Image noise reduction method, device and equipment based on pixel difference value

Abstract

The invention relates to the field of image noise reduction, in particular to an image noise reduction method, device and equipment based on a pixel difference value, and the method comprises the steps: obtaining a target Gaussian function; determining a corresponding independent variable maximum value and a corresponding target fitting function according to the target Gaussian function; determining a median of the independent variable according to the maximum value of the independent variable; calculating a median corresponding point corresponding to the target fitting function at the median of the independent variable; taking the median corresponding point as a center, and performing central symmetry on the part, greater than the independent variable median, of the target fitting function to obtain a correction function segment; combining the modified function segment with the part, greater than the median of the independent variable, of the target fitting function to obtain a target calculation function; in binary calculation, calculating a pixel difference weight of a to-be-processed pixel through the target calculation function by using a shift method; according to the pixel difference value weight, the pixel value of the pixel to be processed after filtering is obtained, the hardware requirement is less, and more circuit area is saved.

Description

Technical Field

[0001] The present invention relates to the field of image noise reduction, and in particular to an image noise reduction method, device and equipment based on pixel difference. Background Art

[0002] The pixel difference denoising method is widely used in various image denoising techniques, such as the bilateral filtering algorithm. Bilateral filtering is widely used in the fields of image denoising, edge preservation, texture enhancement, etc. It can effectively smooth the image while maintaining the image's detail information, making the image look more natural and clear. Specifically, bilateral filtering uses a filter, which consists of two parts: a spatial weight function and a pixel value difference weight function. In the present invention, the focus is on improving the shortcomings of the pixel value difference weight function. The idea of ​​the pixel difference weight function is to first segment or classify the neighborhood to be used for filtering according to the pixel value, and then give the category to which the point belongs a relatively high weight. Pixels with higher difference have lower weights, and pixels with lower difference have higher weights. A Gaussian kernel function is used, and the standard deviations σ of the two kernel functions are different. Finally, the filter will use the product of the two weights as the total weight to obtain the pixel value after bilateral filtering.

[0003] In terms of hardware implementation, the circuit to implement the exponential function requires mathematical operation circuits and logic circuits. Floating-point logic can be used in Verilog to implement the exponential function, but this will take up a lot of chip resources. Therefore, in order to effectively implement the exponential function while not taking up too much circuit resources, numerical approximation techniques, such as Taylor series or polynomial approximation methods, can be used. According to the actual results, the pixel value difference weight filtering algorithm in the prior art selects a piecewise linear function to fit the exponential function, specifically based on the difference between the center pixel and the surrounding points in the window, by shifting and adding the difference, and then making up the value of the corresponding point of the exponential function. In this way, each difference situation requires a corresponding plurality of large-bit-width adders, and if the accuracy needs to be enhanced, the difference range needs to be increased, and the number of adders used will also increase exponentially. The specific schematic diagram is shown in FIG. Figure 1 shown.

[0004] It can be seen that the hardware implementation method of fitting the Gaussian exponential function with a piecewise linear function in the prior art will perform shift addition according to different differences, and requires a large-bit-width selector and adder to implement it in hardware. If the difference range increases, the number of adders used will increase exponentially, wasting a large hardware area and occupying more hardware resources. Therefore, how to reduce the high computing power occupied and the corresponding hardware space occupied during the calculation process of fitting the Gaussian function is an urgent problem to be solved by those skilled in the art. Summary of the invention

[0005] The purpose of the present invention is to provide an image denoising method, device and equipment based on pixel difference, so as to solve the problem that pixel difference denoising occupies large computing power and has high hardware requirements in the prior art.

[0006] In order to solve the above technical problems, the present invention provides an image denoising method based on pixel difference, comprising:

[0007] Get the target Gaussian function;

[0008] Determine the corresponding maximum value of the independent variable and the corresponding target fitting function according to the target Gaussian function; the target fitting function is a gradually decreasing power function with a base of 2;

[0009] Determine the median value of the independent variable according to the maximum value of the independent variable;

[0010] Calculate the median corresponding point of the target fitting function at the median of the independent variable;

[0011] Taking the median corresponding point as the center, the part of the target fitting function that is larger than the median of the independent variable is centrally symmetric to obtain a modified function segment;

[0012] Combining the correction function segment with the portion of the target fitting function that is greater than the median value of the independent variable to obtain a target calculation function;

[0013] In binary calculation, the pixel difference weight of the pixel to be processed is calculated by using the target calculation function using the shift method;

[0014] According to the pixel difference weight, the pixel value of the pixel to be processed after filtering is obtained.

[0015] Optionally, in the image denoising method based on pixel difference, determining the corresponding maximum value of the independent variable and the corresponding target fitting function according to the target Gaussian function includes:

[0016] Determine the maximum value of a plurality of independent variables according to the target Gaussian function;

[0017] Determine the corresponding target fitting function according to the maximum value of the multiple independent variables;

[0018] Accordingly, determining the median value of the independent variable according to the maximum value of the independent variable includes:

[0019] Determine the corresponding median values ​​of the independent variables according to the maximum values ​​of the plurality of independent variables;

[0020] Accordingly, the step of calculating the median corresponding point of the target fitting function at the median of the independent variable includes:

[0021] Calculate a plurality of corresponding points of the median of the target fitting function at the median of each independent variable;

[0022] Accordingly, taking the median corresponding point as the center, the part of the target fitting function that is greater than the median of the independent variable is centrally symmetric, and the correction function segment obtained includes:

[0023] Taking the median corresponding point as the center, the parts of the plurality of target fitting functions that are greater than the median of the independent variable are centrally symmetric to obtain correction function segments corresponding to the respective target fitting functions;

[0024] Accordingly, the step of combining the correction function segment with a portion of the target fitting function that is greater than the median value of the independent variable to obtain a target calculation function comprises:

[0025] Combining the correction function segment with a portion of the corresponding target fitting function that is greater than the corresponding independent variable median value to obtain an alternative calculation function corresponding to the target fitting function;

[0026] According to the target Gaussian function, a target calculation function is selected from a plurality of candidate calculation functions.

[0027] Optionally, in the image denoising method based on pixel difference, determining a corresponding target fitting function according to the target Gaussian function includes:

[0028] According to the target Gaussian function, the target fitting function as follows is determined:

[0029] y=2 k+m-n ;

[0030] Wherein, k is the coefficient of the target fitting function, m is the maximum value of the independent variable, and n is the independent variable.

[0031] Optionally, in the image denoising method based on pixel difference, combining the correction function segment with a portion of the target fitting function that is greater than the median value of the independent variable to obtain a target calculation function comprises:

[0032] The correction function segment is combined with the portion of the target fitting function that is greater than the median value of the independent variable to obtain the target calculation function of the following formula:

[0033] y=2 j -2 k+n n≤m / 2;

[0034] y=2 k+m-n n>m / 2;

[0035] Among them, k is the coefficient of the target calculation function, m is the maximum value of the independent variable, n is the independent variable, and j is the exponent of the normalization coefficient.

[0036] Optionally, in the image denoising method based on pixel difference, the exponent of the normalization coefficient is not less than 4.

[0037] Optionally, in the image denoising method based on pixel difference, in the binary calculation, using the shift method, by using the target calculation function, calculating the pixel difference weight of the pixel to be processed includes:

[0038] Obtaining a pixel difference between a pixel value of a pixel to be processed and a pixel value of a corresponding central pixel;

[0039] Substitute the pixel difference into the corresponding target calculation function and determine the pixel difference weight by the following formula:

[0040] Weight = 2 j -2 k < <diff diff≤m / 2;

[0041] Weight = 2 k <<(m-diff) diff>m / 2;

[0042] Among them, k is the coefficient of the target calculation function, j is the exponent of the normalization coefficient, diff is the pixel difference, Weight is the pixel difference weight, and << is the left shift sign.

[0043] An image noise reduction device based on pixel difference, comprising:

[0044] An acquisition module, used to acquire a target Gaussian function;

[0045] A Gaussian analysis module, used to determine the corresponding maximum value of the independent variable and the corresponding target fitting function according to the target Gaussian function; the target fitting function is a gradually decreasing power function with a base of 2;

[0046] An independent variable median module, used to determine the independent variable median according to the independent variable maximum value;

[0047] A median corresponding point module, used to calculate the median corresponding point of the target fitting function corresponding to the median of the independent variable;

[0048] A symmetric module is used to take the median corresponding point as the center and perform central symmetry on the part of the target fitting function that is larger than the median of the independent variable to obtain a modified function segment;

[0049] A target combination module, used for combining the correction function segment with a portion of the target fitting function that is greater than the median value of the independent variable to obtain a target calculation function;

[0050] A weight module, used to calculate the pixel difference weight of the pixel to be processed by the target calculation function using a shift method in binary calculation;

[0051] The filtering module is used to obtain the pixel value of the pixel to be processed after filtering according to the pixel difference weight.

[0052] An image noise reduction device based on pixel difference, comprising a processor and a computing circuit;

[0053] The processor is used to implement the steps of any one of the above-mentioned methods for image denoising based on pixel difference when executing the computer program;

[0054] The calculation circuit includes a plurality of pixel difference branches;

[0055] Each of the pixel difference branches includes a shifter and an addition calculation component.

[0056] Optionally, in the image denoising device based on pixel difference, each of the pixel difference branches includes a shifter and an adder.

[0057] Optionally, in the image denoising device based on pixel difference, the maximum value of the independent variable is 2 p ; Wherein, p is an integer greater than 2;

[0058] Each of the pixel difference branches includes a shifter, an accumulator and an inverter.

[0059] Optionally, in the image denoising device based on pixel difference, the maximum value of the independent variable is 8.

[0060] The present invention provides an image denoising method based on pixel difference, which obtains a target Gaussian function; determines the corresponding maximum value of an independent variable and a corresponding target fitting function according to the target Gaussian function; the target fitting function is a gradually reduced power function with base 2; determines the median of the independent variable according to the maximum value of the independent variable; calculates the median corresponding point corresponding to the target fitting function at the median of the independent variable; takes the median corresponding point as the center, makes a central symmetry for the part of the target fitting function that is greater than the median of the independent variable to obtain a correction function segment; combines the correction function segment with the part of the target fitting function that is greater than the median of the independent variable to obtain a target calculation function; in binary calculation, uses a shift method to calculate the pixel difference weight of a pixel to be processed through the target calculation function; and obtains the pixel value of the pixel to be processed after filtering according to the pixel difference weight.

[0061] The present invention optimizes and improves the pixel difference weight algorithm with a power function with base 2, and the weight value of the result is highly similar to the original Gaussian weight function. While ensuring the accuracy of the calculation, it also greatly reduces the demand for computing power and hardware. With the increase of the difference situation, that is, the increase of the accuracy, the amount of hardware required to be added is also significantly less than the existing technology, saving a lot of resources. At the same time, less hardware requirements also save more circuit area, which is more conducive to the miniaturization of the device and broadens the versatility of the present invention. The present invention also provides an image denoising device and equipment based on pixel difference with the above beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0063] Figure 1 It is a structural schematic diagram of an image noise reduction device based on pixel difference in the prior art;

[0064] Figure 2 A flowchart of a specific implementation of the image denoising method based on pixel difference provided by the present invention;

[0065] Figures 3 to 6 A schematic diagram of function comparison of a specific implementation of the image denoising method based on pixel difference provided by the present invention;

[0066] Figure 7 A flowchart of another specific implementation of the image denoising method based on pixel difference provided by the present invention;

[0067] Figure 8 A schematic diagram of function comparison of another specific implementation of the image denoising method based on pixel difference provided by the present invention;

[0068] Fig. 9 A schematic structural diagram of a specific implementation of the image denoising device based on pixel difference provided by the present invention;

[0069] Fig.10 , Fig.11 The present invention is a schematic structural diagram of a specific implementation of the image denoising device based on pixel difference provided by the present invention.

[0070] In the figure, 100-includes acquisition module, 200-Gaussian analysis module, 300-independent variable median, 400-median corresponding point, 500-symmetry module, 600-target combination module, 700-weight module, and 800-filtering module. DETAILED DESCRIPTION

[0071] In order to enable those skilled in the art to better understand the scheme of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0072] The core of the present invention is to provide an image denoising method based on pixel difference, a flowchart of a specific implementation method thereof is shown as follows: Figure 2 As shown, it is called specific implementation mode 1, including:

[0073] S101: Obtain a target Gaussian function.

[0074] The target Gaussian function is the target function mentioned above that needs to be fitted and approximated. In the image denoising method based on pixel difference, the image corresponding to the monotonically decreasing interval of the Gaussian function is usually used, and the present invention is no exception. Furthermore, for the convenience of calculation, the starting point of the target fitting function below corresponds to the symmetry axis of the target Gaussian function.

[0075] S102: Determine the corresponding maximum value of the independent variable and the corresponding target fitting function according to the target Gaussian function; the target fitting function is a gradually decreasing power function with a base of 2.

[0076] In the above, the target fitting function of the present invention can be considered as a function that fits the target Gaussian function as much as possible. The target Gaussian function can be simplified as The derivative is Please refer to Figure 3 , Figure 3 The solid line in the middle is the target Gaussian function, and the dotted line is the derivative of the target Gaussian function. It can be seen that the weight first "accelerates and decreases" with the pixel difference, and the slope is the largest when X=0.7 (corresponding function value Y=-0.8577). Then it "decelerates and decreases" and finally converges to 0.

[0077] This function feature means:

[0078] 1) When the pixel difference is too small, the weight is high;

[0079] 2) As the difference increases, the weight decreases rapidly;

[0080] 3) When the difference is large enough, the weight becomes very small, so that it decreases more and more slowly, and finally the weight is close to 0.

[0081] In order to adapt to the characteristic of hardware using binary to calculate 0 and 1, the present invention uses a power function with base 2 to fit the Gaussian function to model the pixel difference weight, rather than using a linear function in other technical methods.

[0082] Combination Figure 3 , the maximum value of the independent variable means that when the independent variable of the target Gaussian function is the maximum value of the independent variable, the y value corresponding to the target Gaussian function can be considered to be non-existent, or infinitely close to 0. Since the target Gaussian function is a function with a limit of 0, the maximum value of the independent variable can be set according to the actual situation.

[0083] As a specific implementation, determining a corresponding target fitting function according to the target Gaussian function includes:

[0084] According to the target Gaussian function, the target fitting function of the following formula (1) is determined:

[0085] y=2 k+m-n ; (1)

[0086] Wherein, k is the coefficient of the target fitting function, m is the maximum value of the independent variable, and n is the independent variable.

[0087] In this specific implementation, the form of the target fitting function is specifically limited, which is still a power function with base 2, but the sign in front of the independent variable n is changed to a sign, so that the power function shows a downward trend as a whole, which is consistent with the target Gaussian function, and the maximum value of the independent variable m specifies the domain of the target fitting function, and k is used as the coefficient of the target fitting function to adjust the shape of the power function to make it closer to the target Gaussian function. This specific implementation simply expresses the target fitting function, while ensuring accuracy and reducing the amount of calculation.

[0088] S103: Determine the median value of the independent variable according to the maximum value of the independent variable.

[0089] The median of the independent variable is also the median of the maximum value of the independent variable, and half of the maximum value of the independent variable can be directly taken.

[0090] S104: Calculate the median corresponding point of the target fitting function corresponding to the median of the independent variable.

[0091] That is, the function value corresponding to the target fitting function is obtained when the independent variable is the median value of the independent variable. This point can be considered as the midpoint of the target fitting function.

[0092] It should be noted that, since we have determined the maximum value of the independent variable, that is, the maximum value of the independent variable of the target Gaussian function, in step S102, the value range of the independent variable of the target Gaussian function is specified. Similarly, the target fitting function only considers images within the same value range of the independent variable, that is, the maximum value of the independent variable also applies to the target fitting function.

[0093] S105: Taking the median corresponding point as the center, the part of the target fitting function that is greater than the median of the independent variable is centrally symmetrical to obtain a modified function segment.

[0094] Please combine Figure 4 understand, Figure 4 For the power function y = 2 k 2 m-n With the target Gaussian function The dotted line is the power function and the solid line is the target Gaussian function. It can be seen that in the second half of the function, the images of the two functions gradually approach each other, that is, when the independent variable exceeds a certain value ( Figure 4 After the value of the two functions is almost the same, Figure 5 for Figure 4 The enlarged schematic diagram of the part where the independent variable is greater than 5 shows that the two functions are almost identical. Therefore, the part of the target Gaussian function after the independent variable is greater than a certain value can be replaced by a power function. Combined with the previous text, the target Gaussian function of the part that can be replaced by the power function has a function change that is first fast and then slow, and the function change rate of the power function corresponding to this part is also first fast and then slow. The previous text points out that the overall function change rate of the target Gaussian function is first slow, then fast, and then slow. The midpoint of the target Gaussian function can be regarded as the point where the change rate changes from fast to slow (of course, the actual situation may not be the case, but it is sufficient as a fitting condition). It is only necessary to make the second half of the target fitting function symmetrical with the corresponding point of the median as the symmetry center, and the function change rate of the obtained symmetrical part (that is, the corrected function segment) also becomes slow at first and then fast, and its function value and trend are both matched with the first half of the target Gaussian function.

[0095] S106: Combine the correction function segment with the portion of the target fitting function that is greater than the median value of the independent variable to obtain a target calculation function.

[0096] Continuing from the above, in this step, the function segment with central symmetry in step S105 is combined with the function segment obtained by central symmetry to obtain a target calculation function whose function change rate is also first slow, then fast, and then slow again, which matches the properties of the target Gaussian function, see Figure 6 , Figure 6is a schematic diagram comparing the target Gaussian function and the target calculation function, wherein the dotted line is the target calculation function and the solid line is the target Gaussian function. So far, we have obtained a suitable power function to fit the target Gaussian function.

[0097] As a specific implementation, this step includes:

[0098] The correction function segment is combined with the part of the target fitting function that is greater than the median value of the independent variable to obtain the target calculation function shown in the following equations (2) and (3):

[0099] y=2 j -2 k+n n≤m / 2; (2)

[0100] y=2 k+m-n n>m / 2; (3)

[0101] Among them, k is the coefficient of the target calculation function, m is the maximum value of the independent variable, n is the independent variable, and j is the exponent of the normalization coefficient.

[0102] Formula (2) and formula (3) represent two segments of the target calculation function, respectively. Formula (3) is the same as the target fitting function, and formula (2) is the expression of the deformed function segment through the central symmetry, that is, the modified function segment. Of course, this power function (that is, the target calculation function) is about the point (m / 2, 2 k+m / 2 )symmetry.

[0103] Since the function needs to be continuous at the point between the two segments, the midpoint of the independent variable is 2 j -2 k+m / 2 =2 k+m / 2 , we can get formula (4):

[0104] k+m / 2=j-1; (4)

[0105] Wherein, m and j are both preset, and the value of k can be calculated by formula (4).

[0106] This specific implementation provides a representation of the target calculation function and ensures lower computing power usage.

[0107] Preferably, the exponent of the normalization coefficient is not less than 4. It is worth noting that in the selection of the normalization coefficient of the Gaussian function on the hardware, selecting a larger normalization coefficient can make the range of the data wider, thereby better retaining the information of the data. For example, if the range of the original data is large, selecting a larger normalization coefficient can map the data to a wider range, avoiding the data from being compressed into a smaller range, thereby better retaining the difference of the data. Therefore, in this preferred embodiment, the exponent of the normalization coefficient is not less than 4, which ensures the difference of subsequent data and improves the calculation accuracy.

[0108] S107: In binary calculation, a shift method is used to calculate the pixel difference weight of the pixel to be processed through the target calculation function.

[0109] As a specific implementation, this step includes:

[0110] A1: Obtain the pixel difference between the pixel value of the pixel to be processed and the pixel value of the corresponding central pixel.

[0111] This step obtains the pixel difference, which can be used to subsequently calculate the weight represented by the difference.

[0112] A2: Substitute the pixel difference into the corresponding target calculation function and determine the pixel difference weight by the following formula (5) (6):

[0113] Weight = 2 j -2 k < <diff diff≤m / 2; (5)

[0114] Weight = 2 k <<(m-diff) diff>m / 2; (6)

[0115] Among them, k is the coefficient of the target calculation function, j is the exponent of the normalization coefficient, diff is the pixel difference, Weight is the pixel difference weight, and << is the left shift sign.

[0116] The above formulas (5) and (6) can better show the simplification of the calculation process by using a power function with base 2. The exponent of the multiplied part is directly converted into an operation on the number of bits of binary data. The pixel difference here is the independent variable n of the target calculation function mentioned above. It can be seen that it is not necessary to calculate the actual value of the power function with base 2 in the calculation. Only a simple calculation and then shifting are required.

[0117] The following is an actual example. k+n k = 2, n = 3, then 2 k+n For 2 5, we get 32, which is converted to binary as 10000. According to the rules of Equation (5) and Equation (6), 2 k+n can also be written as 2 k <<n. Substituting k = 2 and n = 3, we get 2 2 <<3, 2 2 is 4, whose binary representation is 100. Shifting it 3 bits to the left gives exactly 100000, which is also 32 in decimal, the same as the result of normal calculation.

[0118] The target Gaussian function is fitted using a power function with base 2, and the traditional multi-segment linear function fitting is abandoned. The biggest advantage is that in actual electrical devices for operations, binary calculation is mostly used. When calculating a power function with base 2, it is not necessary to actually substitute decimal numbers for calculation. Instead, the calculation can be completed by simply shifting binary numbers, greatly reducing the requirements for computing power and hardware and simplifying the circuit.

[0119] S108: Obtain the filtered pixel value of the pixel to be processed according to the pixel difference weight.

[0120] After obtaining a relatively accurate pixel difference weight according to the foregoing steps, any noise reduction method applicable to the pixel difference weight function, such as the bilateral filtering algorithm, can be used to process the image pixels, which will not be elaborated here.

[0121] A method for image noise reduction based on pixel difference provided by the present invention includes: obtaining a target Gaussian function; determining the corresponding maximum value of the independent variable and the corresponding target fitting function according to the target Gaussian function; the target fitting function is a power function with base 2 that gradually decreases; determining the median value of the independent variable according to the maximum value of the independent variable; calculating the corresponding median corresponding point of the target fitting function at the median value of the independent variable; taking the median corresponding point as the center, making central symmetry for the part of the target fitting function greater than the median value of the independent variable to obtain a modified function segment; combining the modified function segment with the part of the target fitting function greater than the median value of the independent variable to obtain a target calculation function; in binary calculation, using the shifting method, calculating the pixel difference weight of the pixel to be processed through the target calculation function; obtaining the filtered pixel value of the pixel to be processed according to the pixel difference weight. The present invention optimizes and improves the pixel difference weight algorithm using a power function with base 2. The obtained weight value is highly similar to the original Gaussian weight function. While ensuring the calculation accuracy, it also greatly reduces the requirements for computing power and hardware. Moreover, as the number of difference situations increases, that is, the accuracy increases, the number of additional hardware required is significantly less than that of the prior art, saving a large amount of resources. At the same time, fewer hardware requirements also save more circuit area, are more conducive to the miniaturization of devices, and broaden the versatility of the present invention.

[0122] On the basis of the first specific implementation mode, the method for determining the target calculation function is further modified to obtain the second specific implementation mode, the flow chart of which is as follows: Figure 7 As shown, including:

[0123] S201: Obtain a target Gaussian function.

[0124] S202: Determine the maximum values ​​of multiple independent variables according to the target Gaussian function.

[0125] S203: determining a corresponding target fitting function according to the maximum values ​​of the plurality of independent variables; the target fitting function is a gradually decreasing power function with a base of 2.

[0126] As can be seen from the foregoing, the k value in the target fitting function is determined by the maximum value m of the independent variable and the exponent j of the normalization coefficient. Therefore, this step obtains multiple target fitting functions that correspond one-to-one to the maximum values ​​of the independent variables.

[0127] S204: Determine the corresponding median values ​​of the independent variables according to the plurality of maximum values ​​of the independent variables.

[0128] Since there are multiple maximum values ​​of the independent variables, these maximum values ​​of the independent variables can be divided by 2 to obtain multiple median values ​​of the independent variables.

[0129] So far, multiple groups of combinations corresponding to the maximum values ​​of independent variables, median values ​​of independent variables, and target fitting functions have been obtained.

[0130] S205: Calculate a plurality of median corresponding points of the target fitting function corresponding to the median of each independent variable.

[0131] Calculating the value of each target fitting function at the midpoint of the corresponding independent variable actually takes the midpoint of the target fitting function as the symmetry center of the subsequent steps.

[0132] S206: Taking the median corresponding point as the center, perform central symmetry on the parts of the plurality of target fitting functions that are greater than the median of the independent variable to obtain a correction function segment corresponding to each target fitting function.

[0133] After the multiple target fitting functions are centrally symmetric, multiple correction function segments are obtained.

[0134] S207: Combine the correction function segment with the portion of the corresponding target fitting function that is greater than the corresponding independent variable median value to obtain an alternative calculation function corresponding to the target fitting function.

[0135] After combining a plurality of correction function segments with the corresponding function segments of the target fitting function, the starting points and end points of a plurality of alternative calculation functions are approximately the same as the target Gaussian function, and the function change rates are also the same (slow-fast-slow), but the trajectories are slightly different, providing alternative options for subsequent selection.

[0136] S208: According to the target Gaussian function, a target calculation function is selected from a plurality of candidate calculation functions.

[0137] As mentioned above, in step S207, several candidate calculation functions are obtained. The candidate calculation function closest to the target Gaussian function can be determined as the target calculation function through image comparison. The image schematic diagram of the plurality of candidate calculation functions and the target Gaussian function is shown in FIG. Figure 8 As shown, the dashed lines of different styles represent different candidate calculation functions, and the solid line represents the target Gaussian function.

[0138] S209: In binary calculation, the pixel difference weight of the pixel to be processed is calculated by using the target calculation function using the shift method.

[0139] S210: Obtaining the pixel value of the pixel to be processed after filtering according to the pixel difference weight.

[0140] In this specific implementation, a plurality of candidate calculation functions are generated, and then a target calculation function closest to the target Gaussian function is further selected from these candidate calculation functions. The remaining steps are the same as the above specific implementation and will not be repeated here.

[0141] In this preferred embodiment, different maximum values ​​of the independent variables are selected for the target Gaussian function, thereby obtaining different alternative calculation functions with slightly different trajectories. The staff can further select the alternative calculation function whose trajectory is closer to the target Gaussian function as the target calculation function to further improve the calculation accuracy of the pixel difference weight and improve the noise reduction filtering effect.

[0142] The following is an introduction to an image denoising device based on pixel difference provided by an embodiment of the present invention. The image denoising device based on pixel difference described below and the image denoising method based on pixel difference described above can be referred to each other.

[0143] Fig. 9 The structural block diagram of the image denoising device based on pixel difference provided by the embodiment of the present invention is shown in FIG. Fig. 9 The image noise reduction device based on pixel difference may include:

[0144] An acquisition module 100 is used to acquire a target Gaussian function;

[0145] A Gaussian analysis module 200 is used to determine the corresponding maximum value of the independent variable and the corresponding target fitting function according to the target Gaussian function; the target fitting function is a gradually decreasing power function with a base of 2;

[0146] An independent variable median module 300, used to determine the independent variable median according to the independent variable maximum value;

[0147] A median corresponding point module 400 is used to calculate the median corresponding point of the target fitting function corresponding to the median of the independent variable;

[0148] A symmetric module 500 is used to make a central symmetry of the part of the target fitting function that is greater than the median of the independent variable with the median corresponding point as the center to obtain a modified function segment;

[0149] A target combination module 600 is used to combine the correction function segment with the part of the target fitting function that is greater than the median value of the independent variable to obtain a target calculation function;

[0150] The weight module 700 is used to calculate the pixel difference weight of the pixel to be processed by the target calculation function using the shift method in binary calculation;

[0151] The filtering module 800 is used to obtain the pixel value of the pixel to be processed after filtering according to the pixel difference weight.

[0152] As a preferred implementation, the Gaussian analysis module 200 includes:

[0153] A complex independent variable maximum value unit, used to determine the maximum value of the complex independent variables according to the target Gaussian function;

[0154] A complex fitting unit, used for determining a corresponding target fitting function according to the maximum value of the complex independent variables;

[0155] Accordingly, the independent variable median module 300 includes:

[0156] A complex independent variable median unit, used to determine the corresponding independent variable medians according to the maximum values ​​of the complex independent variables;

[0157] Accordingly, the median corresponding point module 400 includes:

[0158] A complex midpoint unit, used to calculate the corresponding points of the multiple target fitting functions at the median of each independent variable;

[0159] Accordingly, the symmetric module 500 includes:

[0160] A complex symmetry unit is used to take the median corresponding point as the center, perform central symmetry on the parts of the plurality of target fitting functions that are greater than the median of the independent variable, and obtain a correction function segment corresponding to each of the target fitting functions;

[0161] Accordingly, the target combination module 600 includes:

[0162] An alternative unit is used to combine the correction function segment with a portion of the corresponding target fitting function that is greater than the corresponding independent variable median value to obtain an alternative calculation function corresponding to the target fitting function;

[0163] The alternative selection unit is used to select and determine a target calculation function from a plurality of the alternative calculation functions according to the target Gaussian function.

[0164] As a preferred implementation, the Gaussian analysis module 200 includes:

[0165] A power function fitting unit is used to determine a target fitting function of the following formula according to the target Gaussian function:

[0166] y=2 k+m-n ;

[0167] Wherein, k is the coefficient of the target fitting function, m is the maximum value of the independent variable, and n is the independent variable.

[0168] As a preferred implementation, the target combination module 600 includes:

[0169] A combining unit is used to combine the correction function segment with the part of the target fitting function that is greater than the median value of the independent variable to obtain the target calculation function of the following formula:

[0170] y=2 j -2 k+n n≤m / 2;

[0171] y=2 k+m-n n>m / 2;

[0172] Among them, k is the coefficient of the target calculation function, m is the maximum value of the independent variable, n is the independent variable, and j is the exponent of the normalization coefficient.

[0173] As a preferred implementation, the weight module 700 includes:

[0174] A difference acquisition unit, used to acquire a pixel difference between a pixel value of a pixel to be processed and a pixel value of a corresponding central pixel;

[0175] The difference weight unit is used to substitute the pixel difference into the corresponding target calculation function and determine the pixel difference weight by the following formula:

[0176] Weight = 2 j -2 k < <diff diff≤m / 2;

[0177] Weight = 2 k <<(m-diff) diff>m / 2;

[0178] Among them, k is the coefficient of the target calculation function, j is the exponent of the normalization coefficient, diff is the pixel difference, and Weight is the pixel difference weight.

[0179] The present invention provides an image denoising device based on pixel difference, which is used to obtain a target Gaussian function through an acquisition module 100; a Gaussian analysis module 200, which is used to determine the corresponding independent variable maximum value and the corresponding target fitting function according to the target Gaussian function; the target fitting function is a gradually reduced power function with a base of 2; an independent variable median module 300, which is used to determine the independent variable median according to the independent variable maximum value; a median corresponding point module 400, which is used to calculate the median corresponding point corresponding to the target fitting function at the independent variable median; a symmetry module 500, which is used to take the median corresponding point as the center, and make a central symmetry for the part of the target fitting function greater than the independent variable median to obtain a correction function segment; a target combination module 600, which is used to combine the correction function segment with the part of the target fitting function greater than the independent variable median to obtain a target calculation function; a weight module 700, which is used to calculate the pixel difference weight of the pixel to be processed by the target calculation function in binary calculation using a shift method; and a filtering module 800, which is used to obtain the pixel value of the pixel to be processed after filtering according to the pixel difference weight. The present invention optimizes and improves the pixel difference weight algorithm with a power function with base 2. The weight value of the result is highly similar to the original Gaussian weight function. While ensuring the calculation accuracy, it also greatly reduces the demand for computing power and hardware. Moreover, as the difference situation increases, that is, the accuracy increases, the amount of hardware required to be added is also significantly smaller than the existing technology, saving a lot of resources. At the same time, less hardware requirements also save more circuit area, which is more conducive to the miniaturization of devices, and broadens the versatility of the present invention.

[0180] The image denoising device based on pixel difference of the present embodiment is used to implement the aforementioned image denoising method based on pixel difference. Therefore, the specific implementation of the image denoising device based on pixel difference can be seen in the embodiment of the image denoising method based on pixel difference in the previous text. For example, the acquisition module 100, the Gaussian analysis module 200, the median of the independent variable 300, the median corresponding point 400, the symmetry module 500, the target combination module 600, the weight module 700, and the filtering module 800 are respectively used to implement steps S101, S102, S103, S104, S105, S106, S107, and S108 in the aforementioned image denoising method based on pixel difference. Therefore, its specific implementation can refer to the description of the corresponding embodiments of each part, which will not be repeated here.

[0181] The present invention also provides an image noise reduction device based on pixel difference, a structural schematic diagram of a specific implementation method thereof is shown in FIG. Fig.10 , Fig.11 As shown, it includes a processor and a computing circuit;

[0182] The processor is used to implement the steps of any one of the above-mentioned methods for image denoising based on pixel difference when executing the computer program;

[0183] The calculation circuit includes a plurality of pixel difference branches 10;

[0184] Each of the pixel difference branches 10 includes a shifter 11 and an addition calculation component.

[0185] The image denoising device based on pixel difference provided by the present invention is to implement the image denoising method based on pixel difference described in the foregoing text. For specific technical details, please refer to the foregoing text and will not be elaborated here.

[0186] As a preferred implementation, each of the pixel difference branches 10 includes a shifter 11 and an adder 12, such as Fig.10 shown.

[0187] Please refer to the above formula (5) and formula (6), the pixel difference weight is calculated by the shift method, which only requires one shift and one subtraction (the subtraction can be implemented by the adder 12). Therefore, for a pixel difference situation (corresponding to a pixel difference branch 10), only one adder 12 and one shifter 11 are needed to calculate the pixel difference weight, which can be compared with Figure 1 The pixel difference branch in the prior art is compared.

[0188] As a preferred implementation, the maximum value of the independent variable is 2 p , where p is an integer greater than 2;

[0189] Each of the pixel difference branches 10 includes a shifter 11 , an accumulator 13 and an inverter 14 .

[0190] like Fig.11 As shown, since the integer powers of 2 are all represented in binary as numbers with the first digit being 1 and the last digits being all 0, when performing subtraction calculations, only one accumulator, one inverter, and one shifter are needed to implement it, which greatly reduces the hardware overhead of the pixel difference branch.

[0191] As a specific implementation, the maximum value of the independent variable is 8;

[0193] Since the binary representation of the decimal 8 is 1000, when the maximum value m of the independent variable is 8, the m-diff calculation process of reference formula (6) can be directly regarded as the inversion of diff plus one. Therefore, only one accumulator 13 (responsible for adding one) and one inverter 14 are needed to realize the function of the adder 12 in the previous preferred embodiment. The hardware overhead and computing power consumption of one accumulator 13 and one inverter 14 are less than that of a single adder 12. 8 matches the number of channels of currently commonly used hardware and has good versatility. Therefore, this preferred embodiment can be regarded as further reducing computing power consumption and hardware overhead, and improving computing efficiency.

[0194] The present invention provides an image denoising device based on pixel difference, comprising a processor and a computing circuit; the processor is used to implement the steps of the image denoising method based on pixel difference as described in any one of the above when executing the computer program; the computing circuit comprises a plurality of pixel difference branches 10; each of the pixel difference branches 10 comprises a shifter 11 and an addition computing component. The present invention optimizes and improves the pixel difference weight algorithm with a power function with base 2, and the weight value of the result is highly similar to the original Gaussian weight function. While ensuring the accuracy of the calculation, it also greatly reduces the demand for computing power and hardware. Moreover, as the number of difference cases increases, that is, the accuracy increases, the amount of hardware required to be added is also significantly less than that of the prior art, saving a lot of resources. At the same time, less hardware requirements also save more circuit area, which is more conducive to the miniaturization of the device and broadens the versatility of the present invention.

[0195] The following is an example of a process in actual production, including:

[0196] The present invention uses the above method to find a suitable power function to fit the Gaussian function, which is as follows: first determine the normalization coefficient 2 jand the independent variable range of the Gaussian function (0, m), and then use the above formula to get the coefficient k, and finally get the fitted power function. It is worth noting that m is not a constant. Its meaning is that if n>m, the y value can be considered negligible, then take this m value as the maximum value. Therefore, when selecting a power function, you can use several groups of m for comparison and select the value closest to the Gaussian function. Figure 8 , the power function from left to right, m ​​is 6, 8, 10, 12, respectively, about the point (3, 2 11 ),(4,2 11 ), (52 11 ), (6, 2 11 ) are symmetrical. For these four curves, it is observed that m=8 is closest to the original Gaussian curve, so m=8 is selected as the power function to be fitted.

[0197] Specifically for the YUV domain denoising module, the present invention mainly optimizes its difference weight calculation. The difference weight of YUV denoising is a Gaussian filter operator. Other methods use linear functions for approximation. The present invention uses the above-mentioned power function method with base 2 for approximation. Figure 8 , select m=8 as the maximum value of pixel difference, that is, if the difference is greater than 9, the weight is considered to be 0. The specific implementation steps are as follows:

[0198] 1. Determine the form of the power function. Take m = 8, j = 12, and calculate k = 7 according to the above formula. The form of the power function is:

[0199] Weight = 4096-2 7 < <diff when diff≤4;

[0200] Weight = 2 7 <<(8-diff) when diff>4;

[0201] 2. Combined with the power function in step 1, determine the hardware implementation in the YUV domain.

[0202] a) Take out a sliding window of n*n (n pixels in the horizontal and vertical directions respectively) for filtering. For the input image data, take out a sliding window of n*n on the image with each pixel as the center for filtering. This requires storing n-1 rows of image data in SRAM (static random access memory) and writing hardware code for taking n*n numbers in the current window in hardware description language (such as Verilog).

[0203] b) Calculate the absolute value of the difference between the n*n-1 numbers in the window and the center point. Because the number of points in the window is n*n-1, the data is relatively large. In hardware, this step can be implemented by first writing a simple difference module composed of a judgment and a subtractor, and then instantiating it n*n-1 times to synchronously calculate the difference of all points in the window.

[0204] c) Calculate the difference weight according to the absolute value of the difference obtained in step b). This step uses the power function with base 2 of the present invention for fitting. According to step 1, firstly, the pixel difference and 4 are judged. After the judgment, if the difference is less than 4, then 2 7 Left shift diff bit, here directly fill the low bit with 0, no resource consumption, for example: diff = 1, shift to get {2 7 ,0}; 4096 minus a number is equivalent to adding the complement of the negative number. The high bit of the original code of the negative number is directly inverted and the last bit is added by 1. The resources consumed are equivalent to some inverters 14 and an accumulator 13. If the difference is greater than 4,8-diff, the highest bit (third bit) of diff can be directly set to 1, and then the high bit of the original code of the negative number is inverted and the last bit is added by 1 to get the complement, 2 7 Shifting left also does not consume resources. The final difference weight is obtained. Fig.11 shown.

[0205] d) The difference weight is multiplied by the spatial domain weight to obtain the final weight, and the pixel value after noise reduction is obtained through weighted sum.

[0206] This matches the hardware characteristics, and the hardware implementation is simple and efficient. Moreover, the power function with base 2 of the present invention has the same function properties as the Gaussian kernel function in terms of monotonicity, rate of change and other function properties, and the weight value is almost the same, thereby effectively improving the computing efficiency of the hardware and saving hardware resources.

[0207] In summary, the present invention optimizes and improves the pixel difference weight algorithm with a power function with a base of 2, invents a modeling method and corresponding hardware implementation, and the weight value of the result is similar to the original Gaussian weight function. Compared with the original method, the advantages of the present invention are: 1. If the number of difference cases is m, the hardware resources of the present invention only use m inverters 14 and accumulators 13 of corresponding bit widths to realize the calculation of the weight value. The original method requires several adders 12 for each difference case, which saves significant area in comparison; 2. If the accuracy requirements of the image processing chip are increased, for example, the number of difference cases is increased by 5 cases on the basis of m, then the present invention only needs to add 5 inverters 14 and accumulators 13 of corresponding bit widths, and the original method needs to add 5 cases, each of which requires several adders 12. The resource occupancy increases exponentially with the increase of the accuracy requirements, while the present invention increases linearly, thereby saving a lot of resources.

[0208] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the image denoising method based on pixel difference as described above are implemented. The image denoising method based on pixel difference provided by the present invention obtains a target Gaussian function; determines the corresponding maximum value of the independent variable and the corresponding target fitting function according to the target Gaussian function; the target fitting function is a gradually reduced power function with a base of 2; determines the median of the independent variable according to the maximum value of the independent variable; calculates the corresponding point of the median corresponding to the target fitting function at the median of the independent variable; takes the corresponding point of the median as the center, makes a central symmetry for the part of the target fitting function greater than the median of the independent variable, and obtains a correction function segment; combines the correction function segment with the part of the target fitting function greater than the median of the independent variable to obtain a target calculation function; in binary calculation, uses the shift method to calculate the pixel difference weight of the pixel to be processed through the target calculation function; obtains the pixel value of the pixel to be processed after filtering according to the pixel difference weight. The present invention optimizes and improves the pixel difference weight algorithm with a power function with base 2. The weight value of the result is highly similar to the original Gaussian weight function. While ensuring the calculation accuracy, it also greatly reduces the demand for computing power and hardware. Moreover, as the difference situation increases, that is, the accuracy increases, the amount of hardware required to be added is also significantly smaller than the existing technology, saving a lot of resources. At the same time, less hardware requirements also save more circuit area, which is more conducive to the miniaturization of devices, and broadens the versatility of the present invention.

[0209] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0210] It should be noted that, in this specification, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0211] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0212] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0213] The above is a detailed introduction to the image denoising method, device and apparatus based on pixel difference provided by the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. An image denoising method based on pixel difference, characterized in that: include: Get the target Gaussian function; Determine the corresponding maximum value of the independent variable and the corresponding target fitting function according to the target Gaussian function; the target fitting function is a gradually decreasing power function with a base of 2; Determine the median value of the independent variable according to the maximum value of the independent variable; Calculate the median corresponding point of the target fitting function at the median of the independent variable; Taking the median corresponding point as the center, the part of the target fitting function that is larger than the median of the independent variable is centrally symmetric to obtain a modified function segment; Combining the correction function segment with the portion of the target fitting function that is greater than the median value of the independent variable to obtain a target calculation function; In binary calculation, the pixel difference weight of the pixel to be processed is calculated by using the target calculation function using the shift method; According to the pixel difference weight, the pixel value of the pixel to be processed after filtering is obtained.

2. The image denoising method based on pixel difference according to claim 1, characterized in that: The determining of the corresponding independent variable maximum value and the corresponding target fitting function according to the target Gaussian function includes: Determine the maximum value of a plurality of independent variables according to the target Gaussian function; Determine the corresponding target fitting function according to the maximum value of the multiple independent variables; Accordingly, determining the median value of the independent variable according to the maximum value of the independent variable includes: Determine the corresponding median values ​​of the independent variables according to the maximum values ​​of the plurality of independent variables; Accordingly, the step of calculating the median corresponding point of the target fitting function at the median of the independent variable includes: Calculate a plurality of corresponding points of the median of the target fitting function at the median of each independent variable; Accordingly, taking the median corresponding point as the center, the part of the target fitting function that is greater than the median of the independent variable is centrally symmetric, and the correction function segment obtained includes: Taking the median corresponding point as the center, the parts of the plurality of target fitting functions that are greater than the median of the independent variable are centrally symmetric to obtain correction function segments corresponding to the respective target fitting functions; Accordingly, the step of combining the correction function segment with a portion of the target fitting function that is greater than the median value of the independent variable to obtain a target calculation function comprises: Combining the correction function segment with a portion of the corresponding target fitting function that is greater than the corresponding independent variable median value to obtain an alternative calculation function corresponding to the target fitting function; According to the target Gaussian function, a target calculation function is selected from a plurality of candidate calculation functions.

3. The image denoising method based on pixel difference according to claim 1, characterized in that: Determining the corresponding target fitting function according to the target Gaussian function includes: According to the target Gaussian function, the target fitting function as follows is determined: y＝2 k+m-n ； Wherein, k is the coefficient of the target fitting function, m is the maximum value of the independent variable, and n is the independent variable.

4. The image denoising method based on pixel difference according to claim 1, characterized in that: The step of combining the correction function segment with the portion of the target fitting function that is greater than the median value of the independent variable to obtain the target calculation function comprises: The correction function segment is combined with the portion of the target fitting function that is greater than the median value of the independent variable to obtain the target calculation function of the following formula: and=2 j -2 k+n n≤m / 2; y=2 k+m-n n>m / 2; Among them, k is the coefficient of the target calculation function, m is the maximum value of the independent variable, n is the independent variable, and j is the exponent of the normalization coefficient.

5. The image denoising method based on pixel difference according to claim 4, characterized in that: The exponent of the normalization coefficient is not less than 4.

6. The image denoising method based on pixel difference according to claim 4, characterized in that: In the binary calculation, using the shift method, by means of the target calculation function, the pixel difference weight of the pixel to be processed is calculated, which includes: Obtaining a pixel difference between a pixel value of a pixel to be processed and a pixel value of a corresponding central pixel; Substitute the pixel difference into the corresponding target calculation function and determine the pixel difference weight by the following formula: Weight=2 j -2 k <<diff diff≤m / 2; Weight=2 k <<(m-diff)diff>m / 2; Among them, k is the coefficient of the target calculation function, j is the exponent of the normalization coefficient, diff is the pixel difference, Weight is the pixel difference weight, and << is the left shift sign.

7. An image denoising device based on pixel difference, characterized in that: include: An acquisition module, used to acquire a target Gaussian function; A Gaussian analysis module, used to determine the corresponding maximum value of the independent variable and the corresponding target fitting function according to the target Gaussian function; the target fitting function is a gradually decreasing power function with a base of 2; An independent variable median module, used to determine the independent variable median according to the independent variable maximum value; A median corresponding point module, used to calculate the median corresponding point of the target fitting function corresponding to the median of the independent variable; A symmetric module is used to take the median corresponding point as the center and perform central symmetry on the part of the target fitting function that is larger than the median of the independent variable to obtain a modified function segment; A target combination module, used for combining the correction function segment with a portion of the target fitting function that is greater than the median value of the independent variable to obtain a target calculation function; A weight module, used to calculate the pixel difference weight of the pixel to be processed by the target calculation function using a shift method in binary calculation; The filtering module is used to obtain the pixel value of the pixel to be processed after filtering according to the pixel difference weight.

8. An image noise reduction device based on pixel difference, characterized in that: including a processor and a computing circuit; The processor is configured to implement the steps of the image denoising method based on pixel difference according to any one of claims 1 to 6 when executing the computer program; The calculation circuit includes a plurality of pixel difference branches; Each of the pixel difference branches includes a shifter and an addition calculation component.

9. The image denoising device based on pixel difference as claimed in claim 8, characterized in that: Each of the pixel difference branches includes a shifter and an adder.

10. The image denoising device based on pixel difference according to claim 8, characterized in that: The maximum value of the independent variable is 2 p ; Wherein, p is an integer greater than 2; Each of the pixel difference branches includes a shifter, an accumulator and an inverter.

11. The image denoising device based on pixel difference according to claim 10, characterized in that: The maximum value of the independent variable is 8.

Images