Underwater image depth-of-field prediction method and system based on Lambert-Beer law

Abstract

The invention discloses an underwater image depth-of-field prediction method and system based on the Lambert-Beer law, and the method comprises the steps: carrying out the preprocessing of a to-be-predicted underwater image, and obtaining a noise reduction image; according to the Lambert-Beer law, calculating a first prediction distance and a second prediction distance corresponding to each pixel point in the noise reduction image; wherein the first prediction distance and the second prediction distance are dual-channel light propagation distances corresponding to pixel points; and according to the first prediction distance and the second prediction distance corresponding to each pixel point, respectively calculating to obtain a depth value of each pixel point. According to the invention, the efficiency of underwater image depth information prediction can be improved.

Description

Technical Field

[0001] The present invention relates to the field of image depth prediction, and in particular to a method and system for predicting the depth of field of an underwater image based on the Lambert-Beer law. Background Art

[0002] When estimating underwater depth information, a common high-precision method is to use high-precision measuring instruments for measurement. First, sound waves are emitted by sonar equipment. When the sound waves encounter the seabed or other obstacles, they are reflected back to the receiver, and the water depth data is calculated based on the propagation time of the sound waves. This process needs to take into account the changes in the propagation speed of sound waves in water, which is usually affected by factors such as water temperature, salinity and pressure. Therefore, in actual operation, these environmental parameters need to be measured synchronously to ensure the accuracy of the data. In addition, the initially collected data may be affected by noise and errors, so data cleaning and correction are required. Common methods include moving average, Kalman filtering and other technologies to filter out outliers and interference signals to obtain a smooth and accurate depth curve. Finally, through multiple measurements and data fusion technology, the reliability and resolution of depth information can be further improved.

[0003] The above method used in the prior art is complicated to operate and is easily interfered by external noise, resulting in reduced accuracy. Therefore, it is necessary to develop a simpler and more efficient method for predicting underwater image depth information. Summary of the invention

[0004] In order to overcome the defects of the above-mentioned prior art, the present invention provides a method for predicting the depth of field of underwater images based on the Lambert-Beer law. After the image to be predicted is preprocessed to obtain a denoised image, the depth value of each pixel is calculated according to the Lambert-Beer law. There is no need to consider a variety of environmental parameters, which can improve the efficiency of underwater image depth information prediction. Another purpose of the present invention is to provide an underwater image depth prediction device that implements the method, which is easy to operate and more efficient.

[0005] The present invention provides a method for predicting depth of field of an underwater image based on the Lambert-Beer law, comprising the following steps:

[0006] Preprocess the underwater image to be predicted to obtain a denoised image;

[0007] According to the Beer-Lambert law, a first predicted distance and a second predicted distance corresponding to each pixel point in the denoised image are calculated; wherein the first predicted distance and the second predicted distance are both dual-channel light propagation distances corresponding to the pixel point;

[0008] The depth value of each pixel point is calculated based on the first predicted distance and the second predicted distance corresponding to each pixel point.

[0009] Furthermore, the preprocessing of the underwater image to be predicted to obtain a denoised image specifically includes:

[0010] Assume that the underwater image to be predicted is A, and define is the pixel value of row i and column j in channel c, then for each pixel Take one The centered n×n matrix Q c (i, j) as the neighborhood, and calculate the mean of all pixels in each of the neighborhoods and mean square error Wherein, n is the preset window size,

[0011] According to the mean value of all pixels in each neighborhood Mean square error Calculate the upper limit of the neighborhood pixels in channel c and the lower limit of the neighborhood pixels

[0012] Determine the pixel value in each neighborhood respectively The pixels other than the above are noise points, and the noise points are smoothed; wherein the smoothing formula is specifically:

[0013]

[0014] in, is the pixel value of the denoised image B in the i-th row and j-th column under channel c;

[0015] When the noise points in each of the neighborhoods are smoothed, the denoised image B is obtained.

[0016] Preferably, the upper limit of the neighborhood pixels in the channel c is and the lower limit of the neighborhood pixels The specific calculation formula is:

[0017]

[0018] Among them, μ is the preset adjustment factor.

[0019] Furthermore, the calculating, according to the Lambert-Beer law, the first predicted distance and the second predicted distance corresponding to each pixel in the denoised image specifically includes:

[0020] The relationship between the pixel value of each pixel in the denoised image under the RGB three channels and the light propagation distance corresponding to the pixel is obtained by Lambert-Beer law; wherein the light propagation distance is the total distance that light propagates from the water surface to the pixel;

[0021] According to the relationship, for each pixel point in the denoised image, the pixel values ​​under any two channels are selected as a calculation combination, and the first predicted distance and the second predicted distance are respectively calculated based on any two of the calculation combinations.

[0022] Preferably, the step of calculating the pixel value of each pixel in the denoised image under the RGB three channels by using the Lambert-Beer law specifically includes:

[0023] set up is the pixel value of the pixel in the ith row and jth column of the denoised image B under channels R, G, and B, then:

[0024]

[0025] Among them, I R ,I G ,I B are the initial light intensities of each channel and their values ​​are equal, a is the preset attenuation coefficient, λ R , G , B are the preset wavelength coefficients of each channel, l i,j is the light propagation distance corresponding to the pixel point in the i-th row and the j-th column.

[0026] Further, the depth value of each pixel point is calculated based on the first predicted distance and the second predicted distance corresponding to each pixel point, specifically including:

[0027] Assume that the first prediction distance and the second prediction distance corresponding to the pixel point in the i-th row and the j-th column are l and 1i,j and l 2i,j ,but:

[0028] l i,j =βl 1i,j +l 2i,j

[0029] Among them, L i,j is the depth value corresponding to the pixel in the i-th row and j-th column, and β is a preset hyperparameter.

[0030] Based on the design concept of the present invention, the present invention also provides an underwater image depth prediction system based on the Lambert-Beer law, including: a noise reduction module, a distance prediction module and a depth calculation module;

[0031] The denoising module is used to pre-process the underwater image to be predicted to obtain a denoised image;

[0032] The distance prediction module is used to calculate the first predicted distance and the second predicted distance corresponding to each pixel point in the denoised image according to the Lambert-Beer law; wherein the first predicted distance and the second predicted distance are both the dual-channel light propagation distances corresponding to the pixel point;

[0033] The depth calculation module is used to calculate the depth value of each pixel point according to the first predicted distance and the second predicted distance corresponding to each pixel point.

[0034] Furthermore, the denoising module is used to preprocess the underwater image to be predicted to obtain a denoised image, which specifically includes:

[0035] Assume that the underwater image to be predicted is A, and define is the pixel value of row i and column j in channel c, then for each pixel Take one The centered n×n matrix Q c (i, j) as the neighborhood, and calculate the mean of all pixels in each of the neighborhoods and mean square error Wherein, n is the preset window size,

[0036] According to the mean value of all pixels in each neighborhood Mean square error Calculate the upper limit of the neighborhood pixels in channel c and the lower limit of the neighborhood pixels

[0037] Determine the pixel value in each neighborhood respectively The pixels other than the above are noise points, and the noise points are smoothed; wherein the smoothing formula is specifically:

[0038]

[0039] in, is the pixel value of the denoised image B in the i-th row and j-th column under channel c;

[0040] When the noise points in each of the neighborhoods are smoothed, the denoised image B is obtained.

[0041] Furthermore, the distance prediction module is used to calculate the first predicted distance and the second predicted distance corresponding to each pixel in the denoised image according to the Lambert-Beer law, specifically including:

[0042] The relationship between the pixel value of each pixel in the denoised image under the RGB three channels and the light propagation distance corresponding to the pixel is obtained by using the Lambert-Beer law; wherein the light propagation distance is the total distance that the light propagates from the water surface to the pixel;

[0043] According to the relationship, for each pixel point in the denoised image, the pixel values ​​under any two channels are selected as a calculation combination, and the first predicted distance and the second predicted distance are respectively calculated based on any two of the calculation combinations.

[0044] Further, the depth calculation module is used to calculate the depth value of each pixel point according to the first predicted distance and the second predicted distance corresponding to each pixel point, specifically including:

[0045] Assume that the first prediction distance and the second prediction distance corresponding to the pixel point in the i-th row and the j-th column are l and 1i,j and l 2i,j ,but:

[0046] L i,j =βl 1i,j +l 2i,j

[0047] Among them, L i,j is the depth value corresponding to the pixel in the i-th row and j-th column, and β is a preset hyperparameter.

[0048] Compared with the prior art, the beneficial effects of this application are:

[0049] The present invention scientifically utilizes the Lambert-Beer law, and based on the Lambert-Beer law, summarizes the relationship between the pixel value of each pixel point in the image to be tested under the RGB three-channel and the light propagation distance corresponding to the pixel point, and quickly predicts the depth information corresponding to the pixel point through the pixel value of the underwater image under the RGB three-channel, which can not only reduce the filtering cost of the collected data, but also ensure the prediction accuracy of the depth information, thereby improving the prediction efficiency of the depth information.

[0050] The present invention further provides an optional system for implementing the method, which designs a noise reduction module, a distance prediction module and a depth calculation module, and is easy and efficient to operate. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic flow chart of a method for predicting depth of field of an underwater image based on the Lambert-Beer law is provided in accordance with an embodiment of the present invention.

[0052] Figure 2 A comparison diagram of noise reduction effects of different window sizes provided by another embodiment of the present invention.

[0053] Figure 3 A schematic diagram of the structure of an underwater image depth prediction system based on the Lambert-Beer law provided by another embodiment of the present invention. DETAILED DESCRIPTION

[0054] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0055] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0056] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0057] Reference Figure 1 , is a flow chart of a method for predicting depth of field of an underwater image based on the Lambert-Beer law provided by an embodiment of the present invention, comprising the following steps:

[0058] S1: preprocess the underwater image to be predicted to obtain a denoised image;

[0059] S2: Calculate the first predicted distance and the second predicted distance corresponding to each pixel point in the denoised image according to the Lambert-Beer law; wherein the first predicted distance and the second predicted distance are both dual-channel light propagation distances corresponding to the pixel point;

[0060] S3: Calculate the depth value of each pixel point according to the first predicted distance and the second predicted distance corresponding to each pixel point.

[0061] For step S1, specifically, preprocessing the underwater image to be predicted to obtain a denoised image specifically includes:

[0062] Assume that the underwater image to be predicted is A, and define is the pixel value of row i and column j in channel c, then for each pixel Take one The centered n×n matrix Q c (i, j) as the neighborhood, and calculate the mean of all pixels in each of the neighborhoods and mean square error Wherein, n is the preset window size,

[0063] According to the mean value of all pixels in each neighborhood Mean square error Calculate the upper limit of the neighborhood pixels in channel c and the lower limit of the neighborhood pixels

[0064] Determine the pixel value in each neighborhood respectively The pixels other than the above are noise points, and the noise points are smoothed; wherein the smoothing formula is specifically:

[0065]

[0066] in, is the pixel value of the denoised image B in the i-th row and j-th column under channel c;

[0067] When the noise points in each of the neighborhoods are smoothed, the denoised image B is obtained.

[0068] Preferably, the upper limit of the neighborhood pixels in the channel c is and the lower limit of the neighborhood pixels The specific calculation formula is:

[0069]

[0070] Among them, μ is the preset adjustment factor.

[0071] In a preferred embodiment, for the original underwater image A to be predicted in RGB format, in order to prevent the influence of local mutation values ​​(noise and impurities), an n×n matrix box is used to define the pixel point and its surrounding points as Q, and the 2 times variance of the statistical distribution is used as a standard to remove the noise points, and then:

[0072]

[0073] in and are the mean and mean square error (MSE) of the c channel respectively, and μ is a parameter that controls the dynamics of the image. and The pixel values ​​of the denoised image are respectively used as the upper and lower limits of the pixel, and the points whose pixel values ​​exceed the upper and lower limits are used as noise points. After smoothing all the noise points, a statistically noise-free image, i.e. the denoised image B, can be obtained.

[0074] At the same time, in order to achieve a satisfactory noise reduction effect, further experiments are needed on the window size, that is, the value of n, to select the optimal window size. Figure 2 , which is a comparison diagram of noise reduction effects of different window sizes provided by another embodiment of the present invention. Figure 2 Seven original images were selected for the experiment, and the values ​​of n were 3, 5, 7, 9, 11, 13 and 15 respectively. Figure 2 It can be seen that when n is equal to 7, the comprehensive visual effect is the best, and when n exceeds 7, different degrees of unclearness can be observed. Therefore, the optimal window size is n=7.

[0075] For step S2, specifically, calculating the first predicted distance and the second predicted distance corresponding to each pixel point in the denoised image according to the Lambert-Beer law specifically includes:

[0076] The relationship between the pixel value of each pixel in the denoised image under the RGB three channels and the light propagation distance corresponding to the pixel is obtained by Lambert-Beer law; wherein the light propagation distance is the total distance that light propagates from the water surface to the pixel;

[0077] According to the relationship, for each pixel point in the denoised image, the pixel values ​​under any two channels are selected as a calculation combination, and the first predicted distance and the second predicted distance are respectively calculated based on any two of the calculation combinations.

[0078] Preferably, the step of calculating the pixel value of each pixel in the denoised image under the RGB three channels by using the Lambert-Beer law specifically includes:

[0079] set up is the pixel value of the pixel in the ith row and jth column of the denoised image B under channels R, G, and B, then:

[0080]

[0081] Among them, I R ,I G ,I B are the initial light intensities of each channel and their values ​​are equal, a is the preset attenuation coefficient, λ R , G , B are the preset wavelength coefficients of each channel, l i,j is the light propagation distance corresponding to the pixel point in the i-th row and the j-th column.

[0082] In a preferred embodiment, the relationship between the pixel value of a certain point in the underwater image and the total distance that the light corresponding to the point propagates in the water (i.e., the light propagation distance) can be obtained from the Lambert-Beer law:

[0083]

[0084] Among them, I R ,I G ,I B is the initial light intensity of different channels, that is, the light intensity of different channels from the sun to the water surface is equal.

[0085] Therefore, for each pixel point in the denoised image, the pixel values ​​under any two channels are selected as the calculation combination. In the preferred embodiment, two groups of R channel and G channel, and B channel and G channel are selected as the calculation combination.

[0086] Compare the pixel value of the pixel in the i-th row and j-th column under the R channel with the pixel value of the pixel under the G channel, and we have:

[0087]

[0088] Among them, l 1i,j This is the first prediction distance. Transforming the above formula yields:

[0089]

[0090] Similarly, comparing the pixel value of the pixel in the i-th row and j-th column under the B channel with the pixel value of the pixel under the G channel, we have:

[0091]

[0092] Among them, l 2i,j This is the second prediction distance. Transforming the above formula, we can get:

[0093]

[0094] In this way, the first predicted distance and the second predicted distance can be obtained.

[0095] For step S3, specifically, the step of calculating the depth value of each pixel point according to the first predicted distance and the second predicted distance corresponding to each pixel point includes:

[0096] Assume that the first prediction distance and the second prediction distance corresponding to the pixel point in the i-th row and the j-th column are l and 1i,j and l 2i,j ,but:

[0097] L i,j =βl 1i,j +l 2i,j

[0098] Among them, l i,j is the depth value corresponding to the pixel in the i-th row and j-th column, and β is a preset hyperparameter.

[0099] Reference Figure 3 , is a schematic diagram of the structure of an underwater image depth prediction system based on the Lambert-Beer law provided by another embodiment of the present invention, comprising: a noise reduction module 101, a distance prediction module 102 and a depth calculation module 103;

[0100] The denoising module 101 is used to pre-process the underwater image to be predicted to obtain a denoised image;

[0101] The distance prediction module 102 is used to calculate the first predicted distance and the second predicted distance corresponding to each pixel point in the denoised image according to the Lambert-Beer law; wherein the first predicted distance and the second predicted distance are both the dual-channel light propagation distances corresponding to the pixel point;

[0102] The depth calculation module 103 is used to calculate the depth value of each pixel point according to the first predicted distance and the second predicted distance corresponding to each pixel point.

[0103] Furthermore, the noise reduction module 101 is used to pre-process the underwater image to be predicted to obtain a noise-reduced image, which specifically includes:

[0104] Assume that the underwater image to be predicted is A, and define is the pixel value of row i and column j in channel c, then for each pixel Take one The centered n×n matrix Q c (i, j) as the neighborhood, and calculate the mean of all pixels in each of the neighborhoods and mean square error Wherein, n is the preset window size,

[0105] According to the mean value of all pixels in each neighborhood Mean square error Calculate the upper limit of the neighborhood pixels in channel c and the lower limit of the neighborhood pixels

[0106] Determine the pixel value in each neighborhood respectively The pixels other than the above are noise points, and the noise points are smoothed; wherein the smoothing formula is specifically:

[0107]

[0108] in, is the pixel value of the denoised image B in the i-th row and j-th column under channel c;

[0109] When the noise points in each of the neighborhoods are smoothed, the denoised image B is obtained.

[0110] Furthermore, the distance prediction module 102 is used to calculate the first predicted distance and the second predicted distance corresponding to each pixel in the denoised image according to the Lambert-Beer law, specifically including:

[0111] The relationship between the pixel value of each pixel in the denoised image under the RGB three channels and the light propagation distance corresponding to the pixel is obtained by Lambert-Beer law; wherein the light propagation distance is the total distance that light propagates from the water surface to the pixel;

[0112] According to the relationship, for each pixel point in the denoised image, the pixel values ​​under any two channels are selected as a calculation combination, and the first predicted distance and the second predicted distance are respectively calculated based on any two of the calculation combinations.

[0113] Further, the depth calculation module 103 is used to calculate the depth value of each pixel point according to the first predicted distance and the second predicted distance corresponding to each pixel point, specifically including:

[0114] Assume that the first prediction distance and the second prediction distance corresponding to the pixel point in the i-th row and the j-th column are l and 1i,j and l 2i,j ,but:

[0115] L i,j =βl 1i,j +l 2i,j

[0116] Among them, L i,j is the depth value corresponding to the pixel in the i-th row and j-th column, and β is a preset hyperparameter.

[0117] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for predicting depth of field of underwater images based on Lambert-Beer law, characterized in that: The steps include: Preprocess the underwater image to be predicted to obtain a denoised image; According to the Beer-Lambert law, a first predicted distance and a second predicted distance corresponding to each pixel point in the denoised image are calculated; wherein the first predicted distance and the second predicted distance are both dual-channel light propagation distances corresponding to the pixel point; The depth value of each pixel point is calculated based on the first predicted distance and the second predicted distance corresponding to each pixel point.

2. The method for predicting depth of field of underwater images based on the Lambert-Beer law as claimed in claim 1, characterized in that: The preprocessing of the underwater image to be predicted to obtain a denoised image specifically includes: Assume that the underwater image to be predicted is A, and define is the pixel of A in the i-th row and j-th column under channel c, then for each pixel Take one The centered n×n matrix Q c (i, j) as the neighborhood, and calculate the mean of all pixels in each of the neighborhoods and mean square error Wherein, n is the preset window size, According to the mean value of all pixels in each neighborhood And mean square error Calculate the upper limit of the neighborhood pixels in channel c and the lower limit of the neighborhood pixels Determine the pixel value in each neighborhood respectively The pixels other than the above are noise points, and the noise points are smoothed; wherein the smoothing formula is specifically: in, is the pixel of the denoised image B in the i-th row and j-th column under channel c; When the noise points in each of the neighborhoods are smoothed, the denoised image B is obtained.

3. The method for predicting depth of field of underwater images based on the Lambert-Beer law as claimed in claim 2, characterized in that: The upper limit of the neighborhood pixels in the channel c and the lower limit of the neighborhood pixels The specific calculation formula is: Among them, μ is the preset adjustment factor.

4. The method for predicting depth of field of underwater images based on the Lambert-Beer law as claimed in claim 1, characterized in that: The calculating, according to the Lambert-Beer law, the first predicted distance and the second predicted distance corresponding to each pixel point in the denoised image specifically includes: The relationship between the pixel value of each pixel in the denoised image under the RGB three channels and the light propagation distance corresponding to the pixel is obtained by Lambert-Beer law; wherein the light propagation distance is the total distance that light propagates from the water surface to the pixel; According to the relationship, for each pixel point in the denoised image, the pixel values ​​under any two channels are selected as a calculation combination, and the first predicted distance and the second predicted distance are respectively calculated based on any two of the calculation combinations.

5. The method for predicting depth of field of underwater images based on the Lambert-Beer law as claimed in claim 4, characterized in that: The relationship between the pixel value of each pixel in the denoised image under the RGB three channels and the light propagation distance corresponding to the pixel is obtained by using the Lambert-Beer law, specifically including: set up is the pixel value of the pixel in the ith row and jth column of the denoised image B under channels R, G, and B, then: Among them, I R ,I G ,I B are the initial light intensities of each channel and their values ​​are equal, a is the preset attenuation coefficient, λ R , G , B are the preset wavelength coefficients of each channel, l i,j is the light propagation distance corresponding to the pixel point in the i-th row and the j-th column.

6. The method for predicting depth of field of underwater images based on the Lambert-Beer law as claimed in claim 5, characterized in that: The calculating the depth value of each pixel point according to the first predicted distance and the second predicted distance corresponding to each pixel point specifically includes: Assume that the first prediction distance and the second prediction distance corresponding to the pixel point in the i-th row and the j-th column are l and 1i,j and l 2i,j ,but: L i,j ＝βl 1i,j +l 2i,j Among them, L i,j is the depth value corresponding to the pixel in the i-th row and j-th column, and β is a preset hyperparameter.

7. An underwater image depth prediction system based on Lambert-Beer law, characterized in that: include: Noise reduction module, distance prediction module and depth calculation module; The denoising module is used to pre-process the underwater image to be predicted to obtain a denoised image; The distance prediction module is used to calculate the first predicted distance and the second predicted distance corresponding to each pixel point in the denoised image according to the Lambert-Beer law; wherein the first predicted distance and the second predicted distance are both the dual-channel light propagation distances corresponding to the pixel point; The depth calculation module is used to calculate the depth value of each pixel point according to the first predicted distance and the second predicted distance corresponding to each pixel point.

8. The underwater image depth prediction system based on the Lambert-Beer law as claimed in claim 7, characterized in that: The denoising module is used to pre-process the underwater image to be predicted to obtain a denoised image, which specifically includes: Assume that the underwater image to be predicted is A, and define is the pixel value of row i and column j in channel c, then for each pixel Take one The centered n×n matrix Q c (i, j) as the neighborhood, and calculate the mean of all pixels in each of the neighborhoods and mean square error Wherein, n is the preset window size, According to the mean value of all pixels in each neighborhood And mean square error Calculate the upper limit of the neighborhood pixels in channel c and the lower limit of the neighborhood pixels Determine the pixel value in each neighborhood respectively The pixels other than the above are noise points, and the noise points are smoothed; wherein the smoothing formula is specifically: in, is the pixel value of the denoised image B in the i-th row and j-th column under channel c; When the noise points in each of the neighborhoods are smoothed, the denoised image B is obtained.

9. The underwater image depth prediction system based on Beer-Lambert law as claimed in claim 7, characterized in that: The distance prediction module is used to calculate the first predicted distance and the second predicted distance corresponding to each pixel point in the denoised image according to the Lambert-Beer law, specifically including: The relationship between the pixel value of each pixel in the denoised image under the RGB three channels and the light propagation distance corresponding to the pixel is obtained by using the Lambert-Beer law; wherein the light propagation distance is the total distance that the light propagates from the water surface to the pixel; According to the relationship, for each pixel point in the denoised image, the pixel values ​​under any two channels are selected as a calculation combination, and the first predicted distance and the second predicted distance are respectively calculated based on any two of the calculation combinations.

10. The underwater image depth prediction system based on Lambert-Beer law as claimed in claim 7, characterized in that: The depth calculation module is used to calculate the depth value of each pixel point according to the first predicted distance and the second predicted distance corresponding to each pixel point, and specifically includes: Assume that the first prediction distance and the second prediction distance corresponding to the pixel point in the i-th row and the j-th column are l and 1i,j and l 2i,j ,but: l i,j ＝βl 1i,j +l 2i,j Among them, l i,j is the depth value corresponding to the pixel in the i-th row and j-th column, and β is a preset hyperparameter.

Images