An image processing method and system based on imaging multiscale analysis

Abstract

The present invention relates to the field of image processing technology, specifically an image processing method and system based on multi-scale imaging analysis, comprising the following steps: performing non-uniformity detection on different imaging devices based on multi-scale image historical storage information, adjusting the different imaging devices according to the detection results to obtain imaging devices with adjusted uniformity; using the adjusted imaging devices to capture original multi-scale images, and setting an image geometric correction model to perform geometric correction to obtain a multi-scale image set; constructing an image component processing and optimization system, analyzing the image set through the image component processing model and component optimization model in the system to obtain corrected brightness components and reflected light components of different images; establishing an image enhancement processing model based on the corrected brightness components and reflected light components, and obtaining a multi-scale image target set based on the model. The present invention is conducive to device adaptive correction and multi-link collaborative optimization to achieve multimodal image processing and enhancement.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to an image processing method and system based on imaging multi-scale analysis. BACKGROUND

[0002] Image processing technology has become a core technology to support the intelligent upgrading of the industry, and multi-scale image analysis, which can capture both microscopic defects and macroscopic structures, has become a key requirement for imaging equipment. However, existing image processing technology faces related problems and contradictions: due to the pixel response nonlinearity of different imaging devices, the image brightness or color is inconsistent when multiple devices work together, the traditional two-point correction method needs frequent calibration and cannot adapt to temperature changes, and it cannot effectively detect the non-uniformity of different imaging devices, thereby affecting the information fusion effect between imaging devices.

[0003] The scale space information analysis technology constructed in the existing method will lose part of the high-frequency information of the image after each sampling, which causes the target data to not meet the actual image acquisition and processing requirements. At the same time, the current global enhancement method cannot distinguish between light changes and real object reflection characteristics, which will cause the loss of details in overexposed areas, and the local method is easy to introduce block artifacts, which are significantly below the human visual perception threshold, and it is difficult to meet the basic requirements of high-precision scenes for image information.

[0004] In order to solve the problems of device regulation function lag, scale space information loss and image enhancement distortion in the existing technology, it is necessary to optimize the existing image processing method to realize the adaptive correction of the imaging device, the effective preservation and fusion enhancement of the multi-scale image information. SUMMARY

[0005] In view of the deficiencies of the existing method and the needs of practical application, in order to solve the problems of imaging device non-uniformity compensation lag, scale space information attenuation and multi-modal enhancement distortion in the existing method, and further guarantee the image imaging quality and processing efficiency.

[0006] In one aspect, the present application provides an image processing method based on imaging multi-scale analysis, the method comprising the following steps: obtaining multi-scale image historical storage information, detecting non-uniformity of different imaging devices based on the multi-scale image historical storage information and obtaining non-uniformity detection results of the different imaging devices, adjusting the different imaging devices according to the non-uniformity detection results to obtain imaging devices after uniformity adjustment; collecting original multi-scale images by using the imaging devices after uniformity adjustment, setting an image geometric correction model to perform geometric correction on the original multi-scale images to obtain a multi-scale image set; constructing an image component processing and optimization system, analyzing the multi-scale image set by using an image component processing model and a component optimization model in the image component processing and optimization system to obtain corrected luminance components and corrected reflected light components of different images in the multi-scale image set; establishing an image enhancement processing model according to the corrected luminance components and the corrected reflected light components, and enhancing different images in the multi-scale image set by using the image enhancement processing model to obtain a multi-scale image target set.

[0007] The present application includes device adjustment, image multi-scale analysis and image enhancement processing links, which are beneficial to guarantee the effectiveness of the output images of the imaging device and ensure the accuracy of the image information related to different target monitoring objects.

[0008] The present application dynamically analyzes the mean and standard deviation of the difference image, provides a key reference for imaging device adjustment, and helps to meet the functional requirements of the imaging system and the accuracy of the collected information; the coordinate correction analysis of the image information breaks through the static correction limitations of the traditional method, can effectively analyze the coordinate mapping relationship of different images, and further guarantees the unity and effectiveness of the images.

[0009] The present application obtains the luminance component by using the luminance component prediction function, and then corrects and optimizes it by using the luminance component optimization function, which can more accurately adjust the image brightness, avoid over-bright or over-dark areas, and is beneficial to improve the definition of different images; the logarithmic transformation of the theoretical model converts the complex multiplication relationship into a simple addition relationship, simplifying the calculation process. At the same time, it can better simulate the nonlinear perception characteristics of light intensity, making the processed image more natural and real.

[0010] The present application obtains the saturation component, the image illumination component and the component adjustment factor based on the multi-scale image set, so that the component optimization model can learn the characteristics and rules of different types of images, thereby being able to process different types of images and meet the diversified application requirements.

[0011] The application analyzes spatial distance, structural similarity and gradient difference according to the device reference image and the multi-scale image set, and establishes an image loss detection function, which can comprehensively and accurately evaluate the loss degree of the enhanced multi-scale image set, and comprehensively judge whether the image has information loss, distortion or over-processing in the enhancement process, so as to effectively guarantee the image quality.

[0012] In order to efficiently execute the image processing method based on imaging multi-scale analysis provided by the application, the application further provides an image processing system based on imaging multi-scale analysis, which comprises a processor, an input device, an output device and a memory, and the processor, the input device, the output device and the memory are connected with each other, wherein the memory is used for storing a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the image processing method based on imaging multi-scale analysis as described in the first aspect of the application. The image processing system based on imaging multi-scale analysis has compact structure and stable performance, and can stably execute the image processing method based on imaging multi-scale analysis provided by the application, thereby improving the overall applicability and practical application ability of the application. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 The flow chart of the image processing method based on imaging multi-scale analysis of the application;

[0014] Figure 2 The image contrast diagram of different imaging devices based on imaging multi-scale analysis of the application;

[0015] Figure 3 The image pixel distribution diagram with or without geometric distortion based on imaging multi-scale analysis of the application;

[0016] Figure 4 The structure diagram of the image processing system based on imaging multi-scale analysis of the application. DETAILED DESCRIPTION

[0017] Reference Figure 1 In order to effectively solve the problems of imaging device non-uniformity compensation lag, scale space information attenuation and multi-modal enhancement distortion in the prior art, further optimize the image imaging quality and processing efficiency, and ensure that the related image quality meets the actual application requirements, the application provides an image processing method based on imaging multi-scale analysis, which comprises the following steps:

[0018] S1, acquire multi-scale image historical storage information, detect non-uniformity of different imaging devices based on the multi-scale image historical storage information, and obtain non-uniformity detection results of the different imaging devices, and then adjust the different imaging devices according to the non-uniformity detection results to obtain the imaging devices after uniformity adjustment.

[0019] Firstly, multi-scale image historical storage information is acquired.

[0020] In the embodiment, the multi-scale image historical storage information required is acquired based on a multi-scale image historical storage database. The database uses efficient and reliable storage technology to store the collected image data. Since the data volume of the multi-scale image historical storage information is huge, a distributed storage system is selected as the storage scheme, and the image data is stored on multiple storage nodes. This not only significantly improves the overall storage capacity of the image data, but also effectively improves the read-write performance of the image data, ensuring efficient access and effective processing of the image information. It should be noted that the plurality in the present application includes two or more.

[0021] At the same time, in order to protect the security and integrity of the image information data, an information backup strategy is formulated. It can periodically backup the storage database according to the established backup period, so as to prevent the risk of image information loss caused by unpredictable factors such as hardware failure and natural disaster, and provide technical support for long-term preservation of image information.

[0022] In the data storage stage, the image data has been indexed and labeled. Therefore, when acquiring the multi-scale image historical storage information, the multi-scale image historical storage information can be acquired by searching and querying the image information. In one embodiment, the relevant information is searched and queried according to the key attributes such as the acquisition time, scene type, and scale range of the image. With the help of the indexing mechanism, the required multi-scale image historical storage information can be quickly located, improving the efficiency and accuracy of acquiring the multi-scale image historical storage information.

[0023] Then, non-uniformity of different imaging devices is detected based on the multi-scale image historical storage information, and non-uniformity detection results of the different imaging devices are obtained.

[0024] The environment status and image acquisition requirement are obtained according to the multi-scale image historical storage information. Based on the obtained multi-scale image historical storage information, the environment status and image acquisition requirement are analyzed. In the embodiment, the multi-dimensional indexes such as scene features, light conditions, and object distribution in the multi-scale image historical storage information are comprehensively analyzed to determine the change trend and characteristics of the environment in the current and future period of time, and the specific application scene, monitoring target, precision requirement, frequency setting, and other factors are combined to determine the specific requirement of image acquisition, including but not limited to the imaging scale, resolution, color mode, and other key parameters of the image.

[0025] The imaging device reference image is set based on the environment status and image acquisition requirement. In the embodiment, the imaging device reference image is pre-set mainly according to the environment status and image acquisition requirement. This process needs to be based on the dynamic change of the environment status and the specific task requirement of image monitoring, and the high-quality image in the multi-scale image historical storage information is taken as a reference blueprint, and the existing image processing algorithm and imaging theory are combined to construct the imaging device reference image that meets the actual requirement. The multi-scale image historical storage information provides an information basis for non-uniformity detection of different imaging devices, the environment status and image acquisition requirement are analyzed based on the historical image data, the imaging device reference image is set, and data support is provided for non-uniformity detection of the imaging device. At the same time, a reference basis is provided for performance evaluation and optimization of different imaging devices in the future.

[0026] The difference image mean value of different images in the multi-scale image historical storage information is analyzed according to the imaging device reference image and the multi-scale image historical storage information.

[0027] The difference image mean value of different images in the multi-scale image historical storage information is analyzed based on the imaging device reference image and the multi-scale image historical storage information. The difference between the imaging device reference image and the monitoring image of different imaging devices in the multi-scale image historical storage information is calculated first, then the difference image is binarized, and the processed image is properly saved.

[0028] Considering the inherent differences of different imaging devices under different conditions, in order to exclude the interference of external conditions on the image contrast result and ensure the accuracy and reliability of the analysis result, not only one image can be used for mean value calculation, but also multiple images of different imaging devices under the same condition should be obtained, based on which the non-uniformity status of different imaging devices can be more accurately judged.

[0029] In the calculation of the difference image, the imaging device reference image is taken as the reference, and the imaging device monitoring image in the multi-scale image history storage information is subtracted from the imaging device reference image, and the difference image obtained at this time presents the characteristics of a noise image with a black background; then the difference image is binarized, and the threshold value is set to 1, so that all non-zero noises are displayed in the form of white pixels on the image.

[0030] The specific implementation steps of the correlation difference image mean are as follows:

[0031] Step 1: Design and calculation of total number of binary image pixels.

[0032] The total number of pixels of the binary image is marked as , which can be calculated by the following formula:

[0033]

[0034] , wherein represents the total number of pixels of the binary image, represents the number of pixels of the binary image in the horizontal direction (x-axis direction), represents the number of pixels of the binary image in the vertical direction (y-axis direction).

[0035] The total number of pixels of the binary image reflects the size of the image and is a basic index for measuring the amount of image information. In the calculation of the standard deviation and mean of the difference image, the total number of pixels is used as the denominator for related operations and relationship analysis. The average processing of the pixel value deviation can reflect the statistical quantity of the overall image characteristics.

[0036] The number of pixels of the binary image in the horizontal direction (x-axis direction) determines the resolution of the image in the horizontal dimension, wherein The larger the value, the more details the image can present in the horizontal direction, and the greater the image width.

[0037] The number of pixels of the binary image in the vertical direction (y-axis direction) determines the resolution of the image in the vertical dimension, wherein The larger the value, the more details the image can present in the vertical direction, and the higher the image height.

[0038] Step 2: Calculation of the mean of the difference image.

[0039] In the embodiment, the number of non-zero pixels after subtraction is set to , and in the case of binarization and threshold value of 1, the pixel value of the non-zero pixel is uniformly marked as , and the pixel values of the remaining pixels are 0 (in the binarized image, the background black pixel value is 0, i.e. ).

[0040] According to the definition of mean, which is the sum of all data divided by the number of data, the formula of the mean of difference image can be derived, that is, all pixel values are composed of the sum of non-zero pixel values and the sum of zero pixel values, the sum of non-zero pixel values is , the sum of zero pixel values is , and the sum of all pixel values is .

[0041] The total number of pixels of the binary image is , and the formula of the mean of difference image is as follows:

[0042]

[0043] wherein, represents the mean of difference image, represents the number of non-zero pixels after subtraction, represents the pixel value of non-zero pixel, represents the background black pixel value, represents the total number of pixels of the binary image.

[0044] When the binary threshold is 1, that is, , , the above formula can be further simplified as:

[0045]

[0046] Then, the standard deviation of different images in the multi-scale image historical storage information is analyzed according to the mean of difference image.

[0047] Third step: calculate the standard deviation based on the mean of difference image and the total number of pixels of the binary image. The standard deviation can measure the dispersion degree of a group of data, which has important significance in image analysis, and the specific implementation process is as follows:

[0048] First, calculate the square of the difference between each data and the mean. For non-zero pixels, the square of the difference between each data and the mean is ; for zero pixels, the square of the difference between each data and the mean is .

[0049] Variance is the average of the square of the difference between each data and the mean, let the variance be , and according to the above content, the formula of the variance is as follows:

[0050]

[0051] wherein, represents the variance, represents the number of pixels after subtraction that are not zero, represents the mean value of the difference image, represents the pixel value of the non-zero pixel, represents the total number of pixels of the binary image, represents the background black pixel value.

[0052] Substituting , , into the above formula, after simplification, we get:

[0053]

[0054]

[0055] Then, the standard deviation σ is calculated Since the variance is the square of the standard deviation, the standard deviation is the square root of the variance, so the standard deviation The calculation formula is as follows:

[0056] To further optimize the formula, set , and , then the standard deviation formula can be simply expressed as: This optimized formula form is more concise, where is convenient for understanding and calculation. This formula is not only suitable for the special case of binary threshold value 1, but also for other binary threshold value settings. Only need to substitute the corresponding and values into the variance calculation formula for derivation, and the accurate standard deviation formula can also be obtained.

[0057] Through analysis, we know that , and Further analysis shows that the function is not monotonically increasing on the interval , but first increases and then decreases, and the function reaches the maximum value when . That is, the larger the standard deviation, the more noise in the image at this time, which means that the non-uniformity of the imaging device is more obvious.

[0058] In summary, in the case of black background pixels, the standard deviation σ of the binary image is calculated, and the larger the standard deviation , the more noise elements the image obtained by the imaging device has, and the more significant the non-uniformity of the imaging device. Based on the above standard deviation and multi-scale image historical storage information, different imaging devices can be comprehensively and accurately detected for non-uniformity, and the non-uniformity detection results of different imaging devices are obtained, which provides a strong basis for performance evaluation and feature optimization of different imaging devices.

[0059] Finally, the different imaging devices are adjusted according to the non-uniformity detection results to obtain the imaging devices after uniformity adjustment.

[0060] Based on the above, in this embodiment, a plurality of groups of images of different imaging devices under the same conditions are obtained. For image comparison of different imaging devices, please refer to Figure 2 , wherein (a) represents the reference image of the imaging device, (b) represents the generated image of the imaging device 1 under the same conditions, (c) represents the generated image of the imaging device 2 under the same conditions, and (d) represents the generated image of the imaging device 3 under the same conditions.

[0061] After obtaining the non-uniformity detection results of different imaging devices, in this embodiment, the different imaging devices are adjusted according to the detection results. For the imaging device with relatively significant non-uniformity, in this embodiment, the standard deviation of the binary image is greater than or equal to a pre-set threshold value (the threshold value can be reasonably adjusted according to the type, performance requirement and expected application scenario of the different imaging devices, according to a large amount of experimental data and actual application scenario analysis) , the imaging device is defined as the imaging device with large standard deviation . When the standard deviation of the imaging device reaches or exceeds the threshold value, the hardware part of the imaging device is first checked to check whether the imaging sensor is damaged, aged or improperly installed, and the non-uniform response of the sensor is an important factor causing the non-uniformity of the image. If it is found that the sensor has a problem, it will be replaced or re-calibrated and installed in time.

[0062] At the same time, for the optical system of the imaging device, the optical performance of the lens needs to be checked, including but not limited to the aberration and distortion of the lens. The imaging quality of the lens is evaluated using an optical detection device. If there is a problem such as too large aberration or distortion, the imaging quality will be improved by adjusting the optical parameters of the lens or replacing a suitable lens to reduce the non-uniformity of the image caused by optical factors.

[0063] For the imaging device with relatively small non-uniformity, the hardware and software of the imaging device need to be regularly maintained and optimized, such as regular cleaning and calibration of the sensor, and updating of the software system of the imaging device to ensure that it is always in the best working state and avoid the gradual aggravation of the non-uniformity problem caused by long-term use or software vulnerabilities.

[0064] As described above, in the case of black background pixels, the standard deviation of the binary image is calculated, and if the standard deviation The larger the value, the more noise elements the image obtained by this imaging device contains, that is, the more significant the non-uniformity of the imaging device. In this embodiment, based on the standard deviation and multi-scale image history storage information, comprehensive and accurate non-uniformity detection can be performed on different imaging devices. The obtained non-uniformity detection results provide a reference for imaging device performance evaluation and feature optimization. Through this series of targeted adjustment measures, the non-uniformity of the imaging device can be effectively improved, resulting in an imaging device with adjusted uniformity, thereby improving image quality and meeting the diverse imaging needs of different fields.

[0065] In an optional embodiment, the non-uniformity of the imaging device is determined according to the size of the image standard deviation.

[0066] In the embodiment, the non-uniformity of different imaging devices is determined based on the size of the image standard deviation. The specific operation process is as follows:

[0067] First, the images collected by different imaging devices are processed according to the above analysis process to calculate the image difference; then binarization is performed; and finally the standard deviation of the binary image is obtained.

[0068] After performing the above processing on images of different imaging devices, the standard deviations of binary images of different imaging devices can be obtained. The standard deviations of binary images obtained in the embodiment are 0.0720, 0.0708, 0.0699 and 0.0711 respectively. In order to more scientifically evaluate the non-uniformity of imaging devices, the distribution test coefficient is introduced. , and define the standard deviation reference index as , and its calculation formula is:

[0069]

[0070] in, represents the standard deviation reference indicator, represents the distribution test coefficient, Represents the standard deviation of binary images of different imaging devices, Indicates the number of imaging devices.

[0071] The above standard deviation reference index is The calculation content is as follows:

[0072]

[0073] in, represents the standard deviation reference indicator, , It represents the distribution test coefficient, and its value is determined based on experimental data and statistical laws. The standard deviation reference indicator is obtained by statistically analyzing the calculation results of images from different imaging devices and is highly representative and reliable.

[0074] According to the standard deviation reference index, the image of the imaging device is judged for eligibility. If the standard deviation of the imaging device image is greater than , it is determined that the imaging device image is unqualified; if the standard deviation of the imaging device image is less than , it is determined that the imaging device image is qualified. At the same time, in order to evaluate the performance of the imaging device more carefully, different judgment levels can be set to distinguish the quality of the imaging device image.

[0075] Further, through the non-uniformity detection results of different imaging devices, it can be determined which devices have non-uniformity problems, so that the hardware and software thereof can be checked, maintained and optimized to improve the performance of the imaging device and obtain the imaging device after uniformity adjustment.

[0076] In the embodiment, the non-uniformity of the imaging device is determined and adjusted according to the image standard deviation, which can effectively reduce the noise and interference in the image, make the edge of the imaging image clearer, and the details more abundant. At the same time, based on the non-uniformity determination and adjustment method of the above image standard deviation, the imaging performance of the imaging device can meet the image acquisition standards and specifications of different industries, and ensure the reliability and effectiveness of different imaging devices in specific application scenarios.

[0077] S2, using the above uniformity adjusted imaging device to collect original multi-scale images, setting an image geometric correction model to perform geometric correction on the original multi-scale images, so that a multi-scale image set can be obtained, and the specific implementation steps and contents are as follows:

[0078] In this embodiment, the original multi-scale image is preprocessed, and a wavelet transform denoising method is used. The specific operation process is as follows: first, wavelet decomposition is performed on the original multi-scale image, and through this process, wavelet coefficients at different scales can be obtained. Since noise has specific statistical characteristics, threshold processing is performed on the wavelet coefficients according to the related characteristics, and the wavelet coefficients corresponding to the noise are removed. Finally, the wavelet reconstruction technology is used to obtain the processed original multi-scale image.

[0079] In order to realize image geometric correction, accurate image coordinate information is needed, and noise will affect the accuracy of coordinate information and reduce the accuracy of geometric correction results. The preprocessing process in the embodiment helps to improve the geometric correction effect, so that the corrected image is more consistent with the actual scene.

[0080] In the field of image processing technology, geometric distortion is a problem that cannot be ignored. The coordinates of corresponding image elements of images with geometric distortion and images without geometric distortion are significantly different. Please refer to Figure 3The right side (b) presents the pixel distribution diagram of the image without geometric distortion, in which the pixels present a uniform and equidistant distribution state; and the left side (a) shows the pixel distribution diagram of the remote sensing image with geometric distortion, in which the pixel distribution presents the characteristics of non-uniform and non-equidistant, and in order to accurately obtain the pixel coordinates without geometric distortion on the image with geometric distortion, the spatial conversion of the two image coordinate systems must be carried out.

[0081] Firstly, the image coordinate information before conversion is obtained according to the original multi-scale image.

[0082] The spatial conversion between different coordinate systems can be described by a binary n-th order polynomial, and the spatial conversion expression of the image coordinate before conversion is as follows:

[0083]

[0084] wherein, and represent the coordinates of the image before conversion (with geometric distortion), and represent the coordinates of the image after conversion (without geometric distortion), and represent the polynomial coefficients, .

[0085] The image coordinate information is the premise of establishing the image geometric correction model, and the spatial conversion relationship between different coordinate systems can be analyzed by using a binary n-th order polynomial after the image coordinate information is determined, which is beneficial to the subsequent establishment of the image geometric correction model.

[0086] In the binary polynomial, for the n-th order polynomial, when i changes from 0 to n, the value range of j needs to satisfy , that is, j changes from 0 to The above value range setting can ensure that all possible n-th order and below polynomial terms are covered, so that the polynomial can more accurately describe the conversion relationship between the coordinates.

[0087] wherein, and respectively represent the horizontal coordinate and the vertical coordinate of the pixel in the image with geometric distortion. In the remote sensing image, due to the comprehensive influence of various factors such as sensor error, earth curvature, atmospheric refraction, etc., the actual position of the pixel in the image deviates from its ideal position, and describe the coordinates after the deviation, and by solving the polynomial coefficients, the coordinates and can be converted into the coordinates without geometric distortion, and thus the correction of the geometric distortion image is realized.

[0088] and respectively represent the horizontal coordinate and the vertical coordinate of a pixel in the image without geometric distortion, that is, the position coordinate of the pixel in the ideal state, which serves as the target coordinate, and the coordinate of the image with geometric distortion and are linked, and the coordinates with distortion can be mapped to the coordinates without distortion through a polynomial transformation.

[0089] and are coefficients of a binary n-th order polynomial, and can be used to describe the mathematical relationship between the coordinates of the image with geometric distortion and the coordinates of the image without geometric distortion Different combinations of i and j correspond to different polynomial terms, and determine the weight or influence degree of each term in the coordinate transformation. In actual applications, a certain number of control points are selected, and mathematical methods such as least squares are used to solve the above coefficients, so that the spatial transformation expression of the image coordinates before transformation can be accurately established, and effective correction of the image with geometric distortion can be further realized.

[0090] Then, the image control points are selected, and the image geometric correction model is established according to the relevant image control points and the image coordinate information before transformation.

[0091] In actual applications, in order to solve the polynomial coefficients and , a certain number of image control points are selected. The above control points are points that can be accurately corresponded on the image with geometric distortion and the image without geometric distortion, and have obvious and easily identifiable features. Among them, road intersections, building corner points, river intersection points, mountain peak points and bridge end points can be selected according to the actual application scene or environmental conditions.

[0092] In this embodiment, m image control points are selected. For each control point k (k = 1, 2,..., m), according to the spatial transformation expression, two equations about and can be obtained respectively. By integrating the equations corresponding to the m control points, a linear equation group about and can be constructed. The equation group is the mathematical expression form of the image geometric correction model, and the specific expression is as follows:

[0093]

[0094] wherein, and denote the horizontal and vertical coordinate values of the kth control point in the distorted coordinate system, denote the highest degree of the binomial polynomial, and denote the coordinates of the transformed (without geometric distortion) image, and denote the polynomial coefficients,

[0095] Image control points are a key link in establishing an image geometric correction model. Selecting corresponding points on the image with geometric distortion and the image without geometric distortion, and using the above control point coordinate information to construct a linear equation system about the polynomial coefficients, i.e., the image geometric correction model, is helpful to realize the conversion of image coordinate information and obtain the geometric corrected image coordinate information.

[0096] The horizontal and vertical coordinate values of the kth control point in the original coordinate system (i.e., the coordinate system with distortion) in the image with geometric distortion, the image control point is a point that can be accurately corresponded on the image with geometric distortion and the image without geometric distortion. By selecting a sufficient number of control points and reasonable distribution, the mapping relationship between the two image coordinates can be established, thereby realizing the correction of the distorted image.

[0097] Combining the equations of the m image control points forms a linear equation system, which can be solved by numerical calculation methods such as least squares method, thereby determining the specific values of the polynomial coefficients and The least squares method finds the best function matching of data by minimizing the sum of squares of errors, which can effectively improve the accuracy and stability of coefficient solving.

[0098] In practical applications, the number of image control points m needs to meet certain conditions. Among them, m should be greater than the total number of polynomial coefficients to ensure that the linear equation system has a unique solution and has certain anti-interference ability and stability. If m is too small, it may lead to underdetermined equation system and cannot obtain accurate solution; while if m is too large, it may increase the calculation amount and calculation complexity, and may also introduce more noise and errors.

[0099] The highest degree of the binomial polynomial is one of the key parameters that affect the performance of the coordinate transformation model. The value of n determines the number of terms contained in the polynomial, which in turn affects the complexity and accuracy of the model. As n increases, the number of terms contained in the polynomial increases, and the model can fit more complex coordinate transformation relationships, but at the same time it may lead to overfitting of the model and reduce the generalization ability of the model.

[0100] In one optional embodiment, when n=1, the polynomial is a linear polynomial, consisting only of a constant term and a linear term. This results in a relatively simple model suitable for situations where coordinate transformations are relatively simple. When n=2, the polynomial is a quadratic polynomial, containing not only a constant term and a linear term but also a quadratic term. This allows for better description of nonlinear coordinate transformations, but also increases computational complexity. Therefore, in actual use, the appropriate value of n should be selected based on factors such as the degree of geometric distortion of the image and the distribution of control points.

[0101] Next, the image coordinate information before transformation is transformed through the image control points and the image geometric correction model to obtain the image coordinate information after geometric correction.

[0102] Based on the selected image control points and the constructed image geometric correction model, a coordinate transformation operation is performed on the coordinate information of the image before transformation to obtain the coordinate information of the image after geometric correction.

[0103] In this embodiment, n=1 is used as an example for detailed description. The optimized expression is as follows:

[0104] When n=1, the optimized expression is:

[0105]

[0106] Three image control points are selected and recorded as ( , , , )、( , , , )、( , , , ), substituting the coordinates of the three image control points into the above expressions, we can obtain the following equations:

[0107]

[0108] A system of equations is a system of equations about unknown coefficients 、 、 、 、 、 A system of linear equations. Among them, The constant coefficient reflects the and The initial offset of the x coordinate when none of them have any effect;

[0109] is the coefficient of the first term of , which indicates the corresponding change in the x coordinate when the y coordinate increases by one unit; is the coefficient of the first term of , which means the change in the x coordinate when the y coordinate increases by one unit; similarly, , , have similar meanings to , , , , , mainly act on the y coordinate.

[0110] Using the least squares method and other methods to solve the equation set can accurately obtain the specific values of , , , , , After determining the above coefficients, the spatial transformation of the image coordinates is realized, and the image with geometric distortion can be corrected to a standard image without geometric distortion, providing accurate and reliable data basis for subsequent image analysis and processing.

[0111] Finally, based on the geometric corrected image coordinate information, a multi-scale image set is obtained, which provides data support for subsequent multi-scale image fusion, feature extraction, image enhancement, etc.

[0112] In practical applications, the original multi-scale images have different scales. In the process of remote sensing image processing, there may be images of different resolutions in the same area; in medical image analysis, there may be slice images of different magnifications. Based on the above image geometric correction method, different images in the original multi-scale images are respectively subjected to geometric correction operation.

[0113] For each original image, according to the steps of selecting control points, constructing correction model, solving coefficients, and performing coordinate transformation as described above, the geometric corrected image coordinate information is obtained. The image coordinate information corresponding to the different scale images after geometric correction is collected and arranged, and the different scale images after correction together constitute a multi-scale image set. The multi-scale image set not only contains detailed information of objects under different scales, but also unifies and standardizes the coordinate relationship between different scale images after geometric correction, which can provide high data support for subsequent multi-scale image fusion, feature extraction, target recognition, etc. tasks, thereby effectively improving the accuracy and reliability of image processing.

[0114] Further, in the embodiment, the detection and correction mode of the original image is only an optional condition of the present application, and in other one or some embodiments, the detection and correction mode of the original image can be optimized according to the actual acquisition conditions of the imaging device and the characteristic attributes of the image. Different imaging devices have differences in imaging principles, sensor performance, optical systems, etc., and the detection and correction mode can be optimized according to the actual acquisition conditions of the imaging device, so that a more suitable correction algorithm can be designed according to the specific distortion characteristics of the device, thereby improving the correction accuracy and making the corrected image more consistent with the actual scene.

[0115] S3, constructing an image component processing and optimization system, analyzing the multi-scale image set through the image component processing model and the component optimization model in the image component processing and optimization system to obtain the corrected luminance component and the corrected reflected light component of different images in the multi-scale image set, and the specific implementation steps and contents are as follows:

[0116] In the above image component processing model, the luminance component prediction function and the reflected light component prediction function are set.

[0117] In an optional embodiment, the luminance component prediction function is set in the image component processing model.

[0118] First, the RETINEX theoretical model is introduced, and the RETINEX theoretical model needs to be logarithmically transformed in the embodiment to obtain the luminance component and the reflected light component decomposition function.

[0119] The RETINEX theoretical model can decompose the image into a luminance component and a reflected light component, providing a basis for further analyzing the image characteristics, and facilitating subsequent separation and calculation of the luminance component and the reflected light component. In the embodiment, the RETINEX theoretical model is logarithmically transformed to obtain the corresponding luminance component and reflected light component decomposition function.

[0120] According to the RETINEX theoretical model, different image sequences in the multi-scale image set are decomposed into a luminance component sequence and a reflected light component sequence, and the specific expression is as follows:

[0121]

[0122] Wherein, represents the pixel component at different image coordinate positions represents the luminance component at different image coordinate positions represents the reflected light component at different image coordinate positions represents the luminance component at different image coordinate positions represents the reflected light component at different image coordinate positions represents the reflected light component at different image coordinate positions represents the matrix multiplication symbol.

[0123] The reflected light component reflects the reflection characteristics of the object surface material to the light, and is related to the inherent properties of the object, such as the material and color.

[0124] The pixel component refers to the intensity value of the actually observed image at the point, which is mainly the result of the joint action of the brightness component of the light and the reflected light component of the object surface.

[0125] The brightness component mainly depends on the properties of the light source, such as the intensity, color, direction of the light source, and the distribution of the light in the scene. In the area under strong light, the brightness component is larger, while in the shadow area, the brightness component is smaller. The brightness component mainly depends on the properties of the light source, such as the intensity, color, direction of the light source, and the distribution of the light in the scene. In the area under strong light, the brightness component is larger, while in the shadow area, the brightness component is smaller.

[0126] In order to simplify the calculation process and realize the dynamic range compression of the multi-scale image set, the above RETINEX theoretical model is logarithmically transformed to obtain the brightness component and the reflected light component decomposition function. After transformation, the RETINEX theoretical model can be converted into the following form:

[0127] The original RETINEX theoretical model is logarithmically transformed to obtain the function for brightness component and reflected light component decomposition as follows:

[0128]

[0129] wherein, represents the logarithmic form of represents the logarithmic form of represents the logarithmic form of represents the logarithmic form of represents the logarithmic form of represents the logarithmic form of

[0130] The numerical range of the pixel component, the reflected light component and the brightness component is compressed by logarithmic transformation, which makes the calculation more convenient and can better handle the dynamic range problem of the image, avoiding the numerical calculation problem caused by the large difference in light intensity.

[0131] Further simplified as:

[0132]

[0133] wherein, represents the reflected light component at different image coordinate positions represents the pixel component at different image coordinate positions represents the brightness component at different image coordinate positions represents the brightness component at different image coordinate positions represents the brightness component at different image coordinate positions represents the brightness component at different image coordinate positions

[0134] ​In the embodiment, the RETINEX theoretical model is optimized and deduced, different images are decomposed into a brightness component (reflecting the illumination condition) and a reflected light component (reflecting the reflection characteristics of the object surface to light), and the influences of the above two factors can be clearly separated.

[0135] In the original RETINEX theoretical model, there is a multiplication relationship among the pixel component, the brightness component and the reflected light component, and the calculation is relatively complex. After the logarithmic transformation in the embodiment, the multiplication operation is converted into a subtraction operation, which greatly simplifies the calculation process. When large-scale image data processing is performed, the processing speed can be significantly improved, and the consumption of calculation resources can be reduced.

[0136] Meanwhile, the logarithmic transformation can compress the numerical range of the pixel component, the reflected light component and the brightness component. By taking the logarithm, the relevant values can be compressed into a more reasonable range, avoiding numerical calculation problems such as overflow or precision loss caused by excessively large or small values, thereby improving the stability and accuracy of the image decomposition and calculation results.

[0137] Then, different scale Gaussian filters are set, and local feature measurement values and weight adjustment factors of different images are obtained according to the multi-scale image set.

[0138] In the prior art, the original information of the image can be directly used to estimate the brightness component. In the embodiment, a multi-scale Gaussian filter is introduced to more accurately capture the brightness information of different images. Specifically, k Gaussian filters of different scales are used , wherein , represents the standard deviation of the Gaussian kernel at the i-th scale.

[0139] To further optimize the accuracy of the image component estimation result and the adaptability of different images, the local feature measurement values of different images are fully considered in the embodiment. Since different images have significant differences in features, such as edge regions and smooth regions in different images, their requirements for extracting brightness information are completely different. Therefore, the weight of the Gaussian filtering result of each scale needs to be dynamically adjusted according to the local region features of different images in the multi-scale image set .

[0140] Let be the local feature measurement value of different images at point , the local gradient amplitude or local variance can be used to measure it, and the local feature adjustment factor of the weight is defined as , which is limited in the range of , and is positively correlated with , that is, the more significant the local feature is, The greater the value of the dynamic weight is, the greater the weight of the i-th scale Gaussian filtering result is.

[0141] Let be the dynamic weight of the i-th scale Gaussian filtering result at point , in the initial setting, , the calculation formula of the dynamic weight based on the adjustment factor related information is as follows:

[0142]

[0143] wherein, be the dynamic weight of the i-th scale Gaussian filtering result at point , be the preset basic weight of the i-th scale Gaussian filtering result, which can be set according to the general importance of the brightness information extracted from different scale images, be the adjustment factor of the weight of the local feature of different images, be the k different scale Gaussian filters, be the preset basic weight of the j-th scale Gaussian filtering result.

[0144] Finally, the brightness component and the reflected light component decomposition function are derived according to the local feature measurement value and the weight adjustment factor, so as to obtain the brightness component prediction function in the image component processing model.

[0145] Further based on , and combined with the local feature measurement value and the weight adjustment factor of different images, the brightness component prediction function is set, and the brightness component prediction function is as follows:

[0146]

[0147] wherein, be the brightness component prediction result of different images, be the k different scale Gaussian filters, be the dynamic weight of the i-th scale Gaussian filtering result at point , be the pixel component at the coordinate position of different images, be the convolution operator symbol, be the different scale Gaussian filters.

[0148] The above dynamic weight is mainly set according to the information extracted from different images. By using the multi-scale Gaussian filter and skillfully combining different weights, the brightness information of the image can be more comprehensively extracted from multiple dimensions, the image features of different frequency components can be effectively adapted, and the accuracy of the brightness component estimation result can be significantly improved.

[0149] For images containing rich low-frequency information, since low-frequency information mainly reflects the overall contour and approximate brightness distribution of the image, a large-scale Gaussian filter can better capture the relevant features, so the weight of the large-scale Gaussian filtering result can be appropriately increased. For multi-scale images rich in details, a small-scale Gaussian filter is more sensitive to detailed information, so the weight of the small-scale Gaussian filtering result can be appropriately increased. Finally, the multi-scale images are decomposed based on the MSRCR (English full name: Multi-Scale Retinex with Color Restore) theory, so that the brightness component and the reflected light component of the multi-scale images can be accurately obtained.

[0150] For example: for images containing more low-frequency information, the weight of the large-scale Gaussian filtering result can be appropriately increased; for multi-scale images rich in details, the weight of the small-scale Gaussian filtering result can be appropriately increased. The brightness component prediction function is used to decompose different images to obtain the brightness component prediction result of different images.

[0151] In an optional embodiment, a reflected light component prediction function is set in the image component processing model.

[0152] First, based on the above multi-scale image set, the local noise level estimation value of different images is obtained;

[0153] In the process of calculating the reflected light component based on the obtained image brightness component, in order to reduce the interference of noise on the calculation result to the greatest extent, a noise suppression factor is introduced. When , the factor will have a certain degree of suppression effect on the brightness component, thereby effectively reducing the transmission of noise in the calculation process of the reflected light component.

[0154] Based on the formula , it can be known that after another brightness component prediction function accurately estimates the brightness component, the reflected light component can be correspondingly obtained. In this embodiment, the noise suppression factor is introduced and the brightness component is moderately suppressed, which can effectively reduce the transmission of noise when calculating the reflected light component and improve the quality of the reflected light component.

[0155] However, in the above method, the noise suppression factor is a fixed value. Since the noise levels of different image regions in the multi-scale image set are significantly different, it is difficult for the noise suppression factor with a fixed value to achieve ideal noise suppression effect in all regions. Therefore, it is necessary to dynamically adjust the noise suppression factor according to the local region noise level of different images in the multi-scale image set.

[0156] For different images at points The local noise level estimate at , which can be obtained by calculating the local standard deviation or using the noise estimation algorithm, defines the noise suppression factor adjustment coefficient of the noise level to the noise suppression factor as , the noise suppression factor adjustment coefficient must satisfy the following That is, when the noise level is higher, the noise suppression factor is smaller, thereby more effectively suppressing the influence of noise on the calculation of the reflected light component.

[0157] Next, a reflected light component prediction function in the image component processing model is constructed based on the local noise level estimation value and the brightness component and reflected light component decomposition functions.

[0158] Based on the above content, we can know that For different images at points There is a specific exponential correlation between the local noise level estimation value of different images and the influence function of the noise level on the noise suppression factor. The specific form of the influence function is as follows:

[0159]

[0160] in, Indicates at point Based on Determine the noise suppression factor adjustment coefficient, represents the noise sensitivity adjustment parameter, Indicates different images at points An estimate of the local noise level at .

[0161] The noise suppression factor adjustment coefficient reflects the influence of the local noise level on the noise suppression factor value; the noise sensitivity adjustment parameter can be used to control the sensitivity of the noise level to the noise suppression factor adjustment.

[0162] The noise sensitivity adjustment parameter is calculated as follows:

[0163]

[0164] in, represents the noise sensitivity adjustment parameter, represents the scaling factor, represents the global noise mean;

[0165] The specific values ​​of the above scaling factors need to be determined based on experimental data to ensure that reasonable noise suppression effects can be achieved in different noise environments. The global noise mean reflects the average noise level of the entire image area and provides a global reference for the calculation of noise sensitivity adjustment parameters.

[0166] Based on the local noise level estimation value and the luminance component and the reflected light component decomposition function, a reflected light component prediction function in the image component processing model is obtained.

[0167] In the embodiment, based on the local noise level estimation value, in combination with the decomposition function of the luminance component and the reflected light component, the reflected light component prediction function in the image component processing model is constructed. This function fully considers the influence of the local noise characteristics of the image on the calculation of the reflected light component, and can more accurately predict the reflected light component.

[0168] The expression of the above reflected light component prediction function is as follows:

[0169]

[0170] wherein, represents the prediction result of the reflected light component of different images, represents the pixel component at the coordinate position of different images, represents the adjustment coefficient of the noise suppression factor based on the noise level estimation value at the point , represents the prediction result of the luminance component of different images.

[0171] By dynamically adjusting the noise suppression factor according to the local noise level of the image, the reflected light component prediction function can effectively suppress the noise transmission in different noise level regions. In the region with higher noise level, the adjustment amplitude of the noise suppression factor can be appropriately increased to reduce the influence of the noise in the luminance component on the calculation of the reflected light component; in the region with lower noise level, a relatively small adjustment amplitude can be maintained to avoid excessive suppression of the luminance component and loss of the reflected light component information. Thus, the accuracy and pertinence of the prediction result of the reflected light component are further improved, and technical support is provided for the subsequent processing and analysis of different images.

[0172] The luminance component prediction function is used to obtain the luminance component of different images in the multi-scale image set, the reflected light component prediction function is used to obtain the reflected light component of different images in the multi-scale image set, and the construction of the above image component processing model provides a theoretical basis and framework for the setting of the luminance component prediction function and the reflected light component prediction function. In the embodiment, the RETINEX theoretical model is introduced and logarithmic transformation is performed to obtain the luminance component and the reflected light component decomposition function, and based on the theoretical basis, the luminance component prediction function is accurately set in combination with factors such as the multi-scale Gaussian filter, the local feature metric value and the weight adjustment factor. At the same time, the reflected light component prediction function is set according to the local noise level estimation value and the noise suppression factor adjustment coefficient.

[0173] ​The image component processing and optimization system also includes a component compensation normalization processing mechanism.

[0174] In the complete system architecture of image component processing and optimization, a component compensation normalization processing mechanism is introduced. This mechanism aims to perform compensation normalization operations on the brightness component prediction results and the reflected light component prediction results of each pixel in the image.

[0175] In order to ensure that the values ​​of the brightness component prediction results and the reflected light component prediction results are within an appropriate range to meet the accuracy requirements of subsequent processing and analysis of different images, a component compensation normalization processing mechanism is constructed in this embodiment. The specific settings are as follows: and Represents the brightness component The minimum and maximum values ​​in the image; at the same time, let and Represents the reflected light components The minimum and maximum values ​​in the image.

[0176] Based on the above settings, the normalized brightness component and reflected light component can be expressed as:

[0177]

[0178]

[0179] in, Represents the brightness component prediction result after normalization, Represents the normalized reflection light component prediction result.

[0180] The above normalization process can ensure that the value range of the brightness component and the reflected light component is within This not only facilitates subsequent image processing and analysis operations, enabling various algorithms to run efficiently under a unified numerical scale, but also effectively avoids calculation errors caused by numerical ranges that are too large or too small, thereby significantly improving the overall quality and reliability of image processing.

[0181] Furthermore, a brightness component optimization function and a reflected light component optimization function are set in the component optimization model.

[0182] Firstly, the saturation components of different images are obtained based on the adaptive nonlinear stretching function and multi-scale image sets.

[0183] Combining an adaptive nonlinear stretching function with a multi-scale image collection allows for the saturation components of different images to be obtained. In complex lighting environments, images are often affected by noise, interference, and distortion, leading to a series of issues such as blurred image details and reduced contrast. To effectively address these issues and practical conditions, this embodiment introduces a component optimization model that helps optimize key factors such as image brightness, contrast, and clarity, further enhancing the visualization of different images and supporting clearer and more accurate analysis and understanding of different image content.

[0184] The color modes of different images in the multi-scale image collection are converted from the RGB color mode to the HSV color mode to obtain the image saturation component. Subsequently, the saturation component of the HSV color mode image is stretched using an adaptive nonlinear stretching function to increase the color information of the different images.

[0185] The initial image saturation component calculation formula is:

[0186]

[0187] in, represents the saturation component of different images in the multi-scale image set, represents the stretch coefficient, Represents the red channel component of different images, Represents the green channel component of different images, Represents the blue channel component of different images, Indicates the maximum value of the three color components of different images, Indicates the minimum value of the three color components of different images, Represents the actual values ​​of the three color components of different images.

[0188] However, in the above formula Using the actual values ​​of the three color components is not conducive to accurate calculation. In order to ensure the credibility of different image analysis results, the embodiment will Adjusted to the average value of the three color components, that is At the same time, in order to make the calculation of the saturation component more stable and avoid the situation where the denominator is close to zero, a positive number is introduced to adjust the denominator.

[0189] Therefore, in this embodiment, based on the adaptive nonlinear stretching function and the multi-scale image set, the saturation components of different multi-scale images can be obtained, and the calculation formula is as follows:

[0190]

[0191] in, a saturation component representing different images in the multi-scale image set, a stretching coefficient, a red channel component representing different images, a green channel component representing different images, a blue channel component representing different images, a maximum value of three color components of different images, a minimum value of three color components of different images, a total value of three color components of different images, a positive number.

[0192] The above-mentioned optimized calculation formula can quickly and accurately obtain the saturation component of different images in the multi-scale image set.

[0193] Then, based on the saturation component and the multi-scale image set, the image illumination component and the component adjustment factor of different images are obtained.

[0194] Based on the image saturation component and the normalized luminance component, the image illumination component is analyzed. In this process, gradient domain guided filtering technology is introduced to improve the accuracy and rationality of the component calculation result. At the same time, in order to adapt to the complex characteristics of the multi-scale image set, a component adjustment factor is introduced, which can dynamically adjust

[0195] Let represent the importance measure of the i-th scale illumination information, and the above-mentioned component adjustment factor needs to satisfy the following relationship:

[0196]

[0197] wherein, represents the component adjustment factor, represents the importance measure of the i-th scale illumination information, represents the number of scales, represents the importance measure of the j-th scale illumination information. Through the formula, the proportion of the importance of the i-th scale illumination information in the whole can be obtained, and then the appropriate component adjustment factor is determined.

[0198] Further, the image illumination component of different images in the multi-scale image set can be obtained, and the calculation formula is as follows:

[0199]

[0200] wherein, represents the image illumination component of different images, represents the saturation component of different images in the multi-scale image set, Indicates the number of scales, represents the component adjustment factor, represents the gradient domain guided filtering, Represents the brightness component prediction result after normalization, represents the scale factor, Represents a constant.

[0201] The above image illumination component calculation formula can comprehensively consider multi-scale information and the interaction of each component, thereby accurately calculating the image illumination component.

[0202] Furthermore, a component optimization model in the image component processing and optimization system is established based on the above-mentioned image illumination components and component adjustment factors. In the embodiment, the component optimization model mainly includes a brightness component optimization function and a reflected light component optimization function.

[0203] In the process of correcting the brightness component and reflected light component of different images in the multi-scale image set based on the image illumination component, in order to enhance the flexibility and adaptability of the reflected light component correction, an adjustment factor is introduced. , this adjustment factor helps to accurately control the image illumination component Prediction results of the normalized reflected light component The degree of influence of the reflected light component is determined, and then an optimization model of the reflected light component is constructed.

[0204] In the reflected light component optimization model, the calculation formula of the corrected reflected light component is as follows:

[0205]

[0206] in, represents the reflected light component of different images after correction, represents the adjustment factor, represents the image illumination component for different images, Represents the normalized reflection light component prediction result.

[0207] At the same time, in order to make the correction process of the brightness component more consistent with the local features of the multi-scale image, the embodiment uses the brightness mean of the local area of ​​the multi-scale image The correction parameter ρ of the brightness component is dynamically adjusted, and a brightness component optimization model is constructed.

[0208] The brightness component optimization function in the component optimization model satisfies the following relationship:

[0209] Among them, represents the brightness component of different images after correction, represents the correction parameter of the luminance component, The corrected reflection light component of the different images is obtained by optimizing the reflection light component of the different images using the reflection light component optimization function.

[0210] In an optional embodiment, the reflection light component of the different images is optimized and corrected using a reflection light component optimization function to obtain a corrected reflection light component; and the luminance component of the different images is optimized and corrected based on the luminance component optimization function and the corrected reflection light component to obtain a corrected luminance component.

[0211] The corrected reflection light component of the different images is obtained by optimizing the reflection light component of the different images using the reflection light component optimization function.

[0212] The corrected reflection light component satisfies the following relationship: ;

[0213] Subsequently, the luminance component of the different images is corrected based on the luminance component optimization function and the corrected reflection light component, and finally a corrected luminance component is obtained, which satisfies the following relationship: .

[0214] Based on the above optimization and correction process, effective compensation of the light of the different images can be achieved. The method can suppress low-frequency information in the different images and enhance high-frequency information, thereby completing the optimization processing of the different images in the multi-scale image set and effectively solving key problems such as excessive noise in dark parts and loss of details in bright parts of images under complex light, so as to improve the image quality of the multi-scale image set.

[0215] The component optimization model of the embodiment is constructed based on the image saturation component, the light component, and the component adjustment factor, which provides a basis for establishing the luminance component and reflection light component optimization functions. The saturation component is obtained by introducing an adaptive nonlinear stretching function, and the light component is calculated by combining the gradient domain guided filtering technology and the component adjustment factor. The component optimization model can optimize the luminance component and the reflection light component, and solve key problems such as excessive noise in dark parts and loss of details in bright parts of images under complex light.

[0216] Furthermore, in the embodiment, the optimization and correction method of the luminance component and the reflection light component of the different images is only an optional condition of the embodiment. In other one or some embodiments, the image component optimization and correction method can be optimized according to the analysis target of the multi-scale image set and the actual situation of the different images, so that the optimization and correction method can better adapt to the characteristics of the different images, thereby improving the effectiveness of the image processing result.

[0217] S4, an image enhancement processing model is established based on the corrected luminance component and the corrected reflection light component, and different images in the multi-scale image set are enhanced by the image enhancement processing model to obtain a multi-scale image target set, and the specific implementation steps and contents are as follows:

[0218] The corrected brightness component and the corrected reflected light component obtained in S3 are the basic data for image enhancement processing in S4. After obtaining more accurate brightness component and reflected light component, the high- and low-frequency filtering information of different images is further analyzed to construct an image enhancement processing model, which is conducive to the enhancement processing of multi-scale image sets.

[0219] First, in this embodiment, high- and low-frequency filtering information of different images is analyzed based on the corrected brightness component and the corrected reflected light component.

[0220] Among them will Defined as the low-frequency filtering information of the multi-scale image, Defined as the high-frequency filtered information of a multi-scale image. For low-frequency filtering, Gaussian low-pass filtering is used, which can smooth the image and retain the low-frequency components of different images. For high-frequency filtering, Laplace high-pass filtering is used, which can effectively extract high-frequency details of different images.

[0221] The low-frequency filtering information of different images in the multi-scale image set satisfies the following formula:

[0222]

[0223] in, Represents the low-frequency filtering information of different images in the multi-scale image set, represents the brightness component of different images after correction, represents the reflected light component of different images after correction, represents the element-wise product symbol, represents the convolution operation, The standard deviation is Gaussian low-pass filter;

[0224] The element-by-element product symbol is used to multiply the corresponding elements of two matrices; the convolution operation is used to convolve the image with the filter; the standard deviation is The Gaussian low-pass filter is adjusted by adjusting the standard deviation You can control the smoothness of the filtering.

[0225] The high-frequency filtering information of different images in the multi-scale image set satisfies the following formula:

[0226]

[0227] in, Represents the high-frequency filtering information of different images in a multi-scale image set, represents the brightness component of different images after correction, represents the reflected light component of the different images after correction, represents the element-wise multiplication symbol, represents the convolution operation, represents the Laplacian operator;

[0228] The Laplacian operator is used to detect edges and detail information in images, where the discrete kernel of the Laplacian operator Δ needs to satisfy the following relationship:

[0229]

[0230] This discrete kernel performs a second-order differential operation on the image, which can highlight the high-frequency variation part in different images, thereby achieving effective extraction of high-frequency information. High-low frequency filter information analysis is the basis for constructing an image enhancement processing model. Obtaining low-frequency filter information and high-frequency filter information of the image can help understand the structural information and detail information of the image. Based on the relevant information, gain coefficients and balance adjustment factors are introduced to construct an image enhancement processing model, which is conducive to achieving effective enhancement of different frequency bands of the image.

[0231] Then, an image enhancement processing model is established based on the gain coefficients and high-low frequency filter information.

[0232] Based on the air volume information of the image, high-low frequency filter information analysis is performed, and then gain coefficients and the above-mentioned high-low frequency filter information of the image are introduced to construct an image enhancement processing model.

[0233] The low-frequency gain coefficient and the high-frequency gain coefficient are introduced, and satisfy The above gain coefficients can flexibly adjust the enhancement intensity of different frequency bands to adapt to the characteristics and processing requirements of different images. At the same time, a balance adjustment factor is introduced, where This adjustment factor is conducive to achieving dynamic balance between low-frequency and high-frequency information, effectively avoiding problems such as over-smoothing (when the weight is excessively biased towards high frequency) or over-enhancement (when the weight is excessively biased towards low frequency) caused by fixed weights. Wherein The larger the value is, the higher the proportion of low-frequency information in the final enhancement result is.

[0234] Based on the high-frequency filter information , the low-frequency filter information of different images, the corrected luminance component and the reflected light component, the homomorphic filtering algorithm is improved to further realize image enhancement, and the expression of the above-mentioned image enhancement processing model is as follows:

[0235]

[0236] Wherein, represents an image in the frequency domain after enhancement processing, represents a low-frequency gain coefficient, represents low-frequency filter information, represents a balance adjustment factor, represents a high-frequency gain coefficient, represents high-frequency filter information, represents the Euclidean distance from a point on the frequency domain to a center point, represents a constant to avoid division by zero.

[0237] The above Gaussian low-pass filtering result reflects the smooth structure information of the image; the Euclidean distance can suppress high-frequency noise, and the farther the distance from the center point, the smaller the corresponding weight; the constant is used to avoid the occurrence of division by zero.

[0238] In order to ensure the quality of images obtained by different imaging devices, the above image processing method based on imaging multi-scale analysis further includes: constructing an image loss detection function to detect the loss of different images.

[0239] In order to ensure the quality stability of images obtained by different imaging devices, an image loss detection mechanism is introduced in the image processing method based on imaging multi-scale analysis in the embodiment, and specific detection of different image loss conditions is realized by constructing an image loss detection function.

[0240] First, according to the device reference image and the multi-scale image set, the characteristics of different images in three key dimensions of spatial distance, structural similarity and gradient difference are analyzed, wherein the spatial distance can reflect the absolute difference of image pixels; the structural similarity can measure the similarity of the structure of the image; and the gradient difference can reflect the change of the edge and detail information of the image.

[0241] Then, based on the spatial distance, image structural similarity and image gradient difference obtained by the above analysis, an image loss detection function is constructed, in order to effectively reduce the gap between the multi-scale generated image and the normal illumination image. In this embodiment, the spatial distance, gradient and structure are combined to construct a fusion loss function, and the mathematical expression is as follows:

[0242]

[0243] wherein, represents the contrast loss result of the device reference image and the device generated image, represents the weight coefficient corresponding to the image spatial distance, represents the device reference image, represents the device generated image, represents the weight coefficient corresponding to the image structural similarity, represents similarity between structure similarity between weight coefficient corresponding to image gradient difference, represent horizontal x and vertical y directions, represent image gradient in direction gradient in direction represent image gradient in direction gradient in direction represent Euclidean norm.

[0244] wherein represents the pixel-level L2 norm square between P and T, i.e. spatial distance loss, which can effectively measure the absolute difference between two images (device reference image and device generated image);

[0245] structure similarity loss converts the structural similarity index into an optimization target, so that the loss function can better reflect the similarity of the structure of the image (device reference image and device generated image);

[0246] gradient difference loss i.e. L2 loss of gradient image in direction i, which can effectively capture the changes in edge and detail information of different images.

[0247] The above image loss detection function constrains the generated image from three aspects of space, structure and gradient, which meets the optimization target of image reconstruction task and helps to improve the quality of the generated image. After the multi-scale image set is processed by the image enhancement processing model, the image loss detection function is used to evaluate the loss degree of the enhanced image. The above image loss detection function evaluates the image from three key dimensions of spatial distance, structure similarity and gradient difference, which can quantify the difference between the enhanced image and the device reference image. Through image loss evaluation, the image with better quality can be selected to form a multi-scale image target set to ensure that the image meets the actual application requirements.

[0248] In an optional embodiment, a multi-scale image target set is obtained, and the relevant implementation steps are as follows:

[0249] First, different images in the multi-scale image set are processed by the image enhancement processing model to obtain an enhanced multi-scale image set, which helps to improve the quality of the multi-scale image set.

[0250] The mathematical expression of the image enhancement processing model is as follows:

[0251]

[0252] Then, the image loss detection function is used to evaluate the loss degree of different images in the enhanced multi-scale image set, so as to obtain a multi-scale image target set.

[0253] After obtaining the enhanced multi-scale image set, the image loss detection function is used to evaluate the loss degree of different images in the set, which aims to quantify the difference between the enhanced image and the device reference image (other reference image), so as to filter out images with better quality, which helps to obtain a multi-scale image target set.

[0254] The expression of the above image loss detection function is as follows:

[0255]

[0256] Based on this, the image enhancement processing model of the embodiment processes different images in the multi-scale image set, which helps to adjust the brightness, contrast, color and other parameters of the image through specific mathematical operations, so that the image is clearer and brighter, and the visual appeal of the image is enhanced, so that the enhanced image can clearly display image details and feature information. At the same time, the model can effectively suppress noise in the image, remove interference information in the image, and improve the signal-to-noise ratio of the image. In digital image processing, due to the complexity of the imaging environment, there are often a large amount of noise in the image. After enhancement processing, the noise can be removed, so that the feature information in the image is more prominent, which is convenient for subsequent image analysis and recognition comparison.

[0257] The image loss detection function of the embodiment can quickly and accurately evaluate the quality of the enhanced image, and filter out images with better quality to form a multi-scale image target set. Based on this, the subjectivity and inefficiency of manual screening are avoided, and the image processing efficiency is improved. At the same time, by setting a suitable loss threshold, it can be ensured that the quality of the filtered image meets the requirements, and the feasibility of the image processing result is improved.

[0258] On the other hand, the parameters of the image enhancement processing model and the image loss detection function can be adjusted according to different application scenarios and image monitoring requirements to obtain the best processing effect. For example, in the field of security monitoring, for scenes with high real-time requirements, the calculation process of the model and function can be appropriately simplified to improve the processing speed; in the field of medical image analysis, for scenes with high accuracy requirements, the complexity of the model and function can be increased to improve the image quality and analysis accuracy.

[0259] Sequentially performing the related steps of S1, S2, S3, and S4 can realize the collection, geometric correction, component processing and optimization, and enhancement processing of multi-scale images, and finally obtain a high-quality multi-scale image target set. The related steps jointly complete the image processing task, so that the related images achieve the expected effect of the scene.

[0260] See Figure 4 In an alternative embodiment, in order to efficiently execute the image processing method based on imaging multi-scale analysis provided by the present application, the present application further provides an image processing system based on imaging multi-scale analysis, wherein the input device, the processor, the output device and the memory in the system are connected with each other, the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to invoke the program instructions to execute the specific steps of the embodiments of the image processing method based on imaging multi-scale analysis provided by the present application. The image processing system based on imaging multi-scale analysis of the present application has a complete structure and objective stability, can efficiently execute the image processing method based on imaging multi-scale analysis of the present application, and improves the overall applicability and practical application ability of the present application.

[0261] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the description of the present application.

Claims

1. An image processing method based on multi-scale imaging analysis, characterized in that: The steps include: Acquiring multi-scale image historical storage information, performing non-uniformity detection on different imaging devices based on the multi-scale image historical storage information and obtaining non-uniformity detection results of the different imaging devices, and adjusting the different imaging devices according to the non-uniformity detection results to obtain imaging devices with adjusted uniformity; Using the imaging device after uniformity adjustment to collect original multi-scale images, setting an image geometric correction model to perform geometric correction on the original multi-scale images to obtain a multi-scale image set; Constructing an image component processing and optimization system, analyzing the multi-scale image set using an image component processing model and a component optimization model in the image component processing and optimization system, and obtaining corrected brightness components and corrected reflected light components of different images in the multi-scale image set; Establishing an image enhancement processing model based on the corrected brightness component and the corrected reflected light component, and performing enhancement processing on different images in the multi-scale image set using the image enhancement processing model to obtain a multi-scale image target set; The performing non-uniformity detection on different imaging devices based on the multi-scale image history storage information and obtaining non-uniformity detection results of different imaging devices includes: Obtaining environmental conditions and image acquisition requirements based on the multi-scale image historical storage information; Setting an imaging device reference image based on the environmental conditions and the image acquisition requirements; Analyzing difference image means of different images in the multi-scale image history storage information according to the imaging device reference image and the multi-scale image history storage information; Analyzing the standard deviations of different images in the multi-scale image history storage information based on the difference image mean; Based on the standard deviation and the multi-scale image history storage information, non-uniformity detection is performed on different imaging devices, and non-uniformity detection results of the different imaging devices are obtained.

2. The image processing method based on imaging multi-scale analysis according to claim 1, characterized in that: The method of acquiring original multi-scale images by using the imaging device after uniformity adjustment and performing geometric correction on the original multi-scale images by setting an image geometric correction model to obtain a multi-scale image set includes: Obtaining image coordinate information before transformation based on the original multi-scale image; Selecting image control points, and establishing an image geometric correction model based on the image control points and the image coordinate information before transformation; Performing coordinate transformation on the image coordinate information before transformation by using the image control points and the image geometric correction model to obtain image coordinate information after geometric correction; A multi-scale image set is obtained based on the geometrically corrected image coordinate information.

3. The image processing method based on imaging multi-scale analysis according to claim 1, characterized in that: The image component processing and optimization system is constructed, and the multi-scale image set is analyzed by the image component processing model and the component optimization model in the image component processing and optimization system to obtain the corrected brightness components and the corrected reflected light components of different images in the multi-scale image set. Setting a brightness component prediction function and a reflected light component prediction function in the image component processing model; Obtaining the brightness components of different images in the multi-scale image set by using the brightness component prediction function, and obtaining the reflected light components of different images in the multi-scale image set by using the reflected light component prediction function; Setting a brightness component optimization function and a reflected light component optimization function in the component optimization model; Correcting and optimizing the reflected light components of different images using the reflected light component optimization function to obtain corrected reflected light components; The brightness components of different images are corrected and optimized based on the brightness component optimization function and the corrected reflected light component to obtain the corrected brightness component.

4. The image processing method based on imaging multi-scale analysis according to claim 3, characterized in that: The setting of the brightness component prediction function and the reflected light component prediction function in the image component processing model comprises: A Retinex theoretical model is introduced and logarithmic transformation is performed on the Retinex theoretical model to obtain decomposition functions of brightness component and reflected light component; Setting Gaussian filters of different scales, and obtaining local feature measurement values ​​and weight adjustment factors of different images according to the multi-scale image set; Decomposition functions of the brightness component and the reflected light component are derived according to the local feature metric value and the weight adjustment factor to obtain a brightness component prediction function in the image component processing model.

5. The image processing method based on imaging multi-scale analysis according to claim 3, characterized in that: The setting of the brightness component prediction function and the reflected light component prediction function in the image component processing model comprises: Obtaining local noise level estimates for different images based on the multi-scale image set; A reflected light component prediction function in an image component processing model is constructed based on the local noise level estimation value and the brightness component and reflected light component decomposition function.

6. The image processing method based on imaging multi-scale analysis according to claim 3, characterized in that: The setting of the brightness component optimization function and the reflected light component optimization function in the component optimization model includes: Obtaining saturation components of different images according to the adaptive nonlinear stretching function and the multi-scale image set; Obtaining image illumination components and component adjustment factors of different images based on the saturation component and the multi-scale image set; A component optimization model in an image component processing and optimization system is established according to the image illumination component and the component adjustment factor. The component optimization model includes a brightness component optimization function and a reflected light component optimization function.

7. The image processing method based on imaging multi-scale analysis according to claim 1, characterized in that: The establishing of an image enhancement processing model based on the corrected brightness component and the corrected reflected light component, and performing enhancement processing on different images in the multi-scale image set by using the image enhancement processing model to obtain a multi-scale image target set includes: Analyzing high- and low-frequency filtering information of different images according to the corrected brightness component and the corrected reflected light component; An image enhancement processing model is established according to the gain coefficient and the high and low frequency filtering information.

8. The image processing method based on imaging multi-scale analysis according to claim 7, characterized in that: The method further comprises: analyzing spatial distances, structural similarities, and gradient differences of different images based on the imaging device reference image and the multi-scale image set; Establishing an image loss detection function based on the spatial distance, the structural similarity and the gradient difference; Processing different images in the multi-scale image set by the image enhancement processing model to obtain an enhanced multi-scale image set; The image loss detection function is used to evaluate the loss of different images in the enhanced multi-scale image set to obtain a multi-scale image target set.

9. An image processing system based on multi-scale imaging analysis, characterized in that: The system includes a processor, an input device, an output device, and a memory, which are interconnected. The memory is used to store a computer program, and the computer program includes program instructions. The processor is configured to call the program instructions to execute the image processing method based on multi-scale imaging analysis according to any one of claims 1 to 8.

Images