Method and system for constructing image model based on random bright spots and irregular stripes

Abstract

The invention provides an image model construction method and system based on random bright spots and irregular stripes, and relates to the technical field of image processing. The method comprises the following steps: acquiring a data set in an image region, observing and analyzing the distribution characteristics of the hot spots in size, intensity and position, establishing a mathematical model of random hot spots, analyzing the complex structure characteristics of irregular stripes in an image, simulating the superposed spatial noise, and carrying out detailed modeling on various background thermal noises which influence the image quality, so as to obtain a random hot spot model. According to the method, common degradation factors in the image processing process are considered, accurate modeling is carried out, obtained different types of noise models are integrated to form a comprehensive high-order thermal infrared image degradation model, various degradation phenomena experienced by the thermal infrared image in the acquisition, transmission and processing processes are comprehensively simulated, the image processing efficiency is improved, and the image quality is improved. And the picture resolution is improved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technologies, and in particular, to a method and system for constructing an image model based on random bright spots and irregular stripes. Background Art

[0002] Current infrared thermal imaging technology detects the infrared radiation on the surface of an object and converts it into a temperature distribution image, which has advantages such as non-contact, high resolution, and real-time performance. It can perform effective detection under different lighting conditions, especially in low-light or no-light environments, and is widely used in power equipment detection (such as temperature monitoring of power lines and transformers), civil structure monitoring (such as thermal anomaly detection of structures like buildings and bridges), automotive thermal management (such as thermal analysis of engines and braking systems), metallurgical and petrochemical equipment evaluation (such as temperature distribution of furnace bodies and pipelines), and medical diagnosis (such as detection of abnormal body temperatures), etc. However, despite the many advantages of infrared thermal imaging technology, its imaging quality is often limited by the hardware performance of the thermal imager. Compared with visible light (VIS) images, thermal infrared (TIR) images usually have a series of inherent degradation characteristics: First, the resolution of thermal infrared images is generally low, which makes it difficult to clearly capture the temperature distribution of the target object in some application scenarios where details are crucial. Second, the noise level of thermal infrared images is high, which easily introduces some irrelevant interference signals, resulting in a decrease in image quality and further affecting subsequent analysis and processing. Third, edge blurring is also a common problem in thermal infrared images, which may be due to factors such as limited sensor accuracy of the imager and insufficient imaging algorithms. During the imaging process, interference phenomena such as random bright spots and uneven stripes may also occur, which will lead to inaccuracy and distortion of the image, affecting the effects in the application fields.

[0003] In the prior art, to overcome the limitations of the hardware performance of thermal imagers, using super-resolution (SR) reconstruction technology to improve the resolution of thermal infrared images has become a more efficient and economical solution. Super-resolution technology reconstructs and enhances low-resolution images by using methods such as multi-frame image information, deep learning models, or signal processing algorithms, so as to achieve the restoration of image details and the improvement of resolution. Compared with hardware upgrades, super-resolution technology not only has low costs and is convenient to implement, but also can significantly improve the clarity of thermal infrared images, reduce noise, enhance image details, and thus improve the accuracy of detection, analysis, and decision-making. Existing super-resolution algorithms [1-6] usually use an ideal downsampling method to generate paired training sets, that is, by performing ideal downsampling processing on high-resolution images , to construct the corresponding relationship between low-resolution and high-resolution images. It is usually assumed that the image degradation process is ideal and simplified, and uniform downsampling and simple interpolation techniques are used to generate low-resolution images. However, this idealized degradation method is difficult to truly reflect the actual degradation characteristics of thermal infrared images. Especially when the performance of the thermal imager is insufficient or affected by the external environment, the image degradation is often more complex. The training set constructed based on the ideal degradation model is difficult to accurately simulate the actual degradation process, resulting in poor performance of the super-resolution algorithm in applications, and may even exacerbate noise or lose details. To address this issue, existing research adopts a real image degradation method based on kernel estimation and noise injection. By establishing a degradation pool to simulate the degradation process, clean high-resolution images are degraded into blurred and noisy images to generate image pairs for training the SR model; the current high-order degradation model extends the applicable range of blind super-resolution methods by modifying the data synthesis process, and can more comprehensively cover the degradation space of the real world.

[0004] However, the current generalized degradation model considers more common degradations in visible light images and fails to fully consider problems such as random bright spots and complex stripes in thermal infrared images, and these factors also play an important role in the actual degradation process of thermal infrared images. Therefore, there is an urgent need to construct a more comprehensive degradation model that conforms to the actual situation of thermal infrared images. Summary of the Invention

[0005] The main object of the present invention is to provide a method and system for constructing an image model based on random bright spots and irregular stripes, so as to solve the technical problem that the prior art fails to comprehensively consider random bright spots, complex stripes and other noise factors in thermal infrared images, resulting in poor resolution effect of thermal infrared images.

[0006] To solve the above technical problems, the technical solution adopted by the present invention is: A method for constructing an image model based on random bright spots and irregular stripes, comprising the following steps: S1: Obtain a data set in the image area and model random bright spots; S2: Model irregular stripes, where the irregular stripes include wide stripes and thin stripes; S3: Model background thermal noise, where the background thermal noise includes background radiation noise, temporal noise, spatial noise, quantization noise and thermal noise; S4: Model traditional degradation noise, where the traditional degradation noise includes blur kernel noise, downsampling noise and compression noise; S5: Combine the models established in S1-S4 to construct a high-order thermal infrared image degradation model. In a preferred embodiment, In a preferred embodiment, the modeling of random bright spots in S1 is specifically: The sizes and intensities of the random bright spots do not differ much and there is no fixed pattern. They show a high degree of randomness in position. The formula is: (1.1); In the formula, is the number of random bright spots, is the probability of random bright spot noise, is the image height, is the image width, is the amplitude, is the bright spot radius, is the maximum intensity of the bright spot itself, is the input image; is a function describing the distribution and influence range of the bright spot. The formula is: (1.2); In the formula, is the unit step function, is to calculate the pixel to the center of the bright spot The distance is: (1.3); In the formula, Ω is the neighboring pixel, defined as: (1.4); In the formula, is the current pixel point, is the maximum intensity of the bright spot itself; is the unit step function. The formula is: (1.5); In the formula, is the maximum intensity of the bright spot itself, is to calculate the pixel to the distance of the center of the bright spot; The center position of the bright spot The formula is: (1.6); In the formula, U(a,b) is a random number uniformly distributed in the interval [a,b], and are the randomly selected coordinates of the center of the bright spot, and are the width and height of the image respectively.

[0007] In the preferred solution, the S2 models the irregular stripes, specifically: Irregular stripe noise usually shows the superposition of low-frequency and high-frequency spatial noise. The formula is: (2.1); In the formula, is the value of the wide stripe noise based on the Gaussian distribution, is the value of the generated fine stripe noise; The formula for the wide stripe noise is: (2.2); In the formula, is the amplitude of the wide stripe noise, controlling the intensity of the stripe; is the center position of the stripe, affecting the position of the stripe in the image; is the standard deviation controlling the stripe width; is the position index (column index) of each pixel in the horizontal direction in the image; The formula for the fine stripe noise is: (2.3); In the formula, i.e., the amplitude of the fine stripe, controlling the intensity of the fine stripe, is the number of superimposed sine waves, determining the complexity of the fine stripe, is the width of the image, determining the frequency of each sine wave and affecting the density of the stripe. In the preferred solution, the background thermal noise is modeled in S3. The overall noise model generated by the thermal imager has the formula: (3.1); In the formula, is the background radiation noise, is the time noise, is the spatial noise, is the quantization noise, is the thermal noise. In the preferred solution, modeling the background thermal noise includes the following steps: S31: Model the background radiation noise: The background noise is white noise because its power is independent of frequency. The formula is: (4.1); In the formula, is the background emissivity, is the background temperature, is the background radiation area, is the Stefan-Boltzmann constant; S32: Model the time noise: The time noise is the noise caused by time factors during the process of the thermal imager acquiring images. It is defined as the sum of photon noise, dark current noise, and readout noise. The formula is: (5.1); wherein, is the photon noise, is the dark current noise, is the readout noise; S33: Model the spatial noise, which is simulated by sine noise, and the formula is: (6.1); wherein, c is a constant, and I is the input image; S34: Model the quantization noise, which is caused by the quantization process of the image. During the digitization process, due to the limited signal precision, quantization errors may be introduced, and its model is: (7.1); wherein, b is the bit depth of the image; S35: Model the thermal noise, which is generated by the thermal motion inside the sensor, and the formula is: (8.1); wherein, k is the Boltzmann constant, T is the temperature of the sensor, R is the resistance of the sensor, is the noise bandwidth. In the preferred solution, in the S32, specifically: The photon noise is the noise caused by photons reaching the sensor, approximately following a Poisson distribution, and the formula is: (5.2); wherein, is the photoelectric conversion efficiency, is the number of incident infrared photons, is the pixel area of the image sensor, is the integration time; The dark current noise comes from the dark current of the sensor itself, approximately following a Poisson distribution, and the formula is: (5.3); wherein, is the pixel area of the image sensor, is the integration time, and D is the number of electrons generated by the sensor per unit time without external radiation; The readout noise comes from the error in the process of the image sensor reading the signal, and its expression is: (5.4); wherein, is the standard deviation of the readout noise. In the preferred solution, in the S4, model the traditional degradation noise, specifically including the following steps: S41: Model the traditional degradation noise with the formula: (9.1); In the formula, is the compression noise, is the downsampling noise, is the blur kernel noise; S42: Model the compression noise with the formula: (10.1); In the formula, is the original image, is the quantization process, is the discrete cosine transform DCT, is the inverse quantization process, i.e., element-wise multiplication; S43: Model the downsampling with the formula: (11.1); In the formula, is the original image, is the interpolation operation to finally restore to the target size, is the interpolation with a random scaling factor, is the random scaling factor, is the scaling factor to finally restore to the target resolution, is the Sinc low-pass filtering operation; S44: Model the blur kernel with the formula: (12.1); In the formula, is the pixel value of the original image, is the convolution operation, is the added noise, while is the blur kernel. In the preferred solution, the specific steps in S5 are as follows: S51: Without introducing classical degradation noise and background thermal noise, the first-order degradation model formula is: (13.1); In the formula, ILR and IHR are the LR and HR images respectively, and s is the scale factor; S52: The first-order degradation model formula with classical degradation noise and background thermal noise introduced is: (13.2); In the formula, is the degradation function at the x-th level, is the overall degradation function at the x-th level, is the background thermal noise, is the traditional degradation noise; S53: Construct a second-order degradation model, the formula is: (13.3); In the formula, n is the noise introduced during the transmission; S54: Construct an n-order degradation model, the formula is: (13.4). In the preferred solution, after the S1 obtains the dataset in the image area, it further includes data preprocessing on the pictures in the dataset. The data preprocessing includes: normalization, data format conversion, color space conversion, image alignment and cropping, and CLAHE processing; Among them, the core of the CLAHE processing lies in local histogram equalization and contrast limitation. The formula for local histogram equalization is: (14.1); In the formula, is the number of pixels with a gray value of in the local area, is the cumulative sum of the number of all pixels with a gray value less than or equal to , and are the pixel values of the image; Then, the local histogram is equalized through a linear transformation to make the pixel value distribution of the image as uniform as possible. The formula is: (14.2); In the formula, is the pixel value of the original image at the position , is the pixel value of the new image after local histogram equalization processing at the position , is the cumulative distribution function value of the minimum pixel value in the local histogram, is the cumulative distribution function value of the maximum pixel value in the local histogram, is the cumulative distribution function value of the pixel value in the local histogram, is the maximum value of the new pixel value range; The contrast limitation formula is: (14.3); In the formula, is the frequency value of the local histogram after contrast limitation processing, is the histogram of the local area, is the contrast limitation parameter.

[0008] An image model construction system based on random bright spots and irregular stripes, comprising: A random bright spot module, configured to obtain a data set in an image area and model the random bright spots; An irregular stripe module, configured to model the irregular stripes, where the irregular stripes include wide stripes and thin stripes; A background thermal noise module, configured to model the background thermal noise, where the background thermal noise includes background radiation noise, temporal noise, spatial noise, quantization noise, and thermal noise; A traditional degradation noise module, configured to model the traditional degradation noise, where the traditional degradation noise includes blur kernel noise, downsampling noise, and compression noise; An image degradation model module, configured to construct a high-order thermal infrared image degradation model by combining the models established by the above modules. The present invention provides a method and system for constructing an image model based on random bright spots and irregular stripes. By obtaining a data set in an image area, observing and analyzing the distribution characteristics of these bright spots in terms of size, intensity, and position, a mathematical model of random bright spots is established. Then, by analyzing the complex structural characteristics of the irregular stripes in the image, simulating this superimposed spatial noise, and detailedly modeling various background thermal noises that affect image quality, considering common degradation factors in the image processing process and accurately modeling them, and integrating the obtained different types of noise models, a comprehensive high-order thermal infrared image (Bright Random Spots and Irregular Stripes, BRSIS) degradation model is formed, comprehensively simulating various degradation phenomena experienced by thermal infrared images during acquisition, transmission, and processing, improving the efficiency of image processing and enhancing the image resolution. Brief Description of the Drawings

[0009] The following further describes the present invention with reference to the drawings and embodiments: Figure 1 is a flowchart of the method for constructing the image model of the present invention; Figure 2 is a diagram of the high-order BRSIS degradation model of the real-world thermal infrared image of the present invention; Figure 3 is an overall architecture diagram of the noise degradation construction in the thermal infrared image of the present invention; Figure 4 is an effect diagram of the thermal infrared image generated after successively adding degradations in the method of the present invention; Figure 5 is a schematic diagram of the high-order thermal infrared degradation model architecture of the present invention. Detailed Embodiments

[0010] Embodiment 1 As Figures 1-5 shown, a method for constructing an image model based on random bright spots and irregular stripes includes the following steps: S1: Obtain a data set in the image area and model the random bright spots.

[0011] S2: Model the irregular stripes, where the irregular stripes include wide stripes and thin stripes.

[0012] S3: Model the background thermal noise, where the background thermal noise includes background radiation noise, time noise, spatial noise, quantization noise, and thermal noise.

[0013] S4: Model the traditional degradation noise, where the traditional degradation noise includes blur kernel noise, downsampling noise, and compression noise.

[0014] S5: Combine the models established in steps S1 - S4 to construct a high - order thermal infrared image degradation model.

[0015] In this embodiment, by obtaining the data set in the image area, observing and analyzing the distribution characteristics of these bright spots in terms of size, intensity, and position, a mathematical model of random bright spots is established. Then, by analyzing the complex structural characteristics of the irregular stripes in the image, simulating this superimposed spatial noise, and detailedly modeling various background thermal noises that affect the image quality, considering the common degradation factors in the image processing process and accurately modeling them, integrating the obtained different types of noise models to form a comprehensive high - order thermal infrared image degradation model, comprehensively simulating various degradation phenomena experienced by thermal infrared images during acquisition, transmission, and processing, improving the efficiency of image processing and the resolution of the pictures.

[0016] In this embodiment, the problems of thermal infrared image quality are roughly divided into three main parts: 1) Random bright spots and irregular stripes: Random bright spots and irregular stripe noises caused by various factors such as background radiation, time noise, and spatial noise exacerbate the degradation of thermal infrared images; 2) Background thermal noise: Noises and interference factors introduced during the process of the thermal imager acquiring images, such as background radiation noise, time noise, spatial noise, quantization noise, and thermal noise, result in the loss of details and an increase in noise in thermal infrared images, and reduce the image quality; 3) Traditional noise: During the transmission of thermal infrared images, due to factors such as signal attenuation, interference during transmission, image compression (such as JPEG compression), and downsampling, thermal infrared images will further exhibit problems such as blurring, degradation, and noise enhancement. To address the above problems, the model of this embodiment is constructed.

[0017] In the preferred solution, in step S1, a model is established for the random bright spots, specifically: samples of the data set are extracted from multiple groups of image regions with uniform colors, and their distribution characteristics are observed: the sizes and intensities of the random bright spots do not differ much and there is no fixed pattern, and they show a high degree of randomness in position. The formula is: (1.1); In the formula, is the number of random bright spots, is the probability of random bright spot noise, is the image height, is the image width, is the amplitude, is the bright spot radius, is the maximum intensity of the bright spot itself, is the input image.

[0018] is a function describing the distribution and influence range of the bright spot. The function formula is: (1.2); In the formula, is the unit step function, is the calculated pixel to the bright spot center distance: (1.3); In the formula, Ω is the neighboring pixel defined as: (1.4); In the formula, is the maximum intensity of the bright spot itself; is the unit step function used to control which pixels are affected by the bright spot. The formula is: (1.5); In the formula, is the maximum intensity of the bright spot itself, is the calculated pixel to the distance of the bright spot center; The center position of the bright spot The formula is: (1.6); In the formula, U(a,b) is a random number uniformly distributed in the interval [a,b], and are the randomly selected bright spot center coordinates, and are the width and height of the image respectively.

[0019] In this embodiment, a complete random bright spot model is constructed by comprehensively considering the characteristics of the number, probability, size, intensity and position of random bright spots. By using uniformly distributed random numbers to determine the center position of the bright spot, the high randomness of the position of the random bright spot is better adapted, thereby improving the accuracy of removing the random bright spot.

[0020] In the preferred solution, step S2 models the irregular stripes, specifically: Uneven stripes present complex structures in the image, and their low-frequency and high-frequency characteristics are manifested as significant peak and trough changes. Irregular stripe noise is usually manifested as the superposition of low-frequency and high-frequency spatial noise, and the formula is: (2.1); In the formula, is the value of wide stripe noise based on Gaussian distribution, is the value of the generated pinstripe noise.

[0021] The formula for wide stripe noise is: (2.2); In the formula, The amplitude of the wide stripe noise controls the intensity of the stripes. The larger the amplitude, the more obvious the stripes appear in the image; the smaller the amplitude, the fainter the stripes. is the center position of the stripe, which affects the position of the stripe in the image; To control the standard deviation of the stripe width, the smaller the standard deviation, the narrower the stripes, and the larger the standard deviation, the wider the stripes; The horizontal position index (column index) of each pixel in the image.

[0022] The formula for stripe noise is: (2.3); In the formula, That is, the amplitude of the fine stripes, which controls the intensity of the fine stripes. The number of superimposed sine waves determines the complexity of the fine stripes. It is the width of the image, determines the frequency of each sine wave, and affects the density of the stripes.

[0023] In this embodiment, since irregular stripes present a complex structure in an image, irregular stripe noise is regarded as a superposition of low-frequency wide stripe noise and high-frequency thin stripe noise, and the wide stripes and thin stripes are modeled respectively.

[0024] In the preferred embodiment, in step S3, the background thermal noise is modeled, and the overall noise model generated by the thermal imager is as follows: (3.1); In the formula, is the background radiation noise, is the time noise, is the spatial noise, is the quantization noise, is the thermal noise.

[0025] In this embodiment, the overall noise model is used to predict the noise generated by the thermal imager during the imaging process, so as to provide more accurate and reliable data input for subsequent image processing and target detection.

[0026] In the preferred solution, modeling the background thermal noise includes the following steps: S31: Model the background radiation noise: The background noise is white noise because the power is independent of the frequency, and its formula is: (4.1); In the formula, represents the background emissivity, which is a dimensionless parameter of the radiation ability of the background material. is the background temperature, with the unit of Kelvin (K). is the background radiation area, with the unit of square meter (m²). is the Stefan-Boltzmann constant, which is used to describe the intensity of radiation.

[0027] S32: Model the time noise: The time noise is the noise caused by time factors during the image acquisition process of the thermal imager, and is defined as the sum of photon noise, dark current noise, and readout noise. The formula is: (5.1); In the formula, is the photon noise, which is the fluctuation of the number of photons; is the dark current noise, which comes from the fluctuation of the electrons in the sensor itself; is the readout noise, which comes from the reading process of the image sensor.

[0028] The photon noise is the noise caused by photons reaching the sensor, and is approximately Poisson distributed. The formula is: (5.2); In the formula, is the photoelectric conversion efficiency, which is the efficiency of each infrared photon converted into an electron; is the number of incident infrared photons, with the unit of photon number; is the pixel area of the image sensor, with the unit of square meter (m²). is the integration time, with the unit of second (s).

[0029] The dark current noise comes from the dark current of the sensor itself and is approximately Poisson distributed. The formula is: (5.3); wherein, is the pixel area of the image sensor, is the integration time, D is the number of electrons generated by the sensor per unit time without external radiation, and the unit is the number of electrons.

[0030] The read noise comes from the error in the process of the image sensor reading the signal, and the formula is: (5.4); wherein, is the standard deviation of the read noise, and the unit is the number of electrons or the signal intensity.

[0031] S33: Model the spatial noise. The stripe noise in this embodiment is generally a periodic noise. This can be simulated by sine noise, and the formula is: (6.1); wherein, c is a constant, and I is the input image; S34: Model the quantization noise. The quantization noise is caused by the quantization process of the image. During the digitization process, due to the limited signal accuracy, quantization errors may be introduced, and its model formula is: (7.1); wherein, b is the bit depth of the image, which is the number of bits per pixel. The smaller the bit depth, the larger the quantization error.

[0032] S35: Model the thermal noise. Noise is generated due to the thermal motion inside the sensor, and the formula is: (8.1); wherein, k is the Boltzmann constant, T is the temperature of the sensor, and the unit is Kelvin (K); R is the resistance of the sensor, and the unit is ohm (Ω); is the noise bandwidth, and the unit is Hertz (Hz).

[0033] Through the above steps, each component of the background thermal noise is modeled, and based on this, a more comprehensive background thermal noise model is constructed, which improves the performance and accuracy of the system for subsequent image processing.

[0034] In the preferred solution, in S4, the traditional degradation noise is modeled, specifically including the following steps and the formula is: S41: Model the traditional degradation noise, and the formula is: (9.1); wherein, is the compression noise, is the downsampling noise, is the blurring kernel noise.

[0035] S42: Model the compression noise with the formula: (10.1); where, is the original image, is the quantization process, is the discrete cosine transform DCT, is the inverse quantization process, i.e., element-wise multiplication.

[0036] S43: Model the downsampling with the formula: (11.1); where, is the original image, is the interpolation operation that finally restores to the target size, is the interpolation with a random scaling factor, is the random scaling factor, is the scaling factor that finally restores to the target resolution, is the Sinc low-pass filtering operation used to remove high-frequency noise and avoid introducing sharpening artifacts during the interpolation process.

[0037] Downsampling refers to reducing the data volume by decreasing the image resolution, and this process will introduce downsampling noise. Downsampling noise is manifested as the loss of image details or a decrease in image resolution, usually appearing as blurring or jagged artifacts.

[0038] S44: Model the blurring kernel with the formula: (12.1); where, is the pixel value of the original image, is the convolution operation, is the added noise, and is the blurring kernel.

[0039] In this embodiment, the specific modeling methods and parameters in the above steps can be adjusted and optimized according to the actual application scenario and image characteristics.

[0040] In the preferred solution, the specific steps in step S5 are as follows: S51: Without introducing classical degradation noise and background thermal noise, the first-order degradation model formula is: (13.1); where, ILR and IHR are the LR and HR images respectively, and s is the scale factor.

[0041] S52: Introduce the first-order degradation model of classical degradation noise and background thermal noise as follows: (13.2); wherein, is the degradation function of the x-th level, is the overall degradation function of the x-th level, is the background thermal noise, is the traditional degradation noise.

[0042] S53: Construct a second-order degradation model, and the formula is: (13.3); wherein, n is the noise introduced during the transmission process, and the functions and are different, only contains classical degradation factors and does not contain infrared degradation noise; S54: Construct an n-order degradation model, and the formula is: (13.4); As Figure 5 shown, it is a schematic diagram of the overall architecture of the high-order thermal infrared degradation model. Through the above steps, a high-order degradation model is constructed. In practical applications, based on specific application scenarios and image characteristics, the complex process of image degradation is highly simulated, and various degradation situations encountered in the actual transmission and processing of images are accurately described. According to specific application scenarios and image characteristics, the number and type of degradation functions are flexibly adjusted, thereby improving the accuracy and practicality of the model. By introducing multiple degradation steps and noise components, the actual process of image degradation can be more accurately simulated, making the model have higher application value in the fields of image restoration, image enhancement, etc. By using the model of this embodiment for image preprocessing, the performance and accuracy of subsequent image processing algorithms can be more effectively improved.

[0043] In a preferred solution, in step S1, a data set in the image region is obtained, and data preprocessing is performed on the pictures in the data set, including: normalization, data format conversion, color space conversion, image alignment and cropping, and CLAHE (Contrast Limited Adaptive Histogram Equalization) processing; Among them, normalization converts the pixel values in the image from their original range to a standardized range; data format conversion converts the channels and data types to ensure that the data format conforms to the input specifications; color space conversion uniformly converts the thermal infrared image into a grayscale image to eliminate the situation of pseudo-color; image alignment and cropping operations ensure that all input images have a unified size.

[0044] In this embodiment, normalization can be performed using linear transformation, and the formula is: ; where I is the original pixel value, and are the minimum and maximum pixel values respectively.

[0045] Data format conversion: The image can be converted from RGB format to grayscale, or from integer data to floating-point data. Color space conversion: For thermal infrared images, since they are already based on grayscale information, the grayscale channel can be directly extracted or linear transformation can be performed to convert them to grayscale images. If the image is pseudo-color, specific color space conversion methods (such as converting from RGB to grayscale space) are required to extract grayscale information.

[0046] Image alignment and cropping: Image alignment is performed using operations such as rotation, translation, or affine transformation to ensure that key features or objects in the image are in the same position. Cropping uses automatic image registration and cropping algorithms.

[0047] CLAHE processing: A technique for enhancing local contrast of an image through local histogram equalization and contrast limitation, aiming to avoid excessive amplification of noise while enhancing the detail information of the image. The core of CLAHE processing lies in local histogram equalization and contrast limitation. The formula for local histogram equalization is: (14.1); where is the number of pixels with grayscale value in the local area, is the cumulative sum of the number of all pixels with grayscale values less than or equal to , and represent the pixel values of the image.

[0048] Then, local histogram equalization is performed through linear transformation to make the pixel value distribution of the image as uniform as possible. The formula is: (14.2); where is the pixel value of the original image at position , is the pixel value of the new image after local histogram equalization processing at position , is the cumulative distribution function value of the minimum pixel value in the local histogram, is the cumulative distribution function value of the maximum pixel value in the local histogram, is the cumulative distribution function value of the pixel value in the local histogram, is the maximum value of the new pixel value range.

[0049] The contrast limit formula is: (14.3); In the formula, is the frequency value of the local histogram after contrast limit processing, is the histogram of the local area, the frequency corresponding to the gray value of each pixel in the current area, is the contrast limit parameter, used to control the contrast enhancement of the local area.

[0050] In use, the proposed high-order thermal infrared degradation model and other degradation processes are respectively applied to this network to verify the effectiveness of the present invention, as shown in Table 1 and Figure 4 as shown in the content:

[0051] Figure 4 (a) represents the original infrared image. The red area in the image is enlarged and shown on the right side of the image to more clearly display the details and differences.

[0052] Figure 4 (b) represents the super-resolution image after adding traditional degradation noise. It can be seen from the figure that adding only traditional degradation noise cannot completely eliminate the noise interference in the thermal infrared image.

[0053] Figure 4 (c) represents that after adding background thermal noise degradation, the image quality is improved, but the interference of bright spots and irregular stripes still exists.

[0054] Figure 4 (d) It can be seen that after adding random bright spot degradation, the bright spot interference in the thermal infrared image has been removed, but the irregular stripes still affect the image quality.

[0055] As Figure 4 (e) shows that when further adding irregular stripe degradation, this embodiment effectively removes the irregular stripes and obtains a super-resolution infrared image.

[0056] At the same time, objective analysis is carried out using two indicators, NIQE and BRISQUE, as shown in Table 1: The technical solution of this embodiment obtains the lowest NIQE and BRISQUE values, which are reduced by 1.2677 and 7.9233 respectively compared with adding only traditional degradation noise and background thermal noise; It can be seen that the present invention can effectively simulate the degradation process of thermal infrared images when applied to the super-resolution network and improve the ability of the super-resolution network to eliminate noise and stripes in thermal infrared images.

[0057] Example 2 Combined with Example 1 for further illustration, a construction system for an image model based on random bright spots and irregular stripes is proposed, which is applicable to a construction method for an image model based on random bright spots and irregular stripes in Example 1, and includes: A random bright spot module, configured to obtain a data set in an image region and model the random bright spots.

[0058] An irregular stripe module, configured to model the irregular stripes, where the irregular stripes include wide stripes and thin stripes.

[0059] A background thermal noise module, configured to model the background thermal noise, where the background thermal noise includes background radiation noise, time noise, spatial noise, quantization noise, and thermal noise.

[0060] A traditional degradation noise module, configured to model the traditional degradation noise, where the traditional degradation noise includes blur kernel noise, downsampling noise, and compression noise.

[0061] An image degradation model module, configured to construct a high-order thermal infrared image degradation model by combining the various models established by the above modules.

[0062] The above embodiments are only the preferred technical solutions of the present invention and should not be regarded as limitations on the present invention. The protection scope of the present invention should be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for constructing an image model based on random bright spots and irregular stripes, characterized in that: The following steps are involved: S1: Obtain a data set in the image area and model random bright spots; S2: Modeling irregular stripes, wherein the irregular stripes include wide stripes and thin stripes; S3: Modeling background thermal noise, where the background thermal noise includes background radiation noise, time noise, space noise, quantization noise and thermal noise; S4: Modeling traditional degradation noise, where the traditional degradation noise includes blur kernel noise, downsampling noise, and compression noise; S5: Combine the models established in S1-S4 to build a high-order thermal infrared image degradation model.

2. The method for constructing an image model based on random bright spots and irregular stripes according to claim 1, characterized in that: In S1, random bright spots are modeled, specifically: Random bright spots have little difference in size and intensity and no fixed pattern, and are highly random in position. The formula is: (1.1)； In the formula, is the number of random bright spots, is the probability of random bright spot noise, is the image height, is the image width, is the amplitude, is the bright spot radius, is the maximum intensity of the bright spot itself, is the input image; To describe the distribution and influence range function of the bright spot, the formula is: (1.2)； In the formula, is a unit step function, To calculate the pixel To the center of the bright spot Distance: (1.3)； Where Ω is the neighboring pixel, defined as: (1.4)； In the formula, is the current pixel, is the maximum intensity of the bright spot itself; is a unit step function, and the formula is: (1.5)； In the formula, is the maximum intensity of the bright spot itself, To calculate the pixel The distance to the center of the bright spot; The center of the bright spot The formula is: (1.6)； Where U(a,b) is a random number uniformly distributed in the interval [a,b]. and is the coordinate of the randomly selected bright spot center, and are the width and height of the image respectively.

3. The method for constructing an image model based on random bright spots and irregular stripes according to claim 1, characterized in that: The S2 models the irregular stripes, specifically: Irregular stripe noise usually appears as the superposition of low-frequency and high-frequency spatial noise, and the formula is: (2.1)； In the formula, is the value of wide stripe noise based on Gaussian distribution, is the value of the generated fine stripe noise; The formula for wide stripe noise is: (2.2)； In the formula, is the amplitude of wide stripe noise, controlling the intensity of stripes; is the center position of the stripe, which affects the position of the stripe in the image; is the standard deviation of the control stripe width; The horizontal position index (column index) of each pixel in the image; The formula for pinstripe noise is: (2.3)； In the formula, That is, the amplitude of the fine stripes, which controls the intensity of the fine stripes. The number of superimposed sine waves determines the complexity of the fine stripes. It is the width of the image, determines the frequency of each sine wave, and affects the density of the stripes.

4. The method for constructing an image model based on random bright spots and irregular stripes according to claim 1, characterized in that: In S3, the background thermal noise is modeled, and the overall noise model generated by the thermal imager is formulated as follows: (3.1)； In the formula, is the background radiation noise, is the temporal noise, is the spatial noise, is the quantization noise, is thermal noise.

5. The method for constructing an image model based on random bright spots and irregular stripes according to claim 4, characterized in that: Modeling background thermal noise involves the following steps: S31: Modeling background radiation noise: Background noise is white noise because the power of is independent of frequency, and the formula is: (4.1)； In the formula, is the background emissivity, is the background temperature, is the background radiation area, is the Stewart-Boltzmann constant; S32: Modeling of temporal noise: Temporal noise is the noise caused by time factors during the image acquisition process of the thermal imager. It is defined as the sum of photon noise, dark current noise and readout noise. The formula is: (5.1)； In the formula, is the photon noise, is the dark current noise, is the read noise; S33: Model the spatial noise by simulating sinusoidal noise. The formula is: (6.1)； In the formula, c is a constant, I is the input image; S34: Modeling the quantization noise. The quantization noise is caused by the quantization process of the image. During the digitization process, due to the limited signal accuracy, quantization errors may be introduced. The model is: (7.1)； Where b is the bit depth of the image; S35: Modeling thermal noise. Noise is generated due to thermal motion inside the sensor. The formula is: (8.1)； Where k is the Boltzmann constant, T is the temperature of the sensor, and R is the resistance of the sensor. is the noise bandwidth.

6. The method for constructing an image model based on random bright spots and irregular stripes according to claim 5, characterized in that: The S32 is specifically as follows: Photon noise is the noise caused by photons reaching the sensor, which is approximately Poisson distribution, and the formula is: (5.2)； In the formula, is the photoelectric conversion efficiency, is the number of incident infrared photons, is the pixel area of ​​the image sensor, is the integration time; Dark current noise comes from the dark current of the sensor itself, which is approximately Poisson distribution, and the formula is: (5.3)； In the formula, is the pixel area of ​​the image sensor, is the integration time, D is the number of electrons generated by the sensor per unit time when there is no external radiation; The read noise comes from the error in the image sensor reading signal, and its expression is: (5.4)； In the formula, is the standard deviation of the read noise.

7. The method for constructing an image model based on random bright spots and irregular stripes according to claim 1, characterized in that: In S4, the traditional degradation noise is modeled, which specifically includes the following steps: S41: Model the traditional degradation noise, the formula is: (9.1)； In the formula, is the compression noise, is the downsampling noise, is the blur kernel noise; S42: Model the compression noise using the formula: (10.1)； In the formula, is the original image, For the quantification process, is the discrete cosine transform DCT, It is the dequantization process, that is, element-wise multiplication; S43: Model downsampling, the formula is: (11.1)； In the formula, is the original image, To finally restore the interpolation operation to the target size, To interpolate with a random scaling factor, is the random scaling factor, is the scaling factor to restore the target resolution. It is a Sinc low-pass filter operation; S44: Model the blur kernel, the formula is: (12.1)； In the formula, is the pixel value of the original image, is the convolution operation, is the added noise, and is the blur kernel.

8. The method for constructing an image model based on random bright spots and irregular stripes according to claim 1, characterized in that: The specific steps in S5 are as follows: S51: Without introducing classical degradation noise and background thermal noise, the first-order degradation model formula is: (13.1); Where ILR and IHR are LR and HR images respectively, and s is the scale factor; S52: The first-order degradation model formula that introduces classical degradation noise and background thermal noise is: (13.2)； In the formula, is the x-th level degradation function, is the overall degradation function of the xth level, is the background thermal noise, is the traditional degradation noise; S53: Construct a second-order degradation model, the formula is: (13.3)； In the formula, n is the noise introduced during the transmission process; S54: Construct an n-order degradation model, the formula is: (13.4)。 9. The method for constructing an image model based on random bright spots and irregular stripes according to claim 1, characterized in that: After the S1 acquires the data set in the image area, it also includes data preprocessing of the images in the data set, and the data preprocessing includes: normalization, data format conversion, color space conversion, image alignment and cropping, and CLAHE processing; Among them, the core of CLAHE lies in local histogram equalization and contrast limitation. The formula of local histogram equalization is: (14.1)； In the formula, The gray value in the local area is The number of pixels, Gray value is less than or equal to The cumulative sum of all pixel numbers, and is the pixel value of the image; Then, the local histogram is equalized through linear transformation to make the pixel value distribution of the image as uniform as possible. The formula is: (14.2)； In the formula, The original image is at position The pixel value at is the new image after local histogram equalization at position The pixel value at is the cumulative distribution function value of the minimum pixel value in the local histogram, is the cumulative distribution function value of the maximum pixel value in the local histogram, is the pixel value The cumulative distribution function value in the local histogram, is the maximum value of the new pixel value range; The contrast limit formula is: (14.3)； In the formula, is the frequency value of the local histogram after contrast limitation processing, is the histogram of the local area, is the contrast limit parameter.

10. A system for constructing an image model based on random bright spots and irregular stripes, characterized in that: include: A random bright spot module is used to obtain a data set in an image area and model the random bright spots; An irregular stripe module, used for modeling irregular stripes, wherein the irregular stripes include wide stripes and thin stripes; A background thermal noise module, used for modeling background thermal noise, wherein the background thermal noise includes background radiation noise, time noise, space noise, quantization noise and thermal noise; A traditional degradation noise module, used for modeling traditional degradation noise, wherein the traditional degradation noise includes blur kernel noise, downsampling noise and compression noise; The image degradation model module is used to combine the models established by the above modules to build a high-order thermal infrared image degradation model.

Images