An image enhancement method based on local image features for pixel value offset

Abstract

The application discloses an image enhancement method based on local image features for pixel value offset, and belongs to the field of imaging logging image processing. Specifically, firstly, an algorithm for randomly traversing each pixel value is determined; then, window sampling is performed on each pixel value to be traversed by specifying a window size; the offset direction and offset size of the pixel value are calculated by using the image data in the window; finally, the pixel value is offset by the offset direction and offset size, the image is visually enhanced, and the non-uniformity of the image is improved. The application can improve the non-uniformity of the image, enhance the visual effect of the image, and improve the precision of a deep learning model by using the uniformity image picture processed by the method to train the deep learning model.

Description

Technical Field

[0001] This invention belongs to the field of imaging logging image processing, and specifically discloses an image enhancement method based on pixel value offsetting of local image features. Background Technology

[0002] Imaging logging technology involves downhole sensor array scanning or rotational scanning to acquire high-density formation resistivity information along the longitudinal, circumferential, or radial directions of the wellbore. This data is then transmitted to the surface and processed using image processing techniques to obtain two-dimensional images of the wellbore or three-dimensional images of the area around the wellbore at a specific depth. Imaging logging can be used for geological applications such as core data relocation, lithology identification, formation dip calculation, fracture evaluation, bedding identification, and sedimentary facies analysis. Imaging logging is more accurate, intuitive, and convenient than conventional logging methods. However, when using imaging logging data, wellbore images often exhibit weak heterogeneity, indistinct imaging features, poor visual effects, and low accuracy in training deep learning models. To address these issues, image enhancement methods can be used to highlight image features, enhance visual effects, and improve the accuracy of trained deep learning models. Summary of the Invention

[0003] The purpose of this invention is to provide an image enhancement method based on pixel value offset of local feature images, in order to improve the visual effect of electro-imaging images and improve the accuracy of deep learning models.

[0004] The present invention specifically adopts the following solution:

[0005] An image enhancement method based on local image features and pixel value offset includes the following steps:

[0006] (1) Obtain the input image and its length, width, and number of channels;

[0007] (2) Calculate the total number of pixels N that need to be traversed. P And use the Knuth-Durstenfeld Shuffle algorithm to shuffle cards of length N. P The sequence is shuffled randomly, and random numbers are traversed through the sequence to ensure that each pixel of the image is randomly traversed;

[0008] (3) Using the pixel position (i,j) mapped by random numbers and the selected window size n, the imaging logging data is traversed in an n*n window (n = 3, 5, 7...).

[0009] (4) Calculate the offset direction and offset amount of each pixel within the window of recorded imaging data. Offset direction:

[0010]

[0011] Where: DIR ij The pixel offset direction, with a value of 1 or -1, d ij D is the pixel value, where D is the pixel value d. ij The surrounding n*n pixel matrix, The average value of matrix D;

[0012] The offset size is represented as:

[0013]

[0014] Where: STEP ij Let max(D) be the magnitude of the offset, max(D) be the maximum value of matrix D, min(D) be the minimum value of matrix D, and μ be the offset magnitude, μ∈(0,1).

[0015] (5) Offset the pixel; the pixel offset result is:

[0016] d ij_N =d ij +DIR ij *STEP ij

[0017] d ij DIR is the pixel value at that point. ij This indicates the pixel offset direction, with a value of 1 or -1. STEP ij This refers to the magnitude of the offset calculated above;

[0018] (6) Repeat step (2) until the loop ends and output the image.

[0019] The present invention has the following beneficial effects:

[0020] This method can enhance the visual effect of images and increase their non-uniformity. Using electro-imaging images preprocessed by this method to train deep learning models can improve the recognition accuracy of the trained models. Attached Figure Description

[0021] Figure 1 Here is the algorithm flowchart for this method;

[0022] Figure 2 A comparison chart showing the processing effects of different image enhancement methods on the test segment;

[0023] Figure 3 A comparison of the effects of different image enhancement methods;

[0024] Figure 4 This is a comparison of grayscale distribution histograms showing the effects of different image enhancement methods on the test segment. Detailed Implementation

[0025] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and specific examples:

[0026] Combination Figure 1 An image enhancement method based on local image features and pixel value offset includes the following steps:

[0027] (1) Obtain the input image and its length, width, and number of channels;

[0028] (2) Calculate the total number of pixels N that need to be traversed. P And the Knuth-Durstenfeld Shuffle algorithm was used to shuffle N P The data is shuffled randomly, and by iterating through these random numbers, we ensure that each pixel of the image is randomly traversed.

[0029] (3) Using the pixel position (i,j) mapped by random numbers and the selected window size n, the imaging logging data is traversed in n*n, n=3, 5, 7... (n=5 has the best effect);

[0030] (4) Calculate the offset direction and offset amount of each pixel within the window of recorded imaging data. Offset direction:

[0031]

[0032] Where: DIR ij The pixel offset direction, with a value of 1 or -1, d ij D is the pixel value, where D is the pixel value d. ij The surrounding n*n pixel matrix, The average value of matrix D;

[0033] The offset size is represented as:

[0034]

[0035] Where: STEP ij Let max(D) be the magnitude of the offset, max(D) be the maximum value of matrix D, min(D) be the minimum value of matrix D, and μ be the offset magnitude, μ∈(0,1).

[0036] (5) Offset the pixel; the pixel offset result is:

[0037] d ij_N =d ij +DIR ij *STEP ij

[0038] d ij DIR is the pixel value at that point. ijThis indicates the pixel offset direction, with a value of 1 or -1. STEP ij This refers to the magnitude of the offset calculated above;

[0039] (6) Repeat step (2) until the loop ends and output the image.

[0040] Based on the algorithm described above, the case of a single traversal is as follows:

[0041] Input an image of 266×250×1 (with 1 channel);

[0042] The total number of points to be traversed is confirmed to be 266×250×1. The shuffling algorithm processes a random sequence array of length 266×250×1, ensuring that the random sequence array corresponds to each pixel point while shuffling its order.

[0043] Traverse the random sequence array. The first value of the random sequence array is 3129, which is mapped to the 129th pixel in the 12th row of the input image (3129 % 250 = 12...129).

[0044] Take the image data of a 5×5 window near the (12, 129) pixel point;

[0045] Calculate the offset direction and offset amount based on the image data within the window;

[0046] Offset the pixel value at that point;

[0047] Determine if the loop has been completed. If not, continue iterating through the random sequence array and performing pixel value offset operations.

[0048] Figure 2 The image enhancement effect of the algorithm is compared with different window lengths. When the window size is 5, the image enhancement improves the visual effect of the image while preserving the details. As the window length increases, the image visual effect improves the non-homogeneity effect, but some details are also erased.

[0049] Figure 3 The figures show a comparison of the processing effects of different image enhancement methods on the test segment. Figure a shows the dynamic imaging of the original data, Figure b shows the imaging effect after processing the test imaging data of this invention, Figure c shows the imaging effect after gamma transform image enhancement of the imaging data of this segment, and Figure d shows the imaging effect after histogram equalization image enhancement processing of the imaging data of this test segment.

[0050] Figure 4 The images show a comparison of grayscale distribution histograms after different image enhancement methods were applied to the test segment. The comparison reveals that the grayscale distribution histogram of the test electro-imaging image enhanced by this invention is more balanced, exhibits greater variance, stronger non-uniformity, and more pronounced features.

[0051] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. An image enhancement method based on pixel value shifting of local image features, characterized in that, Includes the following steps: (1) Obtain the input image and its length, width, and number of channels; (2) Calculate the total number of pixels N that need to be traversed. P And the Knuth-Durstenfeld Shuffle algorithm was used to shuffle N P The data is shuffled randomly, and by traversing these random sequences, we ensure that each pixel of the image is randomly traversed. (3) Using the pixel position (i,j) mapped by random numbers in the random sequence and the selected window size n, the imaging logging data is traversed in windows of n*n, n=3, 5, 7... (4) Calculate the offset direction and offset amount of each pixel within the window of recorded imaging data. Offset direction: Where: DIR ij The pixel offset direction, with a value of 1 or -1, d ij D is the pixel value, where D is the pixel value d. ij The surrounding n*n pixel matrix, The average value of matrix D; The offset size is represented as: Where: STEP ij Let max(D) be the magnitude of the offset, max(D) be the maximum value of matrix D, min(D) be the minimum value of matrix D, and μ be the offset magnitude, μ∈(0,1). (5) Offset the pixel; the pixel offset result is: d ij_N =d ij +DIR ij *STEP ij d ij DIR is the pixel value at that point. ij This indicates the pixel offset direction, with a value of 1 or -1. STEP ij This refers to the magnitude of the offset calculated above; (6) Repeat step (2) until the loop ends and output the image.

Images