Printing ink dosage online detection method based on edge detection and random forest

Abstract

The invention discloses a printing ink consumption online detection method based on edge detection and a random forest. The method comprises the following steps: obtaining an original manuscript RGB image; partitioning the original manuscript RGB image; calculating the L * a * b * value of each ink area; outputting the CMYK ink amount of each ink area through a CMYK ink amount estimation model, and calculating the average CMYK ink amount of each kind of ink in each ink area; processing the manuscript RGB image by using the trained random forest algorithm to obtain the ink thickness corresponding to the manuscript RGB image; establishing a theoretical plane regression model of the image ink thickness and the chromatic value, and determining a plane regression model and parameter estimation; determining an optimal control parameter of the L * a * b * value; establishing a printing ink dosage online detection model; determining a theoretical CMYK ink amount corresponding to the original manuscript RGB image; the average amount of CMYK ink is corrected using the theoretical amount of CMYK ink. According to the invention, accurate ink dosage can be obtained.

Description

Technical Field

[0001] The invention relates to the technical field of printing, and in particular to an online ink usage detection method based on edge detection and random forest. Background Art

[0002] The amount of printing ink used is an important factor affecting the color of printed products. In batch printing, the establishment of the relationship between the color characteristics of the color separation plate image and the printing ink can achieve high-speed, high-quality and stable printing products, providing data guarantee for the high-fidelity color performance on paper and other substrates. However, the existing ink usage estimation is usually based on the traditional judgment method of practitioners' experience, visual inspection of originals and test samples, and roughly estimates the ink usage of each color separation plate. This estimation method is not accurate enough and has large errors. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides an online detection method for ink usage based on edge detection and random forest, which solves the problem of insufficient accuracy of online monitoring of ink usage in the prior art.

[0004] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: an online detection method of ink usage based on edge detection and random forest, comprising the following steps:

[0005] S1, using a CCD camera to obtain an original RGB image, and converting the original RGB image into an L*a*b* color space;

[0006] S2, partitioning the original RGB image according to a preset partitioning rule to obtain at least two ink areas;

[0007] S3, calculating the L*a*b* value of each ink area in at least two ink areas;

[0008] S4, inputting the L*a*b* value of each ink zone into the trained CMYK ink quantity estimation model, outputting the CMYK ink quantity of each ink zone through the CMYK ink quantity estimation model, and obtaining the number of pixels covered by each ink in each ink zone, and calculating the average CMYK ink quantity of each ink in each ink zone according to the CMYK ink quantity of each ink zone and the number of pixels covered by each ink in each ink zone;

[0009] S5, using the trained random forest algorithm to process the original RGB image to obtain the ink thickness corresponding to the original RGB image;

[0010] S6. Establish a theoretical plane regression model of image ink thickness and chromaticity value, substitute L*a*b* value and ink thickness into the theoretical plane regression model, and obtain a plane regression model and parameter estimation;

[0011] S7, performing dimensionality reduction processing on the plane regression model to obtain the optimal control parameters of the L*a*b* value;

[0012] S8. Establish an online detection model for ink usage based on the optimal control parameters;

[0013] S9, substituting the ink thickness corresponding to the original RGB image into the online ink usage detection model to obtain the theoretical CMYK ink amount corresponding to the original RGB image;

[0014] S10. Correct the average CMYK ink amount using the theoretical CMYK ink amount to obtain a final CMYK ink amount corresponding to the original RGB image.

[0015] The beneficial effect of the above scheme is: the present invention proposes an online detection method for ink usage based on edge detection and random forest, which can achieve high-speed, high-quality and stable printing, and provide data guarantee for high-fidelity color performance on paper and other substrates.

[0016] Furthermore, in S1, the formula used to convert the original RGB image into the L*a*b* color space is:

[0017]

[0018] Among them, L * Represents the brightness component, a * The brightness component representing the color change from green to red, b * It represents the brightness component of the color change from blue to yellow. X, Y, and Z represent the three stimulus values ​​of the original RGB image. X0, Y0, and Z0 represent the three stimulus values ​​of a completely diffuse reflector surface under the same lighting conditions.

[0019] Furthermore, in S4, the training process of the trained CMYK ink volume estimation model is as follows:

[0020] Pure color bar images of different colors are taken as input, and the ink amounts corresponding to the pure color bar images of different colors are taken as output, and the neural network is trained to obtain a trained CMYK ink amount estimation model.

[0021] Furthermore, in S5, the training process of the trained random forest algorithm is:

[0022] Obtain a training data set, divide the training set into a ratio of 6:2:2 to obtain a training set, a validation set, and a test set, respectively, and use the training set to train the initial random forest algorithm to obtain a trained random forest algorithm.

[0023] Furthermore, the prediction function f(x) of the initial random forest algorithm is:

[0024]

[0025] Among them, x represents the input data, h j (·) represents a single tree in the random forest algorithm, and J represents the number of trees in the random forest algorithm.

[0026] Furthermore, it is characterized in that, in S6, the theoretical plane regression model is:

[0027] T=k1+k2L*+k3a*+k4b*

[0028] Where T represents the ink thickness, k1 represents the constant term, k2, k3, k4 represent the regression coefficients, and L * Represents the brightness component, a * The brightness component representing the color change from green to red, b * The brightness component representing the color change from blue to yellow.

[0029] Furthermore, in S8, the online detection model of ink usage is:

[0030] Ink CMK:

[0031] Ink Y:

[0032] Where T represents the ink thickness, L * represents the brightness component, b * It represents the brightness component of the color change from blue to yellow, e represents a natural constant, and m and n represent constants.

[0033] The beneficial effect of the above further scheme is that it can obtain more accurate ink dosage, achieve high-speed, high-quality and stable printing, and provide data guarantee for high-fidelity color performance on paper and other substrates. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 The figure is a flow chart of an online ink usage detection method based on edge detection and random forest.

[0035] Figure 2 Schematic diagram of a device for acquiring an RGB image of an original document.

[0036] Figure 3 It is the comparison result of the estimated grayscale image of each color separation plate and the actual production ink usage data. DETAILED DESCRIPTION

[0037] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0038] like Figure 1As shown, an online ink usage detection method based on edge detection and random forest includes the following steps:

[0039] S1. Use a CCD camera to obtain an original RGB image, and convert the original RGB image into an L*a*b* color space.

[0040] In this embodiment, the formula used to convert the original RGB image into the L*a*b* color space is:

[0041]

[0042] Among them, L * Represents the brightness component, a * The brightness component representing the color change from green to red, b * It represents the brightness component of the color change from blue to yellow. X, Y, and Z represent the three stimulus values ​​of the original RGB image. X0, Y0, and Z0 represent the three stimulus values ​​of a completely diffuse reflector surface under the same lighting conditions.

[0043] For example, Figure 2 As shown, Figure 2 Schematic diagram of a device for obtaining an RGB image of an original document. Figure 2 In the embodiment, the device for acquiring the RGB image of the original manuscript may include a computer, an image acquisition card, a CCD camera, a light source and a printed matter. The printed matter may be the RGB image of the original manuscript in this embodiment. The CCD camera in this embodiment may be a linear CCD camera. The data captured by the CCD camera is the RGB value of each pixel in the RGB image of the original manuscript. In a computer screen, there are 175 dots per inch, and the resolution of the CCD camera is 500 dots per inch. If the area of ​​a dot is 50%, a dot can be divided into approximately 4 pixels.

[0044] Optionally, the number of the CCD cameras may be at least one, and the specific number may be selected according to actual conditions and is not specifically limited here.

[0045] The CCD camera is connected to an image acquisition card and continuously inputs the acquired image data into a computer, which receives and processes these digital images.

[0046] S2. Partition the original RGB image according to a preset partition rule to obtain at least two ink areas.

[0047] Exemplarily, the preset partitioning rule may be average partitioning according to the width or length of the original RGB image, or partitioning according to image features. No specific limitation is made here, and the partitioning method can be selected according to actual conditions.

[0048] For example, the original RGB image can be evenly partitioned: the width of each ink zone can be set to d (mm), the resolution of the original RGB image can be set to p (dpi), that is, there are p pixels in one inch, and the number of pixels in the horizontal direction is m. Then the number of pixels D in each ink zone in the digital image can be calculated as: D = dp / 25.4, and the number of ink zones S can be calculated as S = m / D.

[0049] Optionally, the size of the ink area may be 32.5 mm, and the resolution of the CCD camera may be 500 dpi.

[0050] S3. Calculate the L*a*b* value of each ink area in at least two ink areas.

[0051] Exemplarily, the calculation formula for calculating the L*a*b* value of each ink zone can refer to the formula used in S1 to convert the original RGB image into the L*a*b* color space, which will not be described in detail here.

[0052] S4. Input the L*a*b* value of each ink zone into the trained CMYK ink quantity estimation model, output the CMYK ink quantity of each ink zone through the CMYK ink quantity estimation model, and obtain the number of pixels covered by each ink in each ink zone, and calculate the average CMYK ink quantity of each ink in each ink zone according to the CMYK ink quantity of each ink zone and the number of pixels covered by each ink in each ink zone.

[0053] In this embodiment, the training process of the trained CMYK ink volume estimation model is as follows:

[0054] Pure color bar images of different colors are taken as input, and the ink amounts corresponding to the pure color bar images of different colors are taken as output, and the neural network is trained to obtain a trained CMYK ink amount estimation model.

[0055] For example, a pixel is regarded as a solid point, and pure color bars with different CMYK ink quantities can be printed, and the ink quantities corresponding to the pure color bars with different CMYK ink quantities are used as training data to train the neural network, thereby obtaining a trained CMYK ink quantity estimation model. The trained neural network can be saved as "net.mat".

[0056] You can select a standard print, scan it with a CCD camera, and then calculate and save its CMYK ink volume as standard data. While using a linear CCD camera to continuously scan the print, draw an image in the computer at regular intervals to ensure that each image has not only the same size, but also the same starting and ending points. Use multi-threaded programming technology to create two synchronous threads, one for scanning and the other for drawing, so that scanning and drawing images can be performed simultaneously.

[0057] When drawing an image, edge detection technology is used to locate the starting point and the end point using the border of the image to be printed as the detection symbol. The CCD camera starts drawing the image when it scans a border and stops drawing when it scans another border. After that, the drawn image needs to be preprocessed, that is, the captured image is compared with the image to be printed so that they not only have the same size, but also the position of the corresponding pixels is the same, so as to achieve geometric correction of the captured image. The specific method can be: the first step is to perform a geometric transformation on the pixel coordinate space of the original image to ensure that these pixels are in the correct position. The second step is to determine the new pixel grayscale value.

[0058] For example, the number of pixels covered by each ink in each ink zone can be obtained by using a statistical method or a trained statistical neural network model, and the average CMYK ink amount of each ink in each ink zone is calculated according to the CMYK ink amount of each ink zone and the number of pixels covered by each ink in each ink zone.

[0059] For example, the number of pixels covered by cyan ink is N, the total amount of cyan ink is M, the average amount of cyan ink in the ink area is M / N, and the average CMYK ink amount of other colors of ink is calculated in the same way.

[0060] S5. Use the trained random forest algorithm to process the original RGB image to obtain the ink thickness corresponding to the original RGB image.

[0061] In this embodiment, the training process of the trained random forest algorithm is:

[0062] Obtain a training data set, divide the training set into a ratio of 6:2:2 to obtain a training set, a validation set, and a test set, respectively, and use the training set to train the initial random forest algorithm to obtain a trained random forest algorithm.

[0063] In this embodiment, the prediction function f(x) of the initial random forest algorithm is:

[0064]

[0065] Among them, x represents the input data, h j (·) represents a single tree in the random forest algorithm, and J represents the number of trees in the random forest algorithm.

[0066] For example, the training data set can be the original RGB image and its corresponding ink thickness. In the Python environment, the training set can be used to train the initial random forest algorithm, find the prediction function f(x) through the optimization process, and iteratively minimize the estimated loss value EXY(L(Y,f(x))). Each random forest tree uses f(x) to provide a vector of ink prediction values. The prediction function can be defined using the loss function, and the random variable X can be X=(X1,X2,...,X p ) T .

[0067] Optionally, the parameters of the initial random forest algorithm can be set to the parameters in Table 1.

[0068] Table 1 Parameter settings of random forest algorithm

[0069]

[0070] S6. Establish a theoretical plane regression model of image ink thickness and chromaticity value, substitute L*a*b* value and ink thickness into the theoretical plane regression model, and obtain the plane regression model and parameter estimation.

[0071] In this embodiment, the theoretical plane regression model is:

[0072] T=k1+k2L*+k3a*+k4b*

[0073] Where T represents the ink thickness, k1 represents the constant term, k2, k3, k4 represent the regression coefficients, and L * Represents the brightness component, a * The brightness component representing the color change from green to red, b * The brightness component representing the color change from blue to yellow.

[0074] Exemplarily, the L*a*b* value and the ink thickness are substituted into the theoretical plane regression model, and the obtained plane regression model and parameter estimation can be shown in Table 2.

[0075] Table 2 Planar regression model and parameter estimates

[0076]

[0077] S7. Perform dimensionality reduction processing on the plane regression model to obtain the optimal control parameters of the L*a*b* value.

[0078] For example, the p-value of the t-test on L*a*b* using multiple linear regression can be shown in Table 3.

[0079] Table 3 p-values ​​of t-test on L*a*b* using multiple linear regression

[0080]

[0081] It can be concluded from Table 3 that, in the CMYK ink volume, the optimal control parameters for C value are L* and b*; for M value, L* is determined to be the optimal control parameter; similarly, b* is determined to be the optimal control parameter for Y value, and L* is the optimal control parameter for K value.

[0082] S8. Establish an online ink usage detection model based on the optimal control parameters.

[0083] In this embodiment, the online detection model of ink usage is:

[0084] Ink CMK:

[0085] Ink Y:

[0086] Where T represents the ink thickness, L * represents the brightness component, b * It represents the brightness component of the color change from blue to yellow, e represents a natural constant, and m and n represent constants.

[0087] S9. Substitute the ink thickness corresponding to the original RGB image into the online ink usage detection model to obtain the theoretical CMYK ink amount corresponding to the original RGB image.

[0088] S10. Correct the average CMYK ink amount using the theoretical CMYK ink amount to obtain a final CMYK ink amount corresponding to the original RGB image.

[0089] For example, in a specific embodiment, taking a daily necessities packaging bag printed on a roll of BOPP film as an example, the printing volume is 10 rolls, the size of a single film roll is 100000m*900mm, there are 4 color plates in total, CMYK, and it is assumed that the density of the printing ink after mixing the additives is 1g / cm 3 .use Figure 2 The device shown collects RGB images with a resolution of 500dpi. Read the color RGB image and obtain the CMYK ink volume data. The comparison result with the actual ink volume can be as follows: Figure 3 As shown, Figure 3 The grayscale image of each color separation plate is estimated and compared with the actual production ink usage data. Figure 3 It can be seen that the estimated ink volume is close to the actual ink volume and can be used in the production of printed products.

[0090] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the invention.

Claims

1. An online ink usage detection method based on edge detection and random forest, characterized in that: The method comprises: S1, using a CCD camera to obtain an original RGB image, and converting the original RGB image into an L*a*b* color space; S2, partitioning the original RGB image according to a preset partitioning rule to obtain at least two ink areas; S3, calculating the L*a*b* value of each ink area in the at least two ink areas; S4, inputting the L*a*b* value of each ink zone into a trained CMYK ink quantity estimation model, outputting the CMYK ink quantity of each ink zone through the CMYK ink quantity estimation model, and obtaining the number of pixels covered by each ink in each ink zone, and calculating the average CMYK ink quantity of each ink in each ink zone according to the CMYK ink quantity of each ink zone and the number of pixels covered by each ink in each ink zone; S5, using the trained random forest algorithm to process the original RGB image to obtain the ink thickness corresponding to the original RGB image; S6, establishing a theoretical plane regression model of image ink thickness and chromaticity value, substituting the L*a*b* value and the ink thickness into the theoretical plane regression model, and obtaining a plane regression model and parameter estimation; S7, performing dimensionality reduction processing on the plane regression model to obtain the optimal control parameters of the L*a*b* value; S8. Establishing an online ink usage detection model according to the optimal control parameters; S9, substituting the ink thickness corresponding to the original RGB image into the online ink usage detection model to obtain the theoretical CMYK ink amount corresponding to the original RGB image; S10. Correct the average CMYK ink amount using the theoretical CMYK ink amount to obtain a final CMYK ink amount corresponding to the original RGB image.

2. The method according to claim 1, characterized in that In S1, the formula used to convert the original RGB image into the L*a*b* color space is: Among them, L * Represents the brightness component, a * The brightness component representing the color change from green to red, b * It represents the brightness component of the color change from blue to yellow. X, Y, and Z represent the three stimulus values ​​of the original RGB image. X0, Y0, and Z0 represent the three stimulus values ​​of a completely diffuse reflector surface under the same lighting conditions.

3. The method according to claim 1, characterized in that: In S4, the training process of the trained CMYK ink quantity estimation model is as follows: Pure color bar images of different colors are used as inputs, and the ink quantities corresponding to the pure color bar images of different colors are used as outputs, and a neural network is trained to obtain the trained CMYK ink quantity estimation model.

4. The method according to claim 1, characterized in that: In S5, the training process of the trained random forest algorithm is: A training data set is obtained, and the training set is divided into a training set, a validation set, and a test set according to a ratio of 6:2:2 to obtain a training set, a validation set, and a test set, respectively. The training set is used to train an initial random forest algorithm to obtain the trained random forest algorithm.

5. The method according to claim 4, characterized in that The prediction function f(x) of the initial random forest algorithm is: Among them, x represents the input data, h j (·) represents a single tree in the random forest algorithm, and J represents the number of trees in the random forest algorithm.

6. The method according to claim 1, characterized in that In S6, the theoretical plane regression model is: T=k1+k2L*+k3a*+k4b* Where T represents the ink thickness, k1 represents the constant term, k2, k3, k4 represent the regression coefficients, and L * Represents the brightness component, a * The brightness component representing the color change from green to red, b * The brightness component representing the color change from blue to yellow.

7. The method according to claim 1, characterized in that In S8, the ink usage online detection model is: Ink CMK: Ink Y: Where T represents the ink thickness, L * represents the brightness component, b * It represents the brightness component of the color change from blue to yellow, e represents a natural constant, and m and n represent constants.

Images