Self-adaptive image surface texture elimination method based on phase-locked infrared thermal wave nondestructive testing imaging

Abstract

The invention discloses a self-adaptive image surface texture elimination method based on phase-locked infrared thermal wave nondestructive testing imaging, and belongs to the technical field of self-adaptive image surface texture elimination methods.According to the self-adaptive image surface texture elimination method based on phase-locked infrared thermal wave nondestructive testing imaging, three algorithms including a relative total variation (RTV) algorithm, a discrete Fourier transform (DFT) algorithm and a bilateral filtering method are combined to remove textures on the surface of a material, and the texture of the surface of the material is eliminated through the phase-locked infrared thermal wave nondestructive testing imaging based on the phase-locked infrared thermal wave nondestructive testing imaging based on the phase-locked infrared thermal wave nondestructive testing imaging based on the phase-locked infrared thermal wave nondestructive testing imaging based on the phase-locked infrared thermal wave nondestructive testing imaging based on the phase-locked infrared thermal wave nondestructive testing imaging based on the phase-locked infrared thermal wave nondestructive testing imaging. As the three algorithms have advantages on different texture removal effects, the result images obtained by the three algorithms are subjected to image fuzzy degree, texture structure similarity degree and image similarity degree scoring, and the result image with the best texture removal effect is finally output, so that the human eye impression of the infrared thermal wave nondestructive testing result image is improved.

Description

Technical Field

[0001] The invention relates to an adaptive image surface texture elimination method, in particular to an adaptive image surface texture elimination method based on phase-locked infrared thermal wave nondestructive testing imaging, and belongs to the technical field of adaptive image surface texture elimination methods. Background Art

[0002] In high-risk industries such as aerospace, nuclear power, bridges and construction, the safety of materials and structures is critical. Ensuring the integrity of materials is an important measure to prevent catastrophic failures, extend service life, and improve product quality and consistency.

[0003] Therefore, there is an urgent need in this field to find a way to detect whether there are defects inside the material.

[0004] For example, phase-locked infrared thermal wave nondestructive testing technology can promptly detect and locate internal defects of materials, prevent accidents, and improve the safety and reliability of the overall system.

[0005] However, with the demand for detecting subtle defects inside materials, the texture of the material surface will interfere with the detection of subtle defects by infrared thermal wave nondestructive testing technology. Therefore, an adaptive image surface texture elimination method based on phase-locked infrared thermal wave nondestructive testing imaging is designed to solve the above technical problems. Summary of the invention

[0006] The main purpose of the present invention is to provide an adaptive image surface texture elimination method based on phase-locked infrared thermal wave nondestructive testing imaging.

[0007] The purpose of the present invention can be achieved by adopting the following technical solutions:

[0008] An adaptive image surface texture removal method based on phase-locked infrared thermal wave nondestructive testing imaging comprises the following steps:

[0009] Step 1: Set the sine wave signal;

[0010] Step 2: driving the light source to illuminate the sample to be tested;

[0011] Step 3: Collect infrared image sequence;

[0012] Step 4: Calculate and obtain the phase result diagram;

[0013] Step 5: Remove image texture;

[0014] Step 6: Evaluate the texture removal effect and output the comprehensive score;

[0015] Step 7: Get the highest score;

[0016] Step 8: Output the phase result diagram corresponding to the highest score.

[0017] Preferably, in step one, the sinusoidal wave signal is set to generate a sinusoidal wave signal through a computer to drive the heat source to irradiate the surface of the material in a sinusoidal manner.

[0018] Preferably, in step three, the temperature change of the material surface is received by a thermal imager, that is, an infrared image sequence is generated.

[0019] Preferably, in step 4, the infrared image sequence is phase demodulated to obtain the phases of the material defect area and the defect-free area, measure the phase difference, and obtain a phase result map.

[0020] Preferably, in step five, relative total variation algorithm, discrete Fourier transform and bilateral filtering are used to remove image texture.

[0021] Preferably, the relative total variation algorithm comprises the following steps:

[0022] When calculating image gradients, the RTV algorithm measures the relative changes in local areas;

[0023] Assign different weights to different image regions;

[0024] The formula of the RTV algorithm involves image gradient and weighting factor. The RTV model is expressed as:

[0025]

[0026] Formula ① is divided into two parts, formula ② and formula ③, where formula ② is the fidelity term to ensure that the output structure image S p With the input image I p As close as possible;

[0027] Formula ③ is the relative total variation term, D x (p), D y (p) is the gradient of the pixel in the x and y directions and a weight coefficient g p,q The closer to the pixel point, the larger the weight coefficient;

[0028] L x (p), L y (p) is almost 0 in the background area with uniform fluctuations, and the value will increase in the area with obvious edges;

[0029] This item is responsible for distinguishing edges from textures;

[0030] The phase-locked infrared nondestructive testing technology is combined with the above-mentioned algorithm RTV to remove the texture of the material surface in the phase result image, so as to prevent the tiny defects of the material from being affected by the material surface texture and being ignored by the system users;

[0031]

[0032] Preferably, the discrete Fourier transform converts the image into the frequency domain, analyzes the frequency components of the image, identifies and removes unnecessary texture components, and the DFT removes the image texture through the following steps:

[0033] Apply DFT to the image and convert it to the frequency domain. DFT represents each pixel in the image as a weighted sum of a series of sine and cosine waves. The frequency and amplitude of these sine and cosine waves reflect the different frequency components in the image. The DFT formula is shown below:

[0034]

[0035] Among them, f(x,y) is the image pixel value in the spatial domain, F(u,v) is the transformation result in the frequency domain, u and v represent the frequency index in the horizontal and vertical directions respectively, and j is the imaginary unit;

[0036] In the frequency domain, low-frequency components represent the overall brightness and basic structure of the image, while high-frequency components represent details and noise, including surface texture. By analyzing the spectrogram, we can identify which frequency components correspond to unwanted textures.

[0037] Use a low-pass filter or other filtering method to remove high-frequency components. The low-pass filter allows low-frequency components to pass through, while blocking or weakening high-frequency components, thereby removing texture from the image;

[0038] The filtered frequency domain image is then transformed back to the spatial domain using an inverse discrete Fourier transform. The resulting image retains the basic structure of the original image but removes unnecessary high-frequency texture components. The IDFT formula is as follows:

[0039]

[0040] Among them, f(x,y) is the restored spatial domain image, and F(u,v) is the transformation result in the frequency domain.

[0041] Preferably, the bilateral filtering includes removing image noise and texture while retaining the edge details of the image. The bilateral filtering calculates the weighted average of each pixel by combining the information of the spatial domain and the intensity domain, thereby achieving a smooth image. The bilateral filtering is: for each pixel, its new pixel value is the weighted average of the pixel values ​​in its neighborhood, and this weight consists of two parts:

[0042] Spatial distance weight: The weight value decreases as the Euclidean distance between the pixel and the center pixel increases;

[0043] Intensity difference weight: The weight value decreases as the difference between the pixel value and the center pixel value increases;

[0044] The formula for the bilateral filter is as follows:

[0045]

[0046] in:

[0047] I filtered (x) is the pixel value of the filtered image at position x;

[0048] I(x i ) is the original image at position x i The pixel value of

[0049] Ω is the neighborhood of the center pixel x in the image;

[0050] f s (||x i -x||) is the spatial distance weight, usually represented by a Gaussian function:

[0051]

[0052] f T (||I(x i )-I(x)||) is the intensity difference weight, which is usually also represented by a Gaussian function:

[0053]

[0054] W p is a normalization factor that ensures that the weights sum to 1:

[0055]

[0056] Since the bilateral filter takes into account the spatial distance and intensity difference between pixels when calculating the weighted average, it can effectively smooth the texture in the image while keeping the edges clear;

[0057] The spatial distance weight will reduce the weight of the pixels far away from the center in the neighborhood, while the intensity difference weight will reduce the weight of the pixels with large differences from the center pixel value. Therefore, the bilateral filter can remove high-frequency texture components and retain the edge information of the image.

[0058] Beneficial technical effects of the present invention:

[0059] The present invention provides an adaptive image surface texture removal method based on phase-locked infrared thermal wave nondestructive testing imaging. This patent combines three algorithms, namely relative total variation (RTV), discrete Fourier transform (DFT), and bilateral filtering, to remove the texture of the material surface. Since the three algorithms have their own advantages in different texture removal effects, the result images obtained by the three algorithms are scored in terms of image blur, texture structure similarity, and image similarity. Finally, the result image with the best texture removal effect is output to improve the human eye perception of the infrared thermal wave nondestructive testing result image. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A schematic diagram of a phase-locked infrared nondestructive testing process and principle according to a preferred embodiment of a method for adaptively removing image surface texture based on phase-locked infrared thermal wave nondestructive testing imaging of the present invention;

[0061] Figure 2 A schematic diagram of a phase-locked infrared thermal wave nondestructive testing device according to a preferred embodiment of a method for adaptively removing image surface texture based on phase-locked infrared thermal wave nondestructive testing imaging of the present invention;

[0062] Figure 3 A preferred embodiment of an adaptive image surface texture elimination method based on phase-locked infrared thermal wave nondestructive testing imaging according to the present invention. Calculate the phase result map obtained by forming an infrared image sequence from the temperature change of the material surface. (a) The front and back sides of the inspected material. (b) The shooting process of the infrared thermal imager. (c) The infrared thermal image sequence. (d) The phase result map;

[0063] Figure 4 A flowchart of an adaptive image surface texture removal method according to a preferred embodiment of an adaptive image surface texture removal method based on phase-locked infrared thermal wave nondestructive testing imaging of the present invention;

[0064] Figure 5 A schematic diagram of a carbon fiber board (the left figure is the board surface, and the right figure is the board back) according to a preferred embodiment of an adaptive image surface texture removal method based on phase-locked infrared thermal wave nondestructive testing imaging of the present invention;

[0065] Figure 6 A phase result diagram before being processed by a texture removal algorithm according to a preferred embodiment of an adaptive image surface texture removal method based on phase-locked infrared thermal wave nondestructive testing imaging of the present invention;

[0066] Figure 7A phase result diagram before being processed by a texture removal algorithm according to a preferred embodiment of an adaptive image surface texture removal method based on phase-locked infrared thermal wave nondestructive testing imaging of the present invention;

[0067] Figure 8 This is a phase result diagram before texture removal algorithm processing according to a preferred embodiment of an adaptive image surface texture removal method based on phase-locked infrared thermal wave nondestructive testing imaging of the present invention.

[0068] Fig. 9 This is a texture removal effect comprehensive evaluation comparison chart of a preferred embodiment of an adaptive image surface texture removal method based on phase-locked infrared thermal wave nondestructive testing imaging according to the present invention. The comprehensive evaluation scores of RTV and DFT on the surface texture removal effect of the phase result map of a carbon fiber plate with a thickness of 3mm are the same, but because it was previously stipulated that: when the comprehensive evaluation scores are the same, the image with the highest image clarity score is selected as the final result for output, so the adaptive algorithm selects the phase result map after RTV processing as the output result. Next, let's take a look at the texture removal effect of the phase result map of a carbon fiber plate with a thickness of 5mm.

[0069] Fig.10 This is a comparison chart of the comprehensive evaluation of texture removal effects according to a preferred embodiment of an adaptive image surface texture removal method based on phase-locked infrared thermal wave nondestructive testing imaging of the present invention. When facing a carbon fiber plate with a thickness of 5 mm, the texture effect of using RTV and bilateral filtering to remove the phase result image is far inferior to DFT. From the perspective of the human eye, although the surface of the plate in the phase result image processed by RTV looks much cleaner, the pattern of the defects is obviously deformed. The texture removal of bilateral filtering is similar to the result of DFT processing in appearance, but the phase result image after bilateral filtering processing is more blurred and loses more details. The final result also shows that DFT has the highest comprehensive evaluation score, which better fits the perception of the human eye.

[0070] Fig.11 This is a schematic diagram of a texture removal effect comprehensive evaluation comparison of a preferred embodiment of an adaptive image surface texture removal method based on phase-locked infrared thermal wave nondestructive testing imaging according to the present invention. The original carbon fiber board of the phase result diagram of the texture removal effect evaluation comparison is a board with a length*width*height of 10cm*10cm*4mm. There are crack defects, circular defects and square defects inside the board, and the defect depth varies from 2.4mm to 3.6mm. Figure 8It can be seen that the defects detected in the phase result image are greatly affected by the texture. After using the texture removal algorithm, the position signal of the defect can be obtained and analyzed more clearly. From a comprehensive evaluation point of view, the DFT algorithm has the best effect in processing texture and is the choice that conforms to the perception of the human eye. The reason why the comprehensive evaluation score of RTV is similar to that of DFT is that most images processed by the RTV algorithm have a tendency to sharpen, and the texture removal is relatively complete, so the comprehensive evaluation score is often higher. However, under the selection of the adaptive algorithm of this patent, it is possible to make a decision on the texture removal algorithm with the best processing effect. DETAILED DESCRIPTION

[0071] In order to make the technical solution of the present invention more clear and specific to those skilled in the art, the present invention is further described in detail below in conjunction with embodiments and drawings, but the implementation manner of the present invention is not limited thereto.

[0072] Principle of Lock-in Infrared Thermal Wave Nondestructive Testing Technology

[0073] Phase-locked infrared thermal wave nondestructive testing technology uses infrared thermal imaging technology to detect internal defects of materials. Its basic principle is to stimulate the surface of the material through active periodic heating, and track the periodic changes in the surface temperature of the material through an infrared thermal imager to detect whether there are defects inside the material.

[0074] The detection of internal defects in materials is mainly divided into three steps, such as Figure 1 As shown:

[0075] 1. A computer generates a sinusoidal wave signal to drive the heat source (a halogen lamp in phase-locked infrared nondestructive testing) to irradiate the material surface in a sinusoidal manner.

[0076] 2. Receive the temperature change of the material surface through the thermal imager, and generate an infrared image sequence.

[0077] 3. Perform phase demodulation on the infrared image sequence to obtain the phase between the defective area and the defect-free area of ​​the material, measure the phase difference, and obtain the phase result map.

[0078] The phase-locked infrared thermal wave nondestructive testing device set in this patent is as follows Figure 2 As shown in the figure, it is composed of a thermal imager, a tunable heat source, a bracket and a storage rack. In addition, the generation of the signal is controlled by the host computer, which can make the light source change in a sinusoidal manner, so that the sample under test on the storage plate is heated by the tuned excitation source, so as to detect whether there are defects inside the sample.

[0079] The phase-locked infrared thermal wave nondestructive testing technology can detect tiny temperature changes and is suitable for detecting tiny material defects. It does not require direct contact with the object being tested and is suitable for testing materials with various complex shapes and in high-temperature environments.

[0080] Depend on Figure 3 As can be seen in Figures (a) and (d), the phase result map not only reflects the defect location, but also clearly shows the texture information of the material surface. Such a situation is extremely unfavorable for detecting tiny defects inside the inspected material when there is texture on the surface. Therefore, this patent proposes to combine the phase-locked infrared thermal wave nondestructive testing system with the texture removal algorithm to obtain a phase result map that only contains defect information. In addition, different texture removal algorithms have their own advantages in the face of different texture features. Therefore, this patent will use three texture removal algorithms to remove texture from the phase result map, and use the image texture score weight algorithm to adaptively screen out the result map with the best texture removal effect from the three result maps after texture removal, and use this as the result of the phase-locked infrared nondestructive testing.

[0081] The following introduces the image texture removal algorithm and the image texture scoring weight algorithm.

[0082] 2. Relative Total Variation (RTV)

[0083] RTV is an algorithm used in image processing and computer vision, mainly used for tasks such as image texture removal, image segmentation, and image reconstruction. The RTV algorithm analyzes local changes in the image, identifies and retains important structural features in the image, and removes image texture and unimportant details.

[0084] The RTV algorithm is based on the concept of total variation (TV), but adds relative factors to better preserve the edge and structural information of the image when processing the image. The total variation algorithm is usually used for image texture removal to achieve a smoothing effect by minimizing the sum of image gradients.

[0085] The relative total variation (RTV) algorithm improves upon the standard total variation method by:

[0086] 1. Relative variation measurement: When calculating image gradients, the RTV algorithm considers the relative changes in local areas, rather than just the absolute changes. This can better preserve the edges and details in the image.

[0087] 2. Adaptive weights: Assign different weights to different image regions to better balance the relationship between texture removal and detail preservation.

[0088] The core formula of the RTV algorithm usually involves image gradient and weighting factor. The RTV model can be expressed as:

[0089]

[0090] Formula ① is divided into two parts, formula ② and formula ③, where formula ② is the fidelity term to ensure that the output structure image S p With the input image I p As close as possible. Formula ③ is the relative total variation term, D x (p), D y (p) is the gradient of the pixel in the x and y directions and a weight coefficient g p,q The closer to the pixel, the larger the weight coefficient. x (p), L y (p) is almost 0 in the background area with uniform fluctuations, and the value will increase in the area with obvious edges. This item is responsible for distinguishing edges from textures.

[0091] The phase-locked infrared nondestructive testing technology, by combining with the above-mentioned algorithm RTV, can remove the texture of the material surface in the phase result image, thereby preventing tiny defects of the material from being affected by the material surface texture and being ignored by the system user.

[0092]

[0093] 3. Discrete Fourier Transform (DFT)

[0094] Discrete Fourier Transform (DFT) is a mathematical method that converts spatial domain images to frequency domain. By converting the image to the frequency domain, the frequency components of the image can be analyzed, and unwanted texture components can be identified and removed. DFT removes image texture through the following steps:

[0095] 1. Apply DFT to the image and convert it to the frequency domain. DFT represents each pixel in the image as a weighted sum of a series of sine and cosine waves. The frequency and amplitude of these sine and cosine waves reflect the different frequency components in the image. The DFT formula is shown below.

[0096]

[0097] Among them, f(x,y) is the image pixel value in the spatial domain, F(u,v) is the transformation result in the frequency domain, u and v represent the frequency index in the horizontal and vertical directions respectively, and j is the imaginary unit.

[0098] 2. In the frequency domain, low-frequency components usually represent the overall brightness and basic structure of the image, while high-frequency components usually represent details and noise, including surface texture. By analyzing the spectrogram, it is possible to identify which frequency components correspond to unwanted textures.

[0099] 3. Use a low-pass filter or other filtering methods to remove high-frequency components. A low-pass filter allows low-frequency components to pass through, while blocking or weakening high-frequency components, thereby removing texture from the image.

[0100] 4. The filtered frequency domain image is then converted back to the spatial domain using the inverse discrete Fourier transform (IDFT). The resulting image retains the basic structure of the original image but removes the unnecessary high-frequency texture components. The IDFT formula is shown below.

[0101]

[0102] Among them, f(x,y) is the restored spatial domain image, and F(u,v) is the transformation result in the frequency domain.

[0103] 4. Bilateral Filtering

[0104] Bilateral filtering is a nonlinear filtering method that considers both spatial information and pixel value similarity. It can better preserve the edge details of the image while removing image noise and texture. Bilateral filtering calculates the weighted average of each pixel by combining information in the spatial domain and the intensity domain to smooth the image. The basic idea of ​​bilateral filtering is that for each pixel, its new pixel value is the weighted average of the pixel values ​​in its neighborhood, and this weight consists of two parts:

[0105] 1. Spatial distance weight: The weight value decreases as the Euclidean distance between the pixel and the center pixel increases.

[0106] 2. Intensity difference weight: The weight value decreases as the difference between the pixel value and the center pixel value increases.

[0107] The formula for the bilateral filter is as follows:

[0108]

[0109] in:

[0110] 1.I filtered (x) is the pixel value of the filtered image at position x.

[0111] 2. I(x i ) is the original image at position x i The pixel value of .

[0112] 3. Ω is the neighborhood of the center pixel x in the image.

[0113] 4. f s (||x i -x||) is the spatial distance weight, usually represented by a Gaussian function:

[0114]

[0115] 5. f T (||I(x i )-I(x)||) is the intensity difference weight, which is usually also represented by a Gaussian function:

[0116]

[0117] W p is a normalization factor that ensures that the weights sum to 1:

[0118]

[0119] Since the bilateral filter considers the spatial distance and intensity difference between pixels when calculating the weighted average, it can effectively smooth the texture in the image while keeping the edges clear. The spatial distance weight will reduce the weight of the pixels far away from the center in the neighborhood, while the intensity difference weight will reduce the weight of the pixels with a large difference in value from the center pixel. Therefore, the bilateral filter can effectively remove high-frequency texture components but retain the edge information of the image.

[0120] 5. Image Texture Scoring Weight Algorithm

[0121] After obtaining the texture-free images processed by the three algorithms, it is necessary to evaluate the image blur degree, texture structure similarity and image similarity of the result images obtained by the three algorithms.

[0122] 1. Image blur evaluation, using the variance evaluation of the Laplace operator. The Laplace operator is a second-order derivative operator used to detect edges in images. Since the edges in the image will produce large numerical changes under the response of the second-order derivative, the variance of the Laplace operator can be used to evaluate the clarity of the image. A key factor in image clarity is the sharpness of the edge. The edge response of a blurred image is weak, while the edge response of a clear image is strong. The formula for the Laplace operator is:

[0123]

[0124] Where f is the grayscale value of the image, is the image after the Laplace operator is applied. In practical applications, the Laplace operator is usually implemented using a discrete convolution kernel. The following is the process of evaluating image clarity through the variance of the Laplace operator:

[0125] Calculate the Laplace response of an image: Convolve the image with the Laplace operator to obtain the Laplace response image.

[0126] Calculate the variance of the Laplace response: Variance is a statistic that measures the degree of data fluctuation. For a Laplace response image, the larger the variance, the more dramatic the edge changes in the image and the clearer the image; the smaller the variance, the more gradual the edge changes in the image and the blurrier the image. The formula for calculating the variance of the Laplace response is:

[0127]

[0128] N is the total number of pixels in the image. i is the i-th pixel value of the Laplace response image. u is the mean value of the Laplace response image.

[0129] 2. The similarity of texture structure is evaluated by the co-occurrence matrix method. The Gray Level Co-occurrence Matrix (GLCM) is a commonly used method for extracting image texture features. It captures texture information by counting the spatial co-occurrence relationship between gray levels in the image. Using the co-occurrence matrix method to evaluate the similarity of texture structure can quantify the texture features of different images or different regions of the same image, and determine the similarity of texture structure by calculating the similarity of feature vectors.

[0130] The co-occurrence matrix P is an N×N matrix, where N is the number of gray levels in the image. The element P(x,y) in the co-occurrence matrix represents the number of times a pixel pair with gray value i and gray value j in the image appears at a certain distance and direction. The calculation steps of the co-occurrence matrix are as follows:

[0131] Set parameters: Determine the distance d and direction θ (such as 0 degrees, 45 degrees, 90 degrees, 135 degrees) of the co-occurrence matrix.

[0132] Initialize matrix: Initialize an N×N zero matrix P.

[0133] Count co-occurrence pairs: traverse the image, count the grayscale pairs that meet the distance and direction requirements, and update the co-occurrence matrix.

[0134] Normalization: Normalize the co-occurrence matrix so that the sum of its elements is 1.

[0135] Through the co-occurrence matrix, a variety of texture features can be calculated, including energy, contrast, entropy, inverse difference moment (IDM), etc. The following are several commonly used co-occurrence matrix features and their formulas:

[0136] Energy reflects the uniformity of image texture. The larger the value, the more uniform the texture, as shown in the following formula.

[0137]

[0138] Contrast reflects the contrast of image texture. The larger the value, the higher the texture contrast, as shown in the following formula.

[0139]

[0140] Entropy reflects the complexity of image texture. The larger the value, the more complex the texture, as shown in the following formula.

[0141]

[0142] The inverse gap reflects the similarity of image textures. The larger the value, the more similar the textures are, as shown in the following formula.

[0143]

[0144] In order to evaluate the similarity of texture structures of two images or different regions of the same image, their co-occurrence matrix features can be compared. A common method is to calculate the distance between feature vectors, such as Euclidean distance. Assume that the feature vector of image 1 is The feature vector of image 2 is Where k is the number of features. The smaller the distance D, the more similar the texture structures of the two images are; the larger the distance D, the less similar the texture structures of the two images are. The calculation formula is as follows:

[0145]

[0146] 3. The similarity between two images is measured by the Structural Similarity Index (SSIM). SSIM pays special attention to the structural information of the image. It calculates the similarity score of the image by comparing the brightness, contrast and structure of the image. The higher the score, the more similar the images are, which can well reflect the difference in image quality perceived by the human eye. The calculation formula is as follows:

[0147]

[0148] Among them, x and y are the pixel values ​​of the two images, μ x and μ y is the mean of images x and y, σ x and σ y is the variance of images x and y, σ xy The covariance of images x and y, C1 and C2 are stable constants introduced to avoid the denominator being zero, usually taken as C1 = (K1L) 2 and C2=(K2L) 2, where L is the dynamic range of pixel values, usually 255; K1 and K2 are two small constants, usually 0.01 and 0.03 respectively

[0149] We can evaluate the results of different image texture removal algorithms through the image blur degree, texture structure similarity and image similarity, but different image texture removal algorithms have different focuses, resulting in different scores of the three image texture removal algorithms in image blur degree, texture structure similarity and image similarity. Therefore, this patent also designs a nonlinear comprehensive evaluation index to summarize the comprehensive scores of image blur degree, texture structure similarity and image similarity, and adaptively outputs the algorithm with the best image texture removal effect for the phase result map output by this device. The formula of the nonlinear comprehensive evaluation index is shown below.

[0150]

[0151] The above f(x) is as follows:

[0152]

[0153] n(x) represents the results obtained by different algorithms under various evaluation indicators, and max_n represents the maximum value in n(x). If the scores are the same, the image with the best texture removal effect is output with clarity as the primary consideration.

[0154] The above description is only a further embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and concepts of the present invention within the scope disclosed by the present invention, which belong to the protection scope of the present invention.

Claims

1. An adaptive image surface texture removal method based on phase-locked infrared thermal wave nondestructive testing imaging, characterized in that: The steps include: Step 1: Set the sine wave signal; Step 2: driving the light source to illuminate the sample to be tested; Step 3: Collect infrared image sequence; Step 4: Calculate and obtain the phase result diagram; Step 5: Remove image texture; Step 6: Evaluate the texture removal effect and output the comprehensive score; Step 7: Get the highest score; Step 8: Output the phase result diagram corresponding to the highest score.

2. The method for adaptively removing image surface texture based on phase-locked infrared thermal wave nondestructive testing imaging according to claim 1 is characterized by: In step one, the sinusoidal wave signal is set to generate a sinusoidal wave signal through a computer to drive the heat source to irradiate the surface of the material in a sinusoidal manner.

3. The method for adaptively removing image surface texture based on phase-locked infrared thermal wave nondestructive testing imaging according to claim 2, characterized in that: In step three, the temperature change of the material surface is received by the thermal imager, that is, an infrared image sequence is generated.

4. The method for adaptively removing image surface texture based on phase-locked infrared thermal wave nondestructive testing imaging according to claim 3 is characterized by: In step 4, the infrared image sequence is phase demodulated to obtain the phase between the material defect area and the non-defect area, measure the phase difference, and obtain a phase result map.

5. The method for adaptively removing image surface texture based on phase-locked infrared thermal wave nondestructive testing imaging according to claim 4 is characterized in that: In step five, relative total variation algorithm, discrete Fourier transform and bilateral filtering are used to remove image texture.

6. The method for adaptively removing image surface texture based on phase-locked infrared thermal wave nondestructive testing imaging according to claim 5, characterized in that: The relative total variation algorithm includes the following steps: When calculating image gradients, the RTV algorithm measures the relative changes in local areas; Assign different weights to different image regions; The formula of the RTV algorithm involves image gradient and weighting factor. The RTV model is expressed as: Formula ① is divided into two parts, formula ② and formula ③, where formula ② is the fidelity term to ensure that the output structure image S p With the input image I p As close as possible; Formula ③ is the relative total variation term, D x (p), D y (p) is the gradient of the pixel in the x and y directions and a weight coefficient g p,q The closer to the pixel point, the larger the weight coefficient; L x (p), L y (p) is almost 0 in the background area with uniform fluctuations, and the value will increase in the area with obvious edges; This item is responsible for distinguishing edges from textures; The phase-locked infrared nondestructive testing technology is combined with the above-mentioned algorithm RTV to remove the texture of the material surface in the phase result image, so as to prevent the tiny defects of the material from being affected by the material surface texture and being ignored by the system users; 7. The method for adaptively removing image surface texture based on phase-locked infrared thermal wave nondestructive testing imaging according to claim 6 is characterized by: Discrete Fourier transform converts the image into the frequency domain, analyzes the frequency components of the image, identifies and removes unnecessary texture components. DFT removes image texture through the following steps: Apply DFT to the image and convert it to the frequency domain. DFT represents each pixel in the image as a weighted sum of a series of sine and cosine waves. The frequency and amplitude of these sine and cosine waves reflect the different frequency components in the image. The DFT formula is shown below: Among them, f(x,y) is the image pixel value in the spatial domain, F(u,v) is the transformation result in the frequency domain, u and v represent the frequency index in the horizontal and vertical directions respectively, and j is the imaginary unit; In the frequency domain, low-frequency components represent the overall brightness and basic structure of the image, while high-frequency components represent details and noise, including surface texture. By analyzing the spectrogram, we can identify which frequency components correspond to unwanted textures. Use a low-pass filter or other filtering method to remove high-frequency components. The low-pass filter allows low-frequency components to pass through, while blocking or weakening high-frequency components, thereby removing texture from the image; The filtered frequency domain image is then transformed back to the spatial domain using an inverse discrete Fourier transform. The resulting image retains the basic structure of the original image but removes unnecessary high-frequency texture components. The IDFT formula is as follows: Among them, f(x,y) is the restored spatial domain image, and F(u,v) is the transformation result in the frequency domain.

8. The method for adaptively removing image surface texture based on phase-locked infrared thermal wave nondestructive testing imaging according to claim 7 is characterized by: Bilateral filtering involves removing image noise and texture while retaining the edge details of the image. Bilateral filtering calculates the weighted average of each pixel by combining information in the spatial domain and the intensity domain to smooth the image. For each pixel, its new pixel value is the weighted average of the pixel values ​​in its neighborhood. This weight consists of two parts: composition: Spatial distance weight: The weight value decreases as the Euclidean distance between the pixel and the center pixel increases; Intensity difference weight: The weight value decreases as the difference between the pixel value and the center pixel value increases; The formula for the bilateral filter is as follows: in: I filtered (x) is the pixel value of the filtered image at position x; I(x i ) is the original image at position x i The pixel value of Ω is the neighborhood of the center pixel x in the image; f s (||x i -x||) is the spatial distance weight, usually represented by a Gaussian function: f T (||I(x i )-I(x)||) is the intensity difference weight, which is usually also represented by a Gaussian function: W p is a normalization factor that ensures that the weights sum to 1: Since the bilateral filter takes into account the spatial distance and intensity difference between pixels when calculating the weighted average, it can effectively smooth the texture in the image while keeping the edges clear; The spatial distance weight will reduce the weight of the pixels far away from the center in the neighborhood, while the intensity difference weight will reduce the weight of the pixels with large differences from the center pixel value. Therefore, the bilateral filter can remove high-frequency texture components and retain the edge information of the image.

Images