A method for low-light image enhancement that is easily implemented in hardware

Abstract

This invention provides a low-light image enhancement method that is easy to implement in hardware. The method combines point transformation with spatial domain processing for low-light enhancement. A transformation curve is obtained by limiting histogram equalization. Then, the weight W1 of each point is calculated based on the brightness estimate pixelY1; the weight W2 of each point is calculated based on the detail information estimate pixelY2; the final mapping weight W is calculated using W1 and W2; finally, the final result is calculated based on the transformation curve: outI. i，j =w×(L i，j -I i，j )+I i，j , among which, I i，j L represents the grayscale value of the pixel at position i,j in the source image. i，j The gray value I at this point i，j The result after curve transformation processing. Based on a study of existing mainstream low-light image enhancement algorithms, a low-light enhancement method combining point transformation and spatial domain processing is proposed. This method solves problems such as poor adaptability and excessive noise after brightness stretching in point transformation-based algorithms, and the high hardware implementation cost of spatial domain processing algorithms.

Description

Technical Field

[0001] This invention relates to the field of hardware technology, and in particular to a method for low-light image enhancement that is easy to implement in hardware. Background Technology

[0002] In existing technologies, low-light image enhancement mainly involves analyzing the existing information in the source image and enhancing the Y channel of the source image based on the analysis results.

[0003] Traditional low-light enhancement algorithms can be mainly divided into three categories: the first category is point transformation, which only depends on the value of the pixel itself and is independent of its spatial location; the second category is spatial domain processing, which considers neighborhood information during processing; and the third category is transform domain processing, which processes the information in the transform domain of the image within a certain transform domain and then obtains the enhanced image through inverse transformation.

[0004] The main processing methods include grayscale transformation, histogram processing, and filtering; the known technologies involved in this invention include histogram equalization, mean filtering, bilateral filtering, and variance.

[0005] In addition, the commonly used terms in the prior art are as follows:

[0006] Low-light images: When taking photos or monitoring, insufficient light can make the image too dark, resulting in low-light images with insufficient brightness.

[0007] Characteristics of low-light images: low brightness, low contrast, high noise, and few visible details.

[0008] Low-light image enhancement: Processes low-light images to improve brightness and contrast, making more details visible without introducing excessive noise, thus improving image quality and enhancing visual effects.

[0009] Source image: refers to the image to be processed.

[0010] Y channel: Y channel information in YUV format.

[0011] Scene brightness: Scene brightness here refers to a value calculated by this algorithm by statistically analyzing the pixel values ​​of each pixel in the image.

[0012] However, the current low-light image enhancement algorithms have the following problems:

[0013] 1. Algorithms based on point transformation are simple and have few parameters, but due to their rigid form, they suffer from poor adaptability and unsatisfactory results when dealing with most scenarios.

[0014] 2. Spatial domain processing algorithms are implemented by analyzing the overall pixel information and local pixel information of the source image, so they are highly adaptable. However, the process is cumbersome and has too many parameters, which consumes a lot of resources for hardware implementation.

[0015] 3. The transformation domain processing algorithm is relatively complex in its analysis process, has poor robustness, and is not conducive to hardware implementation. Summary of the Invention

[0016] To address the aforementioned problems, the purpose of this application is to propose a low-light enhancement method that combines point transformation and spatial domain processing, based on a study of existing mainstream low-light image enhancement algorithms, thereby solving the following issues:

[0017] 1. Problems such as poor adaptability and excessive noise after brightness stretching exist in point transformation-based algorithms;

[0018] 2. Issues such as the high hardware implementation cost in spatial domain processing algorithms.

[0019] Specifically, this invention provides a low-light image enhancement method that is easy to implement in hardware. The method combines point transformation with spatial domain processing for low-light enhancement. A transformation curve is obtained through restricted histogram equalization. Then, the weight W1 of each point is calculated based on the brightness estimate pixelY1; the weight W2 of each point is calculated based on the detail information estimate pixelY2; the final mapping weight W is calculated using W1 and W2; finally, the final result is calculated based on the transformation curve: outI. i,j =w×(L i,j -I i,j )+I i,j , among which, I i，j L represents the grayscale value of the pixel located at position i,j in the source image. i，j The gray value I at this point i，j The result after curve transformation (L) i,j -I i,j ) represents the brightness change value at points i and j.

[0020] The method includes the following steps:

[0021] S1, Input source image; Obtain Y channel information in YUV format;

[0022] S2, calculate scene brightness, where scene brightness refers to a value calculated by the algorithm by statistically analyzing the pixel values ​​of each pixel in the image;

[0023] S3, perform histogram statistics, and limit histogram equalization according to the scene brightness to obtain the transformation curve;

[0024] S4, perform mean filtering, block brightness, point brightness estimation, and obtain weight W1;

[0025] S5, perform bilateral filtering, block detail estimation, point detail estimation, and obtain another weight W2;

[0026] S6, obtain the final weight W based on W1 and W2;

[0027] S7, obtain the final result based on the change curve and weight W.

[0028] The method further includes:

[0029] S1. Obtain Y channel information from the source image:

[0030] Y channel information is obtained through the RGB to YUV conversion formula;

[0031] S2. Determining scene brightness:

[0032] The average brightness of the Y channel is calculated using the following formula:

[0033]

[0034] S3. Perform histogram statistics on the Y channel and perform histogram equalization. Then, based on the scene brightness judgment result aveY obtained in step S2, obtain the transformation curve denoted as: y=f(x), where x is the input pixel value and y is the output result.

[0035] S4. Perform mean filtering on the Y channel to obtain the mean filtering result meanY. The purpose is to remove the influence of noise on the brightness estimate. Next, divide the filtered result meanY into blocks, calculate the brightness mean lightY within each block, and then perform bilinear interpolation to obtain the brightness estimate pixelY1 for each point. Calculate the weight W1 for each point based on the brightness estimate pixelY1.

[0036] S5. Perform bilateral filtering on the Y channel to obtain the bilateral filtering result bilfY. The purpose is to remove noise while retaining more detail information. Next, divide the bilateral filtering result bilfY into blocks, count the detail information detailY within each block, and then perform bilinear interpolation to obtain the detail information estimate pixelY2 for each point. Based on the detail information estimate pixelY2 for each point, calculate another weight W2 for each point.

[0037] S6. Then, according to the formula:

[0038] w = w1 α *w2 β

[0039] Calculate the final mapping weight W; where α and β are two factors that determine the efficiency of W1 and W2 in calculating W, with a default value of 1.

[0040] S7. Finally, according to the transformation curve: y = f(x), that is: L i，j =f(I i，j );

[0041] Among them, I i，j L represents the grayscale value of the pixel located at position i,j in the source image. i，j The gray value I at this point i，j The result after curve transformation;

[0042] The final calculation result is:

[0043] outI i，j =w×(L i，j -I i，j )+I i，j

[0044] Among them, (L) i，j -I i，j ) represents the brightness change value of points i and j.

[0045] In step S3, the histogram equalization constraint is performed as follows:

[0046] (1) The limiting strength limS is calculated based on aveY:

[0047]

[0048] (2) The histogram equalization is constrained according to the constraint strength limS obtained in (1).

[0049] The conversion formula is based on BT601-6, i.e., the forward and inverse conversion formula between YUV and RGB, as follows:

[0050] T=[0.299,0.587,0.114;-0.1687,-0.3313,0.5;0.5,-0.4187,-0.0813];

[0051] Y=T(1,1)*R+T(1,2)*G+T(1,3)*B;

[0052] U=T(2,1)*R+T(2,2)*G+T(2,3)*B;

[0053] V=T(3,1)*R+T(3,2)*G+T(3,3)*B;

[0054] Ti=[1.0,0,1.402;1.0,-0.344,-0.714;1.0,1.772,0];

[0055] R=Ti(1,1)*Y+Ti(1,2)*(U-128)+Ti(1,3)*(V-128);

[0056] G=Ti(2,1)*Y+Ti(2,2)*(U-128)+Ti(2,3)*(V-128);

[0057] B=Ti(3,1)*Y+Ti(3,2)*(U-128)+Ti(3,3)*(V-128).

[0058] Therefore, the advantages of this application are: the algorithmic implementation of the method makes the analysis process simpler and easier, and more conducive to hardware implementation, solving the problem of large hardware implementation burden in spatial domain processing algorithms; and solving the problems of poor adaptability and excessive noise after brightness stretching in existing point transformation-based algorithms. Attached Figure Description

[0059] The accompanying drawings, which are provided to further illustrate the invention and form part of this application, are not intended to limit the scope of the invention.

[0060] Figure 1 This is a schematic diagram of the main steps involved in the method of this application.

[0061] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0062] To better understand the technical content and advantages of the present invention, the present invention will now be described in further detail with reference to the accompanying drawings.

[0063] like Figure 1 As shown, this application relates to a low-light image enhancement method that is easy to implement in hardware, and the main steps are as follows:

[0064] Input source image;

[0065] Y channel, Y channel information in YUV format;

[0066] Scene brightness (here, scene brightness refers to a value calculated by this algorithm by statistically analyzing the pixel values ​​of each pixel in the image);

[0067] Histogram statistics are performed, and histogram equalization is limited based on scene brightness, and the curve is transformed.

[0068] Constrain histogram equalization and transform the curve;

[0069] Mean filtering, block brightness, point brightness estimation, W1;

[0070] Bilateral filtering, block detail, point detail estimation, W2.

[0071] like Figure 2 As shown, the main steps are described below:

[0072] S1. Obtain Y channel information from the source image:

[0073] The Y channel information is obtained through the RGB to YUV conversion formula;

[0074] S2. Determining scene brightness:

[0075] The average brightness of the Y channel is calculated using the following formula:

[0076]

[0077] S3. Perform histogram statistics on the Y channel and then perform histogram equalization. Based on the scene brightness judgment result aveY obtained in step S2, perform histogram equalization constraints as follows:

[0078] (1) The limiting strength limS is calculated based on aveY:

[0079]

[0080] (2) The histogram equalization is constrained according to the constraint strength limS obtained in (1).

[0081] The resulting transformation curve is denoted as: y = f(x), where x is the input pixel value and y is the output result;

[0082] S4. Perform mean filtering on the Y channel to obtain the mean filtering result meanY. The purpose is to remove the influence of noise on the brightness estimate. Next, divide the filtered result meanY into blocks, calculate the brightness mean lightY within each block, and then perform bilinear interpolation to obtain the brightness estimate pixelY1 for each point. Calculate the weight W1 for each point based on the brightness estimate pixelY1.

[0083] S5. Perform bilateral filtering on the Y channel to obtain the bilateral filtering result bilfY. The purpose is to remove noise while retaining more detailed information. Next, divide the bilateral filtering result bilfY into blocks, count the detailed information detailY within each block, and then perform bilinear interpolation to obtain the estimated value of the detailed information pixelY2 for each point. Calculate another weight W2 for each point based on the estimated value of the detailed information pixelY2.

[0084] S6. Then, according to the formula:

[0085] w = w1 α *w2 β

[0086] Calculate the final mapping weight W; where α and β are two factors that determine the efficiency of W1 and W2 in calculating W, with a default value of 1.

[0087] S7. Finally, based on the transformation curve:

[0088] y = f(x)

[0089] Right now:

[0090] L i，j =f(l i，j )

[0091] Among them, I i，j L represents the grayscale value of the pixel located at position i,j in the source image. i，j The gray value I at this point i，j The result after curve transformation;

[0092] The final calculation result is:

[0093] outI i，j =w×(L i，j -I i，j )+I i，j

[0094] Among them, (L) i，j -I i，j ) represents the brightness change value of points i and j.

[0095] The conversion formula mentioned in step S3 is based on BT601-6, which is the forward and inverse conversion formula between YUV and RGB, as follows:

[0096] T=[0.299,0.587,0.114;-0.1687,-0.3313,0.5;0.5,-0.4187,-0.0813];

[0097] Y=T(1,1)*R+T(1,2)*G+T(1,3)*B;

[0098] U=T(2,1)*R+T(2,2)*G+T(2,3)*B;

[0099] V=T(3,1)*R+T(3,2)*G+T(3,3)*B;

[0100] Ti=[1.0,0,1.402;1.0,-0.344,-0.714;1.0,1.772,0];

[0101] R=Ti(1,1)*Y+Ti(1,2)*(U-128)+Ti(1,3)*(V-128);

[0102] G=Ti(2,1)*Y+Ti(2,2)*(U-128)+Ti(2,3)*(V-128);

[0103] B=Ti(3,1)*Y+Ti(3,2)*(U-128)+Ti(3,3)*(V-128).

[0104] In summary, this algorithm obtains the transformation curve by restricting histogram equalization; calculates the weight W1 of each point based on the brightness estimate pixelY1; calculates the weight W2 of each point based on the detail information estimate pixelY2; obtains the final weight W based on W1 and W2; and finally obtains the final result through the transformation curve.

[0105] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for low-light image enhancement that is easy to implement in hardware, characterized in that, The method is a low-light enhancement that combines point transformation with spatial processing. The transformation curve is obtained by limiting histogram equalization, and the weight W1 of each point is calculated based on the brightness estimate pixelY1 of each point. The weight W2 for each point is calculated based on the estimated pixelY2 of the detailed information for each point; the final mapping weight W is calculated using W1 and W2; finally, the final result is calculated based on the transformation curve. ,in, In the source image, the position is at The grayscale value of a pixel. The grayscale value of that point The result after curve transformation. This represents the brightness change value at points i and j; The method includes the following steps: S1, Input source image; Obtain Y channel information in YUV format; Specifically, obtain Y channel information from source image: Obtain Y channel information through RGB to YUV conversion formula; S2, calculate scene brightness. Scene brightness refers to a value calculated by this algorithm by statistically analyzing the pixel values ​​of each pixel in the image. Specifically, the scene brightness is determined as follows: The average brightness of the Y channel is calculated using the following formula: ； S3, perform histogram statistics and limit histogram equalization according to the scene brightness to obtain the transformation curve; specifically, perform histogram statistics on the Y channel and perform histogram equalization, and then obtain the transformation curve based on the scene brightness judgment result aveY obtained in step S2, denoted as: , where x is the input pixel value and y is the output result; S4. Perform mean filtering, block brightness estimation, and point brightness estimation to obtain weight W1. Specifically, perform mean filtering on the Y channel to obtain the mean filtering result meanY, which aims to remove the influence of noise on the brightness estimation value. Next, divide the filtered result meanY into blocks, calculate the mean brightness value lightY within each block, and then perform bilinear interpolation to obtain the brightness estimation value pixelY1 for each point. Calculate the weight W1 for each point based on the brightness estimation value pixelY1. S5, perform bilateral filtering, block detail estimation, and point detail estimation to obtain another weight W2; specifically, perform bilateral filtering on the Y channel to obtain the bilateral filtering result bilfY, the purpose of which is to remove noise while retaining more detail information. Next, divide the bilateral filtering result bilfY into blocks, count the detail information detailY within each block, and then perform bilinear interpolation to obtain the detail information estimate pixelY2 for each point; calculate another weight W2 for each point based on the detail information estimate pixelY2 for each point; S6, obtain the final weight W based on W1 and W2; specifically, then, according to the formula: ； Calculate the final mapping weight W; where , These two values ​​determine the efficiency of W1 and W2 in calculating W; the default value is 1. S7, obtain the final result based on the transformation curve and weight W; specifically, finally, based on the transformation curve: Right now: ; in, In the source image, the position is at The grayscale value of a pixel. The grayscale value of that point The result after curve transformation; The final calculation result is: ； in, This represents the brightness change value at points i and j.

2. The low-light image enhancement method that is easy to implement in hardware according to claim 1, characterized in that, In step S3, the histogram equalization constraint is performed as follows: (1) The limiting strength limS is calculated based on aveY: ； (2) The histogram equalization is constrained by the constraint strength limS obtained in (1).

3. The low-light image enhancement method that is easy to implement in hardware according to claim 1, characterized in that, The conversion formula is based on BT601-6, i.e., the forward and inverse conversion formula between YUV and RGB, as follows: ； ； ； ； ； ； ； ； ； ； ； 。

Images