Weak light image clustering Binning method based on Gaussian function and terminal

Abstract

The invention discloses a weak light image clustering Binning method and terminal based on a Gaussian function, and the method comprises the steps: obtaining dark field images under different exposure times and gains, and building a dark field noise value feature function according to the dark field images; identifying and clustering signal points and dark field noise points in a weak light image to be processed through the dark field noise value characteristic function; binning window weight distribution is carried out based on a Gaussian function, clustering operation is carried out according to the number of signal points in a Binning window, the gray value and the distributed weight after weight distribution, and a result image is output; according to the method, the dark field noise value characteristic function is created, the detection capacity of weak signals in the Binning window is improved, extremely weak signals cannot be submerged by noise, through clustering of the signal points and the dark field noise points, the signals are enhanced while the dark field noise value in the Binning window is restrained, and the signal-to-noise ratio of the weak light image is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a binning method and terminal for low-light image clustering based on Gaussian functions. Background Technology

[0002] Digital images, as a means of information transmission, are receiving increasing attention in daily life and work, and people's demands for image quality are also rising. However, due to limitations of shooting scenarios, in low-light environments such as cloudy days, nighttime, and microscopic imaging, the light intensity is low, the photosensitive chip receives less light signal, the signal value is small compared to the pixel value in a pure dark field, the dark field noise is high, the weak signals in the image are drowned out by noise, the signal-to-noise ratio is low, resulting in poor image quality and unsatisfactory visual effects in low-light imaging.

[0003] A common method to improve the signal-to-noise ratio of low-light images is to stack pixels using image binning. The basic idea is to directly accumulate the pixels in the binning window according to the set binning window size (e.g., 2x2, 4x4, etc.), and finally output the accumulated value of the binning window as a single pixel.

[0004] This method can amplify the signal by a certain factor, but at the same time, the noise is also amplified proportionally. For example, in a 4x4 window binning operation, a total of 16 pixels are combined. If the weak signal exists only in one pixel, the combined signal may be overwhelmed by the dark field noise value, which is equivalent to the signal not being amplified, but the dark field noise value being amplified by 16 times. In this case, the signal is lost.

[0005] In other words, conventional image binding amplifies the signal but also amplifies the noise proportionally, resulting in limited improvement in the signal-to-noise ratio. Furthermore, for extremely weak signals, conventional binding can cause the signal to be overwhelmed by noise, leading to signal loss. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a low-light image clustering binning method and terminal based on Gaussian function, which can avoid the image signal being overwhelmed by noise and effectively improve the signal-to-noise ratio of low-light images.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0008] A low-light image clustering binning method based on Gaussian function includes the following steps:

[0009] S1. Obtain dark field images under different exposure times and gains, and establish a dark field noise value feature function based on the dark field images;

[0010] S2. The signal points and dark field noise points in the low-light image to be processed are identified and clustered using the dark field noise value feature function.

[0011] S3. Perform a clustering (Binning) operation on the low-light image based on the recognition and clustering results, and output the result image.

[0012] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows:

[0013] A low-light image clustering binning terminal based on Gaussian function includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the low-light image clustering binning method based on Gaussian function described above.

[0014] The beneficial effects of this invention are as follows: The low-light image clustering binning method and terminal provided by this invention can create a dark field noise value feature function based on dark field images under different exposure times and gains, thereby better judging the signal points and dark field noise points in the binning window, clustering the signal points and dark field noise points, greatly improving the detection capability of weak signals, and even extremely weak signal points will not be submerged by noise; thus improving the signal-to-noise ratio of low-light images. Attached Figure Description

[0015] Figure 1 This is a flowchart of a low-light image clustering binning method based on Gaussian function according to an embodiment of the present invention;

[0016] Figure 2 This is a structural diagram of a low-light image clustering binning terminal based on a Gaussian function according to an embodiment of the present invention;

[0017] Figure 3 This is a schematic diagram illustrating the specific process of a low-light image clustering binning method based on Gaussian function according to an embodiment of the present invention.

[0018] Label Explanation:

[0019] 1. A Binning terminal for low-light image clustering based on Gaussian function; 2. Processor; 3. Memory. Detailed Implementation

[0020] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0021] Please refer to Figure 1 as well as Figure 3A Binning method for low-light image clustering based on Gaussian functions includes the following steps:

[0022] S1. Obtain dark field images under different exposure times and gains, and establish a dark field noise value feature function based on the dark field images;

[0023] S2. The signal points and dark field noise points in the low-light image to be processed are identified and clustered using the dark field noise value feature function.

[0024] S3. Perform a clustering (Binning) operation on the low-light image based on the recognition and clustering results, and output the result image.

[0025] As can be seen from the above description, the beneficial effects of the present invention are as follows: The low-light image clustering binning method based on Gaussian function of the present invention can create a dark field noise value feature function according to the dark field image under different exposure time and gain, thereby better judging the signal points and dark field noise points in the binning window, clustering the signal points and dark field noise points, greatly improving the detection capability of weak signals, and even extremely weak signal points will not be submerged by noise; thereby improving the signal-to-noise ratio of low-light images.

[0026] Furthermore, step S1 includes the following steps:

[0027] S11. Obtain dark field images under different exposure times and gains;

[0028] S12. Calculate the mean gray value of each pixel in each dark field image under different exposure times, fit the gray value-exposure time curve, and obtain the first characteristic function of the dark field noise value of each pixel as a function of exposure time:

[0029] dark_value_exp(row,col)=k(row,col)*exp_time+b(row,col);

[0030] Where, dark_value_exp represents the dark field noise value of each pixel under different exposure times, row / col represents the number of rows / columns of the noise point in the image, k represents the slope of the equation for each pixel changing with exposure time, exp_time represents the exposure time, and b represents the intercept of the equation for each pixel changing with exposure time.

[0031] S13. Calculate the mean gray value of each pixel in each dark field image under different gains, fit the gray value-gain curve, and obtain the second characteristic function of the mean dark field noise of the image as a function of gain:

[0032]

[0033] Where, dark_value_gain represents the mean value of dark field noise under different gains, a1, a2, a3, a4 represent the parameters of the exponential equation, and gain represents the gain value;

[0034] S14. Based on the first feature function and the second feature function, solve for the dark field noise value feature function of each pixel:

[0035]

[0036] Here, dark_value represents the dark field noise value.

[0037] As described above, a feature function for dark field noise value is constructed by fitting the gray value-exposure time curve and the gray value-gain curve.

[0038] Furthermore, step S2 specifically includes:

[0039] Based on the relationship between the grayscale value of each pixel in the Binning window of the low-light image and the dark field noise value, signal points and dark field noise points are identified and clustered. The criteria for judging signal points are as follows:

[0040] value_cur(i,j)>dark_value(i,j)+offset;

[0041] i,j∈[1,bin_win];

[0042] Where value_cur(i,j) represents the gray value of the pixel in the i-th row and j-th column of the Binning window of the low-light image, dark_value() represents the feature function of the dark field noise value, bin_win represents the size of the Binning window, and the number of rows and columns of the window are the same, and offset represents the preset offset amount.

[0043] As described above, signal points and dark field noise points are identified and clustered by establishing noise value feature functions.

[0044] Furthermore, step S3 specifically includes:

[0045] Binning window weights are assigned based on Gaussian function, and clustering binning is performed after weight assignment based on the number of signal points, gray values, and assigned weights in the binning window, outputting the resulting image.

[0046] As described above, the Binning window weight allocation based on the Gaussian function reduces the weight of noise values, which can suppress dark noise points in the window while enhancing signal points, thereby improving the signal-to-noise ratio of low-light images.

[0047] Furthermore, the Binning window weight allocation based on the Gaussian function includes the following steps:

[0048] S31. Generate Gaussian weights based on the current gain value and the size of the Binning window.

[0049] S32. Generate window weights based on the number of signal points in the Binning window;

[0050] S33. Generate the weight of each signal point in the Binning window based on the Gaussian weight and the window weight.

[0051] As described above, combining Gaussian weights and window weights to generate the weight of each signal point in the window makes the weight setting more reasonable.

[0052] Further, step S31 includes the following steps:

[0053] S311. Create a Gaussian function based on the Binning window size and the number of signal points:

[0054]

[0055] p∈[1,bin_win 2 ];

[0056] Where w represents the Gaussian function value, p represents the number of signal points in the window, bin_win represents the size of the Binning window, and gauss_para represents the preset adjustment coefficient;

[0057] S312. Normalize the Gaussian function values:

[0058]

[0059] p∈[1,bin_win 2 ];

[0060] Where w_normalize represents the Gaussian function value after normalization;

[0061] S313. Generate Gaussian weights:

[0062]

[0063] Where weight_gauss represents Gaussian weights, and ε represents the preset intensity coefficient.

[0064] As described above, Gaussian weights are generated in the above manner.

[0065] Furthermore, the window weights in step S32 are specifically generated according to the following formula:

[0066]

[0067] Where weight_win represents the window weight, p represents the number of signal points in the window, and bin_win represents the size of the Binning window.

[0068] As described above, window weights are generated in the manner described above.

[0069] Furthermore, the weights of each signal point in the Binning window are generated specifically based on Gaussian weights and window weights as follows:

[0070] weight=(1-weight_gauss(bin_win 2 +1-p))*weight_win;

[0071] Where weight_win represents the window weight, weight_gauss represents the Gaussian weight, p represents the number of signal points in the window, and bin_win represents the size of the Binning window.

[0072] As described above, the final weight data is generated by combining Gaussian weights and window weights in the above manner.

[0073] Furthermore, the clustering and binning operation based on the number of signal points, grayscale values, and assigned weights in the binning window is specifically as follows:

[0074] Choose different clustering binning methods based on the number of signal points in the binning window:

[0075] If the number of signal points in the Binning window is 0, then the grayscale value after Binning clustering is the product of the median of the window data and the total number of pixels in the window:

[0076] value_bin=median_value*bin_win 2 ;

[0077] Where median_value represents the median value of the window data, and bin_win represents the size of the Binning window;

[0078] If the number of signal points in the Binning window is not 0, then the weighted sum of the gray values ​​of the signal points and the weights corresponding to each signal point is used as the gray values ​​after clustering and Binning.

[0079]

[0080] Here, value_signal is the grayscale value corresponding to each signal point in the window.

[0081] As described above, different clustering methods can be selected based on whether there are signal points in the Binning window to improve the effectiveness of Binning clustering.

[0082] Please refer to Figure 2 A low-light image clustering binning terminal based on Gaussian function includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the low-light image clustering binning method based on Gaussian function described above.

[0083] The present invention provides a low-light image clustering binning method and terminal based on Gaussian function, which is applicable to clustering binning of low-light images to improve the signal-to-noise ratio of the images, and is particularly suitable for clustering binning of low-light images with a small number of signal points.

[0084] Please refer to Figure 1 and Figure 3 Embodiment 1 of the present invention is as follows:

[0085] A low-light image clustering binning method based on Gaussian function includes the following steps:

[0086] S1. Obtain dark field images under different exposure times and gains, and establish a dark field noise value feature function based on the dark field images;

[0087] Step S1 includes the following steps:

[0088] S11. Obtain dark field images under different exposure times and gains.

[0089] In this embodiment, the camera is placed in a dark environment (e.g., the camera can be covered with a dust cover). The camera is a monochrome camera. Different exposure times and gains are set to acquire n corresponding dark-field images, where n can be 50 or other values. The exposure time can be set to x segments, where x can be 4 or other values. For example, dark-field images can be captured with exposure times of 2s / 4s / 6s / 8s. The gain can be set to k segments, ranging from 1 to 255, where k can be 10. For example, dark-field images can be captured with gains of 1 / 5 / 10 / 50 / 100 / 150 / 170 / 200 / 240 / 255 at corresponding exposure times.

[0090] S12. Calculate the mean gray value of each pixel in each dark field image under different exposure times, fit the gray value-exposure time curve, and obtain the first characteristic function of the dark field noise value of each pixel as a function of exposure time:

[0091] dark_value_exp(row,col)=k(row,col)*exp_time+b(row,col);

[0092] Where, dark_value_exp represents the dark field noise value of each pixel under different exposure times, row / col represents the number of rows / columns of the noise point in the image, k represents the slope of the equation for each pixel changing with exposure time, exp_time represents the exposure time, and b represents the intercept of the equation for each pixel changing with exposure time.

[0093] S13. Calculate the mean gray value of each pixel in each dark field image under different gains, fit the gray value-gain curve, and obtain the second characteristic function of the mean dark field noise of the image as a function of gain:

[0094]

[0095] Where, dark_value_gain represents the mean value of dark field noise under different gains, a1, a2, a3, a4 represent the parameters of the exponential equation, and gain represents the gain value;

[0096] S14. Based on the first feature function and the second feature function, solve for the dark field noise value feature function of each pixel:

[0097]

[0098] Here, dark_value represents the dark field noise value.

[0099] S2. The signal points and dark field noise points in the low-light image to be processed are identified and clustered using the dark field noise value feature function.

[0100] Step S2 is as follows:

[0101] Based on the relationship between the grayscale value of each pixel in the Binning window of the low-light image and the dark field noise value, signal points and dark field noise points are identified and clustered. The criteria for judging signal points are as follows:

[0102] value_cur(i,j)>dark_value(i,j)+offset;

[0103] i,j∈[1,bin_win];

[0104] Where value_cur(i,j) represents the gray value of the pixel in the i-th row and j-th column of the Binning window of the low-light image, dark_value() represents the feature function of the dark field noise value, bin_win represents the size of the Binning window, and the number of rows and columns of the window are the same, and offset represents the preset offset amount.

[0105] The pixels that meet the above criteria are signal points, and the total number of signal points in the recording window is p. The remaining points are dark noise points.

[0106] In this embodiment, the offset is set to 0.5.

[0107] S3. Perform a clustering (Binning) operation on the low-light image based on the recognition and clustering results, and output the result image.

[0108] Step S3 is as follows:

[0109] Binning window weights are assigned based on Gaussian function, and clustering binning is performed after weight assignment based on the number of signal points, gray values ​​and assigned weights in the binning window, and the resulting image is output.

[0110] Binning window weight allocation based on Gaussian function includes the following steps:

[0111] S31. Generate Gaussian weights based on the current gain value and the size of the Binning window.

[0112] Step S31 includes the following steps:

[0113] S311. Create a Gaussian function based on the Binning window size and the number of signal points:

[0114]

[0115] p∈[1,bin_win 2 ];

[0116] Where w represents the Gaussian function value, p represents the number of signal points in the window, bin_win represents the size of the Binning window, and gauss_para represents the preset adjustment coefficient;

[0117] In this embodiment, the adjustment coefficient gauss_para is set to 20.

[0118] S312. Normalize the Gaussian function values:

[0119]

[0120] p∈[1,bin_win2 ];

[0121] Where w_normalize represents the Gaussian function value after normalization;

[0122] S313. Generate Gaussian weights:

[0123]

[0124] Where weight_gauss represents Gaussian weights, and ε represents the preset intensity coefficient.

[0125] In this embodiment, the Gaussian weights vary with the number of signal points in the Binning window and the current shooting gain; the intensity coefficient ε can be modified according to different cameras, and in this embodiment ε is 0.1.

[0126] S32. Generate window weights based on the number of signal points in the Binning window;

[0127] The window weights in step S32 are specifically generated according to the following formula:

[0128]

[0129] Where weight_win represents the window weight, p represents the number of signal points in the window, and bin_win represents the size of the Binning window.

[0130] S33. Generate the weight of each signal point in the Binning window based on the Gaussian weight and the window weight;

[0131] The weights of each signal point in the Binning window are generated based on Gaussian weights and window weights as follows:

[0132] weight=(1-weight_gauss(bin_win 2 +1-p))*weight_win;

[0133] Where weight_win represents the window weight, weight_gauss represents the Gaussian weight, p represents the number of signal points in the window, and bin_win represents the size of the Binning window.

[0134] The clustering and binning operation is performed based on the number of signal points, grayscale values, and assigned weights in the binning window.

[0135] Choose different clustering binning methods based on the number of signal points in the binning window:

[0136] In this embodiment, taking a 4*4 window as an example, the number of signal points in the window is counted according to the steps above, and different clustering binning methods are selected according to the different numbers.

[0137] If the number of signal points in the Binning window is 0, meaning the Binning window contains only dark noise values, then the grayscale value after Binning clustering is the product of the median of the window data and the total number of pixels in the window:

[0138] value_bin=median_value*bin_win 2 ;

[0139] Where median_value represents the median value of the window data, and bin_win represents the size of the Binning window.

[0140] In this embodiment, bin_win equals 4.

[0141] If the number of signal points in the Binning window is not 0, to ensure that the signal points are not overwhelmed by noise after Binning in a 4*4 window, the gray value of the signal point and the weighted sum of the corresponding weight of each signal point are used as the gray value after clustering and Binning.

[0142]

[0143] Here, value_signal is the grayscale value corresponding to each signal point in the window.

[0144] In this embodiment, the output after the clustering binning is completed is the cumulative value of a 4*4 window. If the mean value needs to be output, it can be directly divided by the total number of pixels in the window, which is 16.

[0145] The following are the signal-to-noise ratio (SNR) data obtained using the conventional Binning method and the clustering Binning method. Compared with conventional Binning, clustering Binning can significantly improve the image SNR.

[0146]

[0147]

[0148] Please refer to Figure 2 Embodiment two of the present invention is as follows:

[0149] A low-light image clustering binning terminal 1 based on Gaussian function includes a processor 2, a memory 3, and a computer program stored in the memory 3 and executable on the processor 2. When the processor 2 executes the computer program, it implements the steps in the low-light image clustering binning method based on Gaussian function described above.

[0150] In summary, the low-light image clustering binning method and terminal provided by this invention can create a dark-field noise value feature function based on dark-field images under different exposure times and gains. This allows for better identification of signal points and dark-field noise points within the binning window, and clustering of these points significantly improves the detection capability of weak signals, ensuring that even extremely weak signal points are not overwhelmed by noise. Furthermore, the Gaussian function-based binning window weight allocation reduces the weight of noise values, suppressing dark-field noise points within the window while enhancing signal points, thereby improving the signal-to-noise ratio of low-light images. This method is simple to implement, computationally inexpensive, and has high application potential.

[0151] This invention proposes a clustering binning approach, which enhances the signal while suppressing the dark noise value in the binning window by clustering signal points and dark noise points, thus greatly improving the signal-to-noise ratio of low-light images.

[0152] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A Binning method for low-light image clustering based on Gaussian functions, characterized in that, Including the following steps: S1. Obtain dark field images under different exposure times and gains, and establish a dark field noise value feature function based on the dark field images; S2. The signal points and dark field noise points in the low-light image to be processed are identified and clustered using the dark field noise value feature function. S3. Perform a clustering (Binning) operation on the low-light image based on the recognition and clustering results, and output the result image.

2. The Binning method for low-light image clustering based on Gaussian functions according to claim 1, characterized in that, Step S1 includes the following steps: S11. Obtain dark field images under different exposure times and gains; S12. Calculate the mean gray value of each pixel in each dark field image under different exposure times, fit the gray value-exposure time curve, and obtain the first characteristic function of the dark field noise value of each pixel as a function of exposure time: dark_value_exp(row,col)=k(row,col)*exp_time+b(row,col); Where, dark_value_exp represents the dark field noise value of each pixel under different exposure times, row / col represents the number of rows / columns of the noise point in the image, k represents the slope of the equation for each pixel changing with exposure time, exp_time represents the exposure time, and b represents the intercept of the equation for each pixel changing with exposure time. S13. Calculate the mean gray value of each pixel in each dark field image under different gains, fit the gray value-gain curve, and obtain the second characteristic function of the mean dark field noise of the image as a function of gain: Where, dark_value_gain represents the mean value of dark field noise under different gains, a1, a2, a3, a4 represent the parameters of the exponential equation, and gain represents the gain value; S14. Based on the first feature function and the second feature function, solve for the dark field noise value feature function of each pixel: Where, dark_value represents the dark field noise value.

3. The low-light image clustering binning method based on Gaussian function according to claim 1, characterized in that, Step S2 is as follows: Based on the relationship between the grayscale value of each pixel in the Binning window of the low-light image and the dark field noise value, signal points and dark field noise points are identified and clustered. The criteria for judging signal points are as follows: value_cur(i,j)>dark_value(i,j)+offset; i,j∈[1,bin_win]; Where value_cur(i,j) represents the gray value of the pixel in the i-th row and j-th column of the Binning window of the low-light image, dark_value() represents the feature function of the dark field noise value, bin_win represents the size of the Binning window, and the number of rows and columns of the window are the same, and offset represents the preset offset amount.

4. The low-light image clustering binning method based on Gaussian function according to claim 1, characterized in that, Step S3 is as follows: Binning window weights are assigned based on Gaussian function, and clustering binning is performed after weight assignment based on the number of signal points, gray values, and assigned weights in the binning window, outputting the resulting image.

5. The low-light image clustering binning method based on Gaussian function according to claim 4, characterized in that, Binning window weight allocation based on Gaussian function includes the following steps: S31. Generate Gaussian weights based on the current gain value and the size of the Binning window. S32. Generate window weights based on the number of signal points in the Binning window; S33. Generate the weight of each signal point in the Binning window based on the Gaussian weight and the window weight.

6. The low-light image clustering binning method based on Gaussian function according to claim 5, characterized in that, Step S31 includes the following steps: S311. Create a Gaussian function based on the Binning window size and the number of signal points: p∈[1,bin_win 2 ]; Where w represents the Gaussian function value, p represents the number of signal points in the window, bin_win represents the size of the Binning window, and gauss_para represents the preset adjustment coefficient; S312. Normalize the Gaussian function values: p∈[1,bin_win 2 ]; Where w_normalize represents the Gaussian function value after normalization; S313. Generate Gaussian weights: Where weight_gauss represents Gaussian weights, and ε represents the preset intensity coefficient.

7. The Binning method for low-light image clustering based on Gaussian functions according to claim 5, characterized in that, The window weights in step S32 are specifically generated according to the following formula: Where weight_win represents the window weight, p represents the number of signal points in the window, and bin_win represents the size of the Binning window.

8. The low-light image clustering binning method based on Gaussian function according to claim 5, characterized in that, The weights of each signal point in the Binning window are generated based on Gaussian weights and window weights as follows: weight=(1-weight_gauss(bin_win 2 +1-p))*weight_win; Where weight_win represents the window weight, weight_gauss represents the Gaussian weight, p represents the number of signal points in the window, and bin_win represents the size of the Binning window.

9. The Binning method for low-light image clustering based on Gaussian functions according to claim 4, characterized in that, The clustering and binning operation is performed based on the number of signal points, grayscale values, and assigned weights in the binning window. Choose different clustering binning methods based on the number of signal points in the binning window: If the number of signal points in the Binning window is 0, then the grayscale value after Binning clustering is the product of the median of the window data and the total number of pixels in the window: value_bin=median_value*bin_win 2 ; Where median_value represents the median value of the window data, and bin_win represents the size of the Binning window; If the number of signal points in the Binning window is not 0, then the weighted sum of the gray values ​​of the signal points and the weights corresponding to each signal point is used as the gray values ​​after clustering and Binning. Here, value_signal is the grayscale value corresponding to each signal point in the window.

10. A low-light image clustering binning terminal based on a Gaussian function, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps in the weak light image clustering binning method based on Gaussian function as described in any one of claims 1-9.

Images