0047] (13)

[0048] (14)

[0049] (15)

[0050] (16)

[0051] (2) Using the above indicators, perform custom error set score calculation. Normalize the values ​​of each indicator and then multiply them to get our final custom error set score. The calculation process is shown in formula (17).

[0052] (17)

[0053] in TOTAL ni It corresponds to the first n Custom recovery error set score of the i-th filter group consisting of a set of preferred filters;

[0054] (3) Using formula (18), select the optimal filter group under L, M and S weighting respectively, and then compare the custom error set scores of the three filter groups to select the optimal filter group.

[0055] (18)

[0056] G n This represents the optimal filter set for the current number of channels;

[0057] (4) When the number of channels in the multispectral imaging system increases, the number of filters in the filter group should also increase accordingly. At this time, the three optimal filter groups selected in (3) under L, M, and S weighting are used as the initial filter groups and the remaining filters for exhaustive combination. The operation of formula (6)-(18) is repeated until a filter group that meets the channel number requirements of the multispectral imaging system is selected.

[0058] Application of a spectral reconstruction method based on human visual screening filters in the evaluation of printed matter quality.

[0059] A method for evaluating the quality of printed materials, comprising the following steps:

[0060] (a) Obtaining printed materials: Randomly sample the printed materials obtained from the printing press to obtain test printed materials, and then place the test printed materials on the printing inspection table;

[0061] (b) Calculate the spectral reflectance R of the printed matter to be detected by using a spectral reconstruction method based on human visual screening of filters as described in the foregoing text, and obtain the spectral reflectance R;

[0062] (c) Compare the spectral reflectance R of the printed matter to be detected with the spectral reflectance of the printed original, and calculate the root mean square error value RMSE between the two. The calculation formula is as follows:

[0063]

[0064] In the formula, r is the spectral reflectance of the printed original, and n is the wavelength dimension;

[0065] (d) When the root mean square error value is within 0.05, it is a qualified product; when the root mean square error value is outside 0.05, it is an unqualified product. The beneficial technical effects of the present invention:

[0066] According to the spectral reconstruction method involved in the present invention, the original filter is weighted by using the LMS cone response function, and the weighted area screening method is used to select three filters that are most matched with the LMS cone response function of the human visual system as the initial filters; after determining the initial filters, the remaining filters are respectively and exhaustively combined with the three initial filters one by one; then the filter sets generated by the exhaustive combination are substituted into the spectral reconstruction model to obtain the spectral recovery error and chromaticity error of each set of filters; then, according to the custom recovery error set score ranking, the optimal filter sets under L weighting, M weighting and S weighting are selected, and the optimal filter set is selected from the three optimal filter sets through the custom error score ranking. Therefore, the present invention provides a spectral reconstruction algorithm with simple calculation, high spectral reconstruction accuracy, and relatively convenient for users to use. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 is a flowchart of a spectral reconstruction method based on human visual screening of filters according to the present invention:

[0068] Figure 2 is a schematic diagram of the placement positions of the hyperspectral camera, the light source and the color sample during shooting; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0069] In order to make the technical means, creative features, achieved purposes and effects of the present invention easy to understand, the following embodiments are specifically described with reference to the accompanying drawings for the present invention.

[0070] As Figure 1 shown, a spectral reconstruction method based on human visual screening of filters provided by the present invention includes the following steps:

[0071] Step 1, Determine the initial filters: Use the LMS cone response function to weight the original filters, calculate the area reduction rate of the filters before and after weighting, and select the initial filters based on the area reduction rate. Select the three filters that best match the LMS cone response function of the human visual system as the initial filters. These filters are determined based on the shape and mathematical characteristics of the spectral transmittance curve of the weighted filters.

[0072] The specific steps for determining the initial filter in step 1 are as follows:

[0073] (1) The original filter is weighted using the LMS cone response function. The weighting process is shown in formula (1):

[0074] (1)

[0075] In the formula b The original transmittance function of the filter; i Indicates the first i One filter; V Represent the response functions of the three cones, L, M, and S; n The first element in the LMS cone response function represents the... n The response function of a cone. B This represents the filter spectral transmittance function weighted by the LMS cone response function.

[0076] (2) Weight the filters based on the band distance between the peak value of the filter and the peak value of the LMS curve. First, determine the band corresponding to the peak of each of the three cone response functions (L, M, S), and then determine the band corresponding to the peak of each filter. Calculate the distance between the band corresponding to the peak of each filter and the band corresponding to the peak of the three cone response functions, and weight the filters based on this distance. The process is shown in formulas (2) and (3).

[0077] (2)

[0078] (3)

[0079] In the formula, b maxi Indicates the peak band of the filter response value; Vmax Represents the peak band of the LMS cone response function; c n This indicates the band distance between the filter peak and the LMS curve peak. Cn This represents the filter transmittance function weighted by wavelength range. Multiplying the two weighted filter transmittance functions yields the final weighted filter spectral transmittance function. Dn , as in equation (4).

[0080] (4)

[0081] (3) Using the spectral band as the horizontal axis and the filter spectral transmittance as the vertical axis, calculate the area before and after weighting the filter spectral transmittance curves. Then subtract the original area from the weighted area to obtain the area difference. Divide the area difference by the original area; the filter with the smallest area reduction rate is the preferred filter we need. The selection process is as described in formula (5);

[0082] (5)

[0083] S Represents the area of ​​the unweighted filter; S w The area represents the weighted filter. Using formula (5), the three preferred filters that best match the L, M, and S cone cell response functions of the human eye are obtained;

[0084] Step 2, exhaustive combination of filters: After determining the initial filters, exhaustively combine the remaining filters with the three initial filters one by one to form a multi-channel filter group.

[0085] The specific operation of exhaustively combining the filters is as follows:

[0086] Will b 1 , b 2 and b 3 These are defined as the three preferred filters that best match the human eye after weighting. The remaining filters are then exhaustively combined with these three filters, as shown in formula (6).

[0087]

[0088] (6)

[0089]

[0090] In the formula K 1 , K 2 and K 3 They represent respectively by b 1 , b 2 and b 3 The remaining filter group consists of [the following].

[0091] Step 3, Spectral Reconstruction: Substitute each filter group into the multispectral imaging system to perform spectral reconstruction, obtaining the spectral recovery error and chromaticity error of that filter group. Root mean square error (RMSE), goodness of fit (GFC), peak signal-to-noise ratio (PSNR), spectral angle (SAM), and chromatic difference (∆E) are selected as reference indicators.

[0092] The specific steps for spectral reconstruction are as follows:

[0093] (1) During the spectral reconstruction process after obtaining the filter set, the similarity between the training samples and the test samples will also affect the final recovery accuracy. Therefore, the Euclidean distance between the response values ​​of the training samples and the test samples is used as the weighting function to optimize the spectral reconstruction process, as shown in formula (7);

[0094] (7)

[0095] In the formula, It is the response value of the test sample; It is the response value of the best local training sample; subscript j It is the training sample of the first j One sample; s j Representing the j The Euclidean distance between training samples and test samples.

[0096] (2) Arrange the training samples and test samples in ascending order of distance, select the first N (1≤N≤j) training samples as local optimal training samples, and calculate the inverse distance weighted (IDW) coefficient for each selected local optimal training sample, as shown in formula (8).

[0097] (8)

[0098] subscript in formula k It is the first of the locally optimal training samples k The sample; is the first sample. k The Euclidean distance between a local optimum training sample and a test sample; e It is a very small additional value to avoid division by zero, using e =0.001. Weighted matrix. W Defined as formula (9).

[0099] (9)

[0100] (3) After optimizing the sample, the pseudo-inverse method is used for spectral reconstruction.

[0101] (10)

[0102] (11)

[0103] The superscript "-1" indicates a matrix inverse; R Train It is the spectral reflectance of the selected local training sample; P Train It is the normalized response value of the training samples; P Test This represents the normalized response value of the test sample; R It is the reconstructed spectral reflectance.

[0104] Step 4, Custom Error Set Sorting: Normalize the calculated custom error reference index and then select the optimal filter group under L-weighted, M-weighted and S-weighted methods according to the set sorting. Select the optimal filter group from the three optimal filter groups through the custom error set ranking.

[0105] The specific steps for the custom error set sorting are as follows:

[0106] (1) Based on the reconstructed spectral reflectance, the spectral reconstruction error and chromaticity error are calculated. We select root mean square error (RMSE), goodness of fit (GFC), peak signal-to-noise ratio (PSNR), spectral angle (SAM), and color difference (∆E) as custom error calculation indicators. The calculation of the above custom error calculation indicators is shown in formulas (12)-(16).

[0107] (12)

[0108] (13)

[0109] (14)

[0110] (15)

[0111] (16)

[0112] (2) Using the above indicators, perform custom error set score calculation. Normalize the values ​​of each indicator and then multiply them to get our final custom error set score. The calculation process is shown in formula (17).

[0113] (17)

[0114] in TOTAL ni It corresponds to the first n Custom recovery error set score of the i-th filter group consisting of a set of preferred filters;

[0115] (3) By using Formula (18), the optimal filter sets under L, M, and S weightings are respectively selected, and then the custom error set scores of the three groups of filter sets are compared to select the optimal group of filter sets.

[0116] (18)

[0117] G n Represents the optimal filter set for the current channel number.

[0118] (4) When the number of channels of the multispectral imaging system increases, the number of filters in the filter set should also increase accordingly. At this time, the three groups of optimal filter sets under L, M, and S weightings selected in (3) are respectively used as the initial filter sets and exhaustively combined with the remaining filters, and the operations of Formulas (6)-(18) are repeated until a filter set that meets the requirements of the number of channels of the multispectral imaging system is selected.

[0119] Application of a spectral reconstruction method for filtering based on human eye vision in the evaluation of printed matter quality.

[0120] A method for evaluating the quality of printed matter, characterized in that the method is implemented by the following steps:

[0121] (a) Obtain the printed matter: Randomly sample the printed matter obtained by the printing press to obtain the test printed matter, and then place the test printed matter on the printing proofing table;

[0122] (b) Use a spectral reconstruction method for filtering based on human eye vision described in the foregoing text to calculate the spectral reflectance R of the printed matter to be detected, and obtain the spectral reflectance R;

[0123] (c) Compare the spectral reflectance R of the printed matter to be detected with the spectral reflectance of the printing original, and calculate the root mean square error value RMSE between the two. The calculation formula is as follows:

[0124]

[0125] In the formula, r is the spectral reflectance of the printing original, and n is the wavelength dimension.

[0126] (d) When the root mean square error value is within 0.05, it is a qualified product; when the root mean square error value is outside 0.05, it is an unqualified product. The technical features of measuring the color information of the printed matter are all known technologies to those skilled in the art.

[0127] Functions and effects of the embodiments

[0128] A real-world experiment was conducted in a darkroom to further validate the performance of the proposed method. In this experiment, the IT8.7 / 3 color chart was selected as the data sample, comprising 928 color patches. The spectral reflectance of the sample groups was measured in intervals from 400 to 700 nm using an X-rite Eye One spectrophotometer. The response value for each color patch was obtained using a Shot 5.0 multispectral camera at ISO 50, an f-number aperture of F5.6, and an exposure time of 1 / 10 s. The true response values ​​were extracted in the sRGB color space. The power distribution of the light sources in the shooting environment was measured using a CS2000 spectroradiometer.

[0129] During the shooting process, the data sample should be placed in the center of the camera's field of view, and also in the center of both light sources, ensuring that the light shines on the sample at a 45° angle. The sample should be placed 1.5 meters away from the camera lens. Figure 2 As shown. Before shooting, allow the light source to warm up for 30 minutes to allow the light intensity to stabilize.

[0130] The experimental results are recorded in Table 1.

[0131]

[0132] In the table, CIE DE1976 and CIE DE2000 represent the color difference results of spectral reconstruction; the larger the value, the worse the colorimetric effect of the spectral reconstruction. RMSE represents the root mean square error between the reconstructed spectral reflectance and the original spectral reflectance; the smaller the value, the better the spectral reconstruction effect. GFC represents the goodness of fit between the reconstructed spectral reflectance and the original spectral reflectance; the larger the value, the more similar the two spectral curves are, and the better the reconstruction effect.

[0133] The results show that the method proposed in this invention outperforms existing methods in both spectral recovery and color difference. At different channel numbers, the maximum and average color difference values ​​of this method are smaller than those of the other three methods, indicating that the color reconstruction of the spectrum by this method is more accurate. The maximum and average RMSE values ​​of this method are smaller than those of the other three methods, while the GFC values ​​are larger, indicating that the spectral reconstruction accuracy of this method is higher.

[0134] The results of the actual experiment demonstrate that the method proposed in this study can be applied to real-world scenarios.

[0135] The spectral reconstruction process based on filter selection using the human visual system is computationally simple, has high spectral reconstruction accuracy, and is relatively convenient for users.

[0136] The above embodiments are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention.

Claims

1. A method of spectral reconstruction for screening filters based on human eye vision, characterized by: Comprising the following steps: Step S1, determining the initial filter: The original filter is weighted using the LMS cone response function, the area reduction rate before and after the filter is weighted is calculated, and the initial filter is selected according to the area reduction rate. The three filters that best match the LMS cone response function of the human visual system are selected as the initial filters. These filters are determined based on the shape and mathematical characteristics of the spectral transmittance curve of the weighted filter. Step S2, filter exhaustive combination: After determining the initial filter, the remaining filters are combined with the three initial filters one by one to form a multi-channel filter group. Step S3, spectral reconstruction: Each filter group is substituted into the multispectral imaging system to perform spectral reconstruction and obtain the spectral recovery error and chrominance error of the filter group. Root mean square error (RMSE), goodness of fit (GFC), peak signal-to-noise ratio (PSNR), spectral angle (SAM), and color difference (ΔE) are selected as reference indicators. Step S4, self-defined error set sorting: The calculated self-defined error reference indicators are normalized and sorted according to the set points. The optimal filter group under L weighting, M weighting, and S weighting is selected, and the optimal filter group is selected from the three optimal filter groups based on the self-defined error set ranking. When the number of channels increases, the respective optimal filter groups under L weighting, M weighting, and S weighting are combined with the remaining filters, and the above steps are repeated according to the number of channels until the number of filters in the selected filter group meets the number of channels.

2. The spectral reconstruction method for screening filters based on human eye vision according to claim 1, wherein: The specific steps for determining the initial filter in step S1 are as follows: S1.1, use the LMS cone response function to weight the original filter; the weighting process is shown in formula (1): （1） wherein b represents the original transmittance function of the filter; i represents the first i filter; V represents the L, M, S cone response functions; n represents the first n cone response function of the LMS cone response functions, B represents the filter spectral transmittance function weighted by the LMS cone response functions; S1.2, weight the filter according to the wavelength distance between the filter peak and the LMS curve peak. First, determine the wavelength corresponding to the peak of each of the L, M, and S cone response functions, and then determine the wavelength corresponding to the peak of each filter. Calculate the distance between the wavelength corresponding to the peak of each filter and the wavelength corresponding to the peak of the three cone response functions, and weight according to this distance. The process is shown in formulas (2) and (3): （2） （3） wherein b maxi peak wavelength band representing the peak wavelength of the filter response; Vmax peak wavelength band representing the peak wavelength of the LMS cone response function; c n wavelength band representing the distance between the peak wavelength of the filter and the peak wavelength of the LMS curve, Cn weighted filter transmittance function represented by the wavelength band distance; multiplying the two weighted filter transmittance functions to obtain the final weighted filter spectral transmittance function Dn as in equation (4); （4） S1.3, take the spectral wavelength as the horizontal axis and the filter spectral transmittance as the vertical axis, and calculate the area of the filter spectral transmittance curve before and after weighting. Then, subtract the original area from the weighted area to obtain the area difference. Divide the area difference by the original area, and the filter with the smallest area reduction rate is the preferred filter. The selection process is shown in formula (5). （5） S Area representing unweighted filter; S w Area representing weighted filter; Through formula (5), the three preferred filters that best match the L, M, and S cone cells of the human eye are obtained.

3. The method of claim 1, wherein: The specific steps for filter exhaustive combination in step S2 are as follows: Let b 1 , b 2 and b 3 be defined as the three most preferred filters that best match the human eye after weighting; the remaining filters are exhaustively combined with these three filters, respectively, as shown in equation (6). （6）； wherein K 1 , K 2 and K 3 respectively represent the remaining filter groups consisting of b 1 , b 2 and b 3 .

4. The method of claim 3, wherein: The specific steps for spectral reconstruction in step S3 are as follows: S3.1、In the process of spectral reconstruction after obtaining the filter set, the similarity between the training samples and the test samples also affects the final recovery accuracy; therefore, the Euclidean distance between the response values of the training samples and the test samples is used as a weighted function to optimize the spectral reconstruction process, as shown in equation (7); （7） wherein p test is a response value of a test sample; p train is a response value of a best local training sample; subscript j is the j th sample of a training sample; S j represents the Euclidean distance between the j th training sample and the test sample; S3.2、According to the ascending order of the distance between the training samples and the test samples, the first N (1≤N≤j) training samples are selected as the locally optimal training samples, and the inverse distance weighted (IDW) coefficient is calculated for each selected locally optimal training sample, as shown in equation (8); （8） where the subscript k is the k th sample of the local optimal training sample; is the Euclidean distance between the k th local optimal training sample and the test sample; ε is a very small additional value to avoid the case of division by zero, and ε = 0.001; the weighting matrix W is defined as formula (9); （9）； S3.3、After optimizing the samples, the spectral reconstruction is performed using the pseudo-inverse method （10） （11） The superscript "-1 " represents the matrix inverse; R Train is the spectral reflectance of the selected local training sample; P Train is the normalized response value of the training sample; P Test normalized response values representative of the test sample; R is the reconstructed spectral reflectance.

5. The method of claim 4, wherein: The specific steps of sorting the self-defined error set in the step S4 are as follows: S4.1、According to the reconstructed spectral reflectance, the spectral reconstruction error and the chroma error are calculated, and we select the root mean square error (RMSE), goodness of fit (GFC), peak signal-to-noise ratio (PSNR), spectral angle (SAM), and color difference (∆E) as the self-defined error calculation indicators. The calculation of the above self-defined error calculation indicators is shown in equations (12)-(16); （12） （13） （14） （15） （16） wherein Rtest represents the true spectral reflectance of the test sample, and R represents the reconstructed spectral reflectance; S4.2、Using the above indicators, the self-defined error set is calculated; the values of the indicators are normalized and then multiplied to obtain the final self-defined error set value, and the calculation process is shown in equation (17); （17） wherein TOTALn i is a custom recovery error set score corresponding to the ith filter set consisting of the first n preferred filter. S4.3、Through equation (18), the optimal filter set under L, M, and S weighting is selected, and then the self-defined error set values of the three filter sets are compared to select the optimal filter set; （18） G n represent the best filter set for the current number of channels; S4.4、When the number of channels of the multispectral imaging system increases, the number of filters in the filter set also increases accordingly; at this time, the three optimal filter sets selected in (3) under L, M, and S weighting are respectively used as the initial filter set and the remaining filter set for exhaustive combination, and the operations of equations (6)-(18) are repeated until the filter set that meets the channel number requirement of the multispectral imaging system is selected.

6. The use of a method of spectral reconstruction of a filter based on human eye vision for the evaluation of the quality of printed matter, characterized in that ： A printed matter quality evaluation method, comprising the following steps: (a) obtaining printed matter: randomly sampling the printed matter printed by the printing machine to obtain a detection printed matter, and then placing the detection printed matter on a printing sample viewing table; (b) calculating the spectral reflectance R of the detection printed matter using the spectral reconstruction method based on the human eye visual screening filter according to any one of claims 1-5 to obtain the spectral reflectance R; (c) comparing the spectral reflectance R of the detection printed matter with the spectral reflectance of the printed original manuscript, and calculating the root mean square error value RMSE between them, the calculation formula is as follows: ； In the formula, r is the spectral reflectance of the printed original manuscript, and n is the wavelength dimension; (d) when the root mean square error value is within 0.05, it is a qualified product; when the root mean square error value is outside 0.05, it is an unqualified product.

Images