A method for suppressing stripe noise based on a single image

Abstract

The present application relates to the technical field of image processing, and particularly relates to a stripe noise suppression method based on a single image. Direction-specific smoothing processing is performed on the input image to generate a direction-specific smoothed image, which facilitates subsequent data processing. The gradient of the input image is calculated to obtain a gradient image, the gradient image is binarized to generate a binary image, and the binary image is subjected to morphological processing to obtain an effective region image. The column mean values of the input image and the direction-specific smoothed image in the effective region image are extracted, the column mean values of the input image and the direction-specific smoothed image are differentiated to obtain stripe noise column estimation values, and a stripe noise estimation image is generated based on the stripe noise column estimation values, so that the stripe noise is efficiently and accurately extracted. The input image and the stripe noise estimation image are differentiated to obtain a denoised image in which the stripe noise is eliminated, so that the stripe noise is removed without damaging the edge detail information of the image, and the quality of the image is ensured.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically to a method for suppressing stripe noise in a single image. Background Technology

[0002] In image acquisition systems, due to differences in the operating characteristics of CCD devices on different readout lines, the photoelectric array receives the same radiation, resulting in different signals read from each row or column. This is the root cause of stripe noise. Stripe noise is characterized by roughly the same noise intensity in each row or column, but varying noise intensity between different columns or rows. This noise significantly impacts image quality and hinders subsequent image analysis. Therefore, there is an urgent need for a method to suppress or remove stripe noise from single images. Summary of the Invention

[0003] The purpose of this invention is to overcome the above-mentioned shortcomings and provide a method for suppressing stripe noise based on a single image. This method can extract stripe noise efficiently and accurately, remove stripe noise without destroying the edge details of the image, and ensure the quality of the denoised image.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] A method for suppressing stripe noise based on a single image includes the following steps:

[0006] S1. Perform orientation-specific smoothing on the input image I to generate an orientation-specific smoothed image I. s Calculate the gradient of the input image I to obtain the gradient image I. g For gradient image I g Perform binarization processing to generate a binary image I. a For binarized image I a Morphological processing is performed to obtain the effective region image I. b ;

[0007] S2. Extract the input image I and the orientation-specific smoothed image I s Image I in the effective region b The column mean and orientation-specific smoothing of the input image I. s The column mean of the input image I and the orientation-specific smoothed image I are used to calculate the column mean. s The column mean is differiated to obtain the stripe noise column estimate N_column, and the stripe noise estimated image Noise is generated based on the stripe noise column estimate N_column;

[0008] S3. Difference the input image I with the stripe noise estimation image Noise to obtain the denoised image with stripe noise removed.

[0009] All pixels of the input image I described in S1 need to be direction-specific median filtering processing.

[0010] S1 described in the calculation of the gradient of the input image I, the gradient image I g First, the input image I is edge detection, edge detection using Sobel operator as follows:

[0011]

[0012]

[0013] Where, G x is the vertical gradient operator, G y is the horizontal gradient operator, G x The image after spatial filtering is I gx , G y The image after spatial filtering is I gy , then the value of each pixel in the gradient image I g It can be expressed as follows:

[0014] S1 described in the gradient image I g Do binary processing to generate binary image I a The process is as follows:

[0015] Based on adaptive threshold k * The image is binarized, and the image is divided into two parts: flat area and detail area, threshold k * Is automatically generated by Otsu algorithm, based on the gradient threshold k * The binary image I a , the area with gradient less than threshold k * The value is flat area, assigned to 0, the area with gradient greater than and equal to threshold k * The value is the detail area, then assigned to 1.

[0016] The formula of Otsu algorithm is as follows:

[0017]

[0018]

[0019]

[0020]

[0021] Where, m G is the global mean of the image, m is the cumulative mean of the area with lower gray level divided by the gradient threshold k, Hi H is the i-th value in the normalized histogram. i Let L be the probability of a pixel with grayscale value i appearing, and L be the pixel level of the image.

[0022] S1 describes the binarization of image I a The morphological processing procedure is as follows:

[0023]

[0024] This formula represents dilating the input image I using structuring element B. a Effective region image I b For input image I a To obtain the result using B-dilation, the origin of the structuring element B is shifted to the input image I. a The position (,y) of B, if B is in the input image I a The position (,y) is at the same location as I a If the intersection of the two regions is not empty, then the effective region image I is output. b The corresponding position (,y) is assigned a value of 1, otherwise it is assigned a value of 0, and the structuring element B is:

[0025] S2 describes the extraction of input image I and orientation-specific smoothed image I. s Image I in the effective region b The column mean and orientation-specific smoothing of the input image I. s The column mean is as follows:

[0026]

[0027]

[0028] Among them, Ave I () represents the mean of the j-th column of the input image I, Ave Is () represents the orientation-specific smoothed image I. s The mean of the j-th column, 1≤j≤n, where m is the number of rows in the image and n is the number of columns in the image.

[0029] S2 describes smoothing the input image I by taking its column mean and orientation specificity. s The specific formula for differencing column means is: N_column(j) = Ave I ()-Ave Is (),1≤j≤n.

[0030] The stripe noise estimation image Noise generated based on the stripe noise column estimate N_column described in S2 is specifically: Noise(i,j)= column(j),1≤i≤m,1≤j≤n.

[0031] The formula described in S3 for obtaining a denoised image with stripe noise removed by subtracting the input image I from the stripe noise estimation image Noise is: I output =-Noise, where I output A denoised image to remove stripe noise.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0033] The method of this invention is as follows: Orientation-specific smoothing is applied to the input image I to generate an orientation-specific smoothed image I. s This facilitates subsequent data processing. The gradient of the input image I is calculated to obtain the gradient image I. g For gradient image I g Perform binarization processing to generate a binary image I. a For binarized image I a Morphological processing is performed to obtain the effective region image I. b Extract the input image I and the orientation-specific smoothed image I. s Image I in the effective region b The column mean and orientation-specific smoothing of the input image I. s The column mean of the input image I and the orientation-specific smoothed image I are used to calculate the column mean. s The column mean of the input image I is subtracted to obtain the estimated column value N_column of the stripe noise. Based on the estimated column value N_column, a stripe noise estimation image Noise is generated, thus efficiently and accurately extracting the stripe noise. The input image I is then subtracted from the stripe noise estimation image Noise to obtain a denoised image with stripe noise removed, thereby removing stripe noise without destroying the edge details of the image and ensuring image quality. Attached Figure Description

[0034] Figure 1 This is a flowchart of the stripe noise suppression method of the present invention;

[0035] Figure 2 Image I contains stripe noise;

[0036] Figure 3 This is the orientation-specific smoothing image I of the present invention. s ;

[0037] Figure 4 The effective region image I obtained after morphological processing according to the present invention b ;

[0038] Figure 5 This is the noise image of the present invention;

[0039] Figure 6 is the final result of the present application. DETAILED DESCRIPTION

[0040] The present application is further described below with reference to the accompanying drawings.

[0041] As Figure 1 shown is a flow chart of the method of the present application.

[0042] A method for suppressing stripe noise based on a single image, comprising the following steps:

[0043] S1. Direction-specific smoothing processing is performed on the input image I to generate a direction-specific smoothed image I s , facilitating subsequent data processing. As Figure 2 shown is an image I containing stripe noise according to the present embodiment, and as Figure 3 shown is a direction-specific smoothed image I s of the present application. S2. The gradient of the input image I is calculated to obtain a gradient image I g S3. Binary processing is performed on the gradient image I g to generate a binary image I a S4. Morphological processing is performed on the binary image I a to obtain an effective region image I b , the result of which is shown in Figure 4 .

[0044] Further, all pixels of the input image I need to be subjected to direction-specific median filtering processing.

[0045] Preferably, direction-specific smoothing processing is performed according to the direction of the stripe noise in the input image. In this embodiment, the input image I containing vertical stripes is taken as an example, and all pixels of the input image I are subjected to direction-specific median filtering processing. The size of the direction-specific median filtering window is [1, 2+1], and N is valued according to the thickness of the stripes, generally being 6, 7, or 8, which can achieve better results.

[0046] S2. The gradient of the input image I is calculated to obtain a gradient image I g Before this, edge detection is performed on the input image I, and the Sobel operator used for edge detection is as follows:

[0047]

[0048]

[0049] wherein G x is the vertical gradient operator, G y is the horizontal gradient operator, and G xThe image filtered in the spatial domain is I gx , G y The image filtered in the spatial domain is I gy The value of each pixel in the gradient image I g may be expressed by the following formula:

[0050]

[0051] Further, the gradient image I g is binarized to generate a binarized image I a The process is as follows:

[0052] Based on the adaptive threshold k * , the image is binarized, and the image is divided into two parts, a flat area and a detail area, the threshold k * is automatically generated by the Otsu algorithm, and the binarized image I * based on the gradient threshold k a , in which the area with a gradient lower than the threshold k * is a flat area, and is assigned a value of 0, and the area with a gradient greater than and equal to the threshold k * is a detail area, and is assigned a value of 1.

[0053] The Otsu algorithm calculation formula is as follows:

[0054]

[0055]

[0056]

[0057]

[0058] wherein m G is the global mean value of the image, m is the cumulative mean value of the area with a lower gray value divided by the gradient threshold k, H i is the i-th value in the normalized histogram, i.e. the probability of the pixel with a gray value of i, and L is the pixel level of the image.

[0059] The process of morphological processing on the binarized image I a is as follows:

[0060]

[0061] The formula indicates that the input image I a is dilated by the structural element B, the effective area image I b is the input image I a is the result obtained by dilating B, and the origin of the structural element B is shifted to the input image I aThe position (,y) of B, if B is in the input image I a The position (,y) is at the same location as I a The intersection of B and I is not empty, meaning the position of the element with a value of 1 in B corresponds to the position of the element with a value of 1 in B. a If at least one of the image values ​​is 1, then the output valid region image I is obtained. b The corresponding position (,y) is assigned a value of 1, otherwise it is assigned a value of 0, and the structuring element B is:

[0062] S2. Extract the input image I and the orientation-specific smoothed image I s Image I in the effective region b The column mean and orientation-specific smoothing of the input image I. s The column mean of the input image I and the orientation-specific smoothed image I are used to calculate the column mean. s The column mean is differiated to obtain the estimated value N_column of the striped noise column, as shown in the following figure. Figure 5 As shown, a stripe noise estimation image Noise is generated based on the stripe noise column estimate N_column, thereby efficiently and accurately extracting stripe noise.

[0063] Extracting input image I and orientation-specific smoothed image I s Image I in the effective region b The column mean and orientation-specific smoothing of the input image I. s The column mean is as follows:

[0064]

[0065]

[0066] Among them, Ave I () represents the mean of the j-th column of the input image I, Ave Is () represents the orientation-specific smoothed image I. s The mean of the j-th column, 1≤j≤n, where m is the number of rows in the image and n is the number of columns in the image.

[0067] Furthermore, the column mean and orientation-specific smoothing of input image I are applied to image I. s The specific formula for differencing column means is: N_column(j) = Ave I ()-Ave Is (),1≤j≤n.

[0068] Furthermore, the stripe noise estimation image Noise is generated based on the stripe noise column estimate N_column as follows: Noise(i,j) = column(j) ,1≤i≤m,1≤j≤n.

[0069] S3. The input image I is subtracted from the stripe noise estimation image Noise to obtain a denoised image without stripe noise, so that the stripe noise is removed without destroying the edge detail information of the image, and the quality of the denoised image is ensured, as shown in the following formula: Figure 6 The method effectively removes the stripe noise while well preserving the detail information of the image.

[0070] The formula for obtaining the denoised image without stripe noise by subtracting the input image I from the stripe noise estimation image Noise is as follows: output = -Noise, wherein I output is the denoised image without stripe noise.

[0071] Preferably, the present application performs direction-specific smoothing processing on the input image; performs Sobel gradient operator processing on the input image I and the direction-specific smoothed image I s ; distinguishes the edge and the smooth area; obtains the noise estimation value by column mean for the smooth area respectively; generates a noise image based on the noise estimation value; and subtracts the input image from the noise image to obtain the result after denoising. The method can efficiently and accurately extract the stripe noise, can remove the stripe noise without destroying the edge detail information of the image, and ensures the quality of the denoised image.

[0072] In summary, the above is only a preferred embodiment of the present application, and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for suppressing stripe noise based on a single image, characterized by, The method comprises the following steps: S1. performing directional specific smoothing on an input image to generate a directional specific smoothed image S2. calculating a gradient of the input image to generate a gradient image S3. performing binaryzation on the gradient image to generate a binary image S4. performing morphological processing on the binary image to generate an effective region image ; S2. Extract the input image With orientation-specific smoothing images Image in the effective region Input image within Column mean and orientation-specific smoothed image The column mean of the input image Column mean and orientation-specific smoothed image By differencing the column means, we obtain the estimated value of the stripe noise column. Based on stripe noise column estimates Generate stripe noise estimation image ; S2 said extracting the input image a direction-specific smoothed image in the effective region image the input image the column mean of the direction-specific smoothed image the column mean of the direction-specific smoothed image wherein is the mean of the first column of the input image is the mean of the first column of the direction-specific smoothed image , is the number of image rows, is the number of image columns;​​​​ S2 said to the input image column mean and direction-specific smoothed image The specific formula for differencing the column mean of ; S2 the stripe noise column estimation value generating a stripe noise estimation image Specifically, ; S3. Subtract the input image from the stripe noise estimate image to obtain a denoised image with stripe noise removed.

2. A method for suppressing the stripe noise based on a single image as claimed in claim 1, wherein, The input image of S1 All pixels of the image need to be direction-specifically median filtered.

3. A method for suppressing the stripe noise based on a single image as claimed in claim 1, wherein, S1 the computing input image gradient, resulting in a gradient image before the input image is edge detected, the Sobel operator used for the edge detection is as follows: wherein is a gradient operator in the vertical direction, is a gradient operator in the horizontal direction, the spatially filtered image is , the spatially filtered image is then the gradient image the value of each pixel in the gradient image can be expressed as .

4. A method for suppressing the stripe noise based on a single image as claimed in claim 1, wherein, S1 the gradient image S1 the gradient image The process is as follows: Based on adaptive threshold The image is binarized, and the image is divided into two parts of flat area and detail area Automatically generated by Otsu algorithm, based on gradient threshold Binarized image Among them, the area with gradient lower than the threshold The value of the flat area is 0, and the value of the area with gradient greater than and equal to the threshold The value of the detail area is 1.

5. A method of fringe noise suppression based on a single image as claimed in claim 4, wherein, The Otsu algorithm calculation formula is as follows: The Otsu algorithm calculation formula is as follows: wherein, is the global mean of the image, is the cumulative mean of the regions with lower gray level divided by the gradient threshold is the cumulative mean of the regions with lower gray level divided by the gradient threshold is the value at the th percentile of the normalized histogram, i.e. is the probability of a pixel with gray level is the pixel level of the image.​ 6. A method for suppressing the stripe noise based on a single image as claimed in claim 1, wherein, The binaryzation image of S1 The morphological processing is done as follows: This formula represents the use of structural elements. Dilatation of input image Effective area image For the input image For use The result of the expansion is that the structuring element The origin is shifted to the input image. of Location, if Input image of Location and If the intersection of the two regions is not empty, then the valid region image is output. Corresponding position The structure element is assigned a value of 1 if it is not assigned a value of 0 otherwise. Take as: .

7. A method for suppressing the stripe noise based on a single image as claimed in claim 1, wherein, S3 said input image with the stripe noise estimation image The formula for obtaining the denoised image with the stripe noise removed by differencing is: wherein is the denoised image with the stripe noise removed.

Images