A method and system for calculating image complexity considering multi-scale fusion

Abstract

The present invention belongs to the field of image processing technology, and discloses a method and system for calculating image complexity considering multi-scale fusion. The method comprises: judging whether the pixel values ​​of width and height of the input image are the same, determining the width and number of iterations of two hyperparameter filters, and calculating the value range of the number of iterations; performing coarse-graining processing on the image to obtain a renormalized image obtained after the th coarse-graining; expanding the renormalized image obtained after the th coarse-graining to obtain an expanded image of the renormalized image obtained after the th coarse-graining. The present invention expands the application scope of the technology, making it more suitable for ordinary users to analyze the complexity of conventional images.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to an image complexity calculation method and system considering multi-scale fusion. Background Art

[0002] The measurement of image complexity has important applications in many fields, including computer vision, pattern recognition, image compression, biomedical image analysis, etc. The quantification of complexity can not only help improve the accuracy of image processing and analysis, but also provide important information about the structure and content of the image. The definition of complexity is usually related to the structural characteristics and content density of the image, reflecting the amount of information contained in the image and how it is organized.

[0003] There are many existing methods for calculating image complexity, ranging from simple edge detection to feature extraction using deep learning. Through these methods, the complexity of images can be quantitatively analyzed, and processing strategies and information processing efficiency can be optimized according to specific needs. Image complexity calculation has a very critical practical application value in improving image processing technology, enhancing user experience, and optimizing resource utilization.

[0004] Among them, the existing technology Bagrov AA, Iakovlev IA, Iliasov AA, et al. Multiscalestructural complexity of natural patterns[J]. Proceedings of the National Academy of Sciences, 2020, 117(48): 30241-30251 provides a complexity calculation method through coarse-grained processing. This method first downsamples the original image step by step to generate a series of images with gradually decreasing resolution. The width and height of these images are adjusted to be consistent, and then the complexity of the image is calculated using the initial and final generated images. By analyzing the changes between images of different resolutions, this method attempts to capture the overall complexity characteristics of the image. Although this method is simple and easy to implement, it has certain limitations.

[0005] Through the above analysis, the problems and defects of the prior art are as follows:

[0006] (1) This method requires that the image has the same width and height before processing, which is not suitable for ordinary users to analyze the complexity of conventional images;

[0007] (2) This method only uses the initial and final renormalized images to calculate the complexity, ignoring the image information at all scales in the coarse-grained process, which leads to an incomplete description of image complexity;

[0008] (3) The final complexity calculation result is usually between 0 and 1. This small numerical range makes it difficult to distinguish the complexity differences between different images, limiting its effectiveness in practical applications.

[0009] (4) This method has low computational efficiency when processing high-resolution images or large-scale data sets, which becomes a bottleneck in practical applications.

[0010] These deficiencies indicate that existing coarse-grained methods have room for improvement in terms of accuracy, comprehensiveness, and computational efficiency. Summary of the invention

[0011] In order to overcome the problems existing in the related art, the disclosed embodiments of the present invention provide a method and system for calculating image complexity considering multi-scale fusion.

[0012] The technical solution is as follows: an image complexity calculation method considering multi-scale fusion, including:

[0013] S1, determine whether the pixel values ​​of the width and height of the input image are the same. If they are not the same, fill the image background. If the larger value is an odd number, the image size is increased by 1 to become an even number;

[0014] S2, determines the width of two hyperparameter filters With the number of iterations , calculate the number of iterations The value range of

[0015] S3, performs coarse-grained processing on the image, and obtains the Renormalized image after sub-coarse granularity ;

[0016] S4, for The renormalized image obtained after the second coarse granularity Expand and get the extended image ;

[0017] S5, calculate the The renormalized image obtained after the second coarse granularity Extended image of With The renormalized image obtained by sub-coarse granularization The overlap between ;

[0018] S6, Calculate image complexity .

[0019] In step S1, after the image background is filled, the image data is normalized. The background filling method is divided into edge filling and positioning filling. Positioning filling is to place the original image on the left side of the background filling area for comparative analysis. The normalization calculation formula is:

[0020] ;

[0021] ;

[0022] ;

[0023] In the formula, They are the first Row, No. The normalized values ​​of the RGB channels of the column pixels, They are the first Row, No. RGB channel values ​​of the column pixels.

[0024] In step S2, the filter width , filter width With the number of iterations are two hyperparameters in the coarse-graining process. The width and height of the renormalized image obtained by each coarse-graining are reduced. times, and the minimum width and height is 1, the number of iterations The calculation formula for the value range of is:

[0025] ;

[0026] In the formula, For the The side length of the image obtained at the second coarse granularity is an integer, is the side length of the original image.

[0027] In step S3, the image is subjected to coarse-grained processing, including:

[0028] In the coarse-graining process, the image is divided into multiple sizes The pixel block is formed by adding the values ​​of each pixel in the pixel block and taking the average value to obtain a pixel point, which is used to replace the original pixel block; each pixel block is traversed to complete a coarse-graining. The renormalized image obtained after the second coarse granularity , the calculation formula is:

[0029] ;

[0030] In the formula, For the The renormalized image obtained by the second coarse-graining is Row, No. Column pixels; is the image before coarse-graining Row, No. The first pixel block Row, No. Column pixels; For the The renormalized image obtained by the second coarse-graining is Row, No. Column pixels.

[0031] In step S4, The renormalized image obtained after the second coarse granularity The nearest neighbor upsampling expansion is used to obtain the The renormalized image obtained by sub-coarse granularization Extended image of , the calculation formula for the nearest neighbor upsampling expansion is:

[0032] ;

[0033] In the formula, for No. Row, No. Column pixels, for No. Row, No. Column pixels, for No. Row, No. Column pixels, for No. Row, No. Column pixels.

[0034] Furthermore, renormalize the image The side length is The renormalized image obtained after the second coarse granularity of times, expand the image and The size is consistent; and meets ;in, For the The side length of the image obtained at the next coarser granularity is an integer, is the side length of the expanded image, for The extended image of .

[0035] In step S5, calculate the The renormalized image obtained after the second coarse granularity Extended image of With The renormalized image obtained by sub-coarse granularization The overlap between ,include:

[0036] for No. Row, No. Column pixel points, a vector, and The product of is the vector product; the calculation formula is:

[0037] ;

[0038] In the formula, For the The renormalized image obtained by -1 times of coarse granularity is Row, No. Column pixels.

[0039] In step S6, the image complexity The calculation formula is:

[0040] ;

[0041] ;

[0042] In the formula, is the component of complexity C, For the The renormalized image obtained by sub-coarse granularization Extended image of With The renormalized image obtained by sub-coarse granularization The overlap between .

[0043] Furthermore, the image complexity By exponentially magnifying , the expression is:

[0044] ;

[0045] In the formula, is the complexity of the enlarged image, C is the complexity of the original image, is the base of natural logarithms.

[0046] Another object of the present invention is to provide an image complexity calculation system considering multi-scale fusion, the system implements the image complexity calculation method considering multi-scale fusion, the system comprising:

[0047] The image background filling module is used to determine whether the pixel values ​​of the width and height of the input image are the same. If they are not the same, the image background is filled. If the larger value is an odd number, the image size is increased by 1 to become an even number.

[0048] Iteration number range determination module, used to determine the width of two hyperparameter filters With the number of iterations , calculate the number of iterations The value range of

[0049] Renormalized image acquisition module, used to determine the two hyperparameter filter widths With the number of iterations After that, the image is processed into coarse grains, and the Renormalized image after sub-coarse granularity ;

[0050] Expand the renormalized image acquisition module to The renormalized image obtained after the second coarse granularity Expand and get the extended image ;

[0051] The overlap calculation module calculates the The renormalized image obtained after the second coarse granularity Extended image of With The renormalized image obtained by sub-coarse granularization The overlap between ;

[0052] Image complexity calculation module, used to calculate image complexity , amplified by an exponential function

[0053] In combination with all the above technical solutions, the beneficial effects of the present invention are as follows:

[0054] Regarding the image size limitation: the current technology requires that the processed image must have the same width and height, which limits the applicability of the method in common applications. The present invention proposes a complexity evaluation method that can process images with any aspect ratio, thereby expanding the application scope of the technology and making it more suitable for ordinary users to analyze the complexity of conventional images.

[0055] Regarding the comprehensiveness of complexity evaluation: existing methods only consider the initial and final states of the image, ignoring the information of each intermediate scale in the coarse-graining process. This invention proposes a new computational framework that incorporates images of all or multiple intermediate coarse-grained scales into complexity calculation to provide a more comprehensive and detailed description of image complexity, thereby more accurately reflecting the true complexity of the image.

[0056] Regarding the problem of result differentiation: the complexity values ​​generated by traditional methods are between 0 and 1, which makes it difficult to identify the complexity differences between different images. The present invention aims to develop a new metric or adjust the existing algorithm to increase the range of the output complexity value, so that it is easier to distinguish the complexity differences between different images.

[0057] On the issue of computational efficiency: Traditional methods have low computational efficiency when processing high-resolution images or large-scale image sets. The present invention proposes a more efficient algorithm implementation or adopts a more optimized computing resource management solution to improve the computational efficiency of the coarse-graining process and complexity evaluation, so as to be more suitable for the processing requirements of high-resolution images or large-scale image sets.

[0058] In view of the inconsistent width and height of the original image, a padding strategy is adopted to adjust the image so that all processed images have uniform width and height, ensuring the versatility and adaptability of the algorithm.

[0059] The present invention not only considers the complexity of the final coarse-grained image, but also integrates the complexity changes of the images in each intermediate stage into the final complexity evaluation, and evaluates the overall complexity by integrating these continuous changes. By adjusting the numerical range of the complexity output (for example, from 0-1 to a wider value such as 21, 43, 72), the results are more distinguishable and easier to analyze and interpret; by optimizing the algorithm and simplifying the calculation process, the efficiency of complexity calculation is greatly improved. Improve accuracy and scope of application: The present invention uses multi-scale fusion analysis to more accurately evaluate image complexity, and can be applied to multiple fields such as automatic classification, evaluation and search optimization of images. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings herein are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description, serve to explain the principles of the present disclosure;

[0061] Figure 1 is a schematic diagram of a method for calculating image complexity considering multi-scale fusion provided by an embodiment of the present invention;

[0062] Figure 2 is a simple example principle diagram of image expansion provided by an embodiment of the present invention;

[0063] Figure 3 is a schematic diagram of an image complexity calculation system considering multi-scale fusion provided by an embodiment of the present invention;

[0064] Figure 4 This is an experimental example 1 of edge filling provided by an embodiment of the present invention, with the filling background being black;

[0065] Figure 5 This is the second experimental example of edge filling provided by the embodiment of the present invention, with the filling background being black;

[0066] Figure 6 This is the third experimental example of edge filling provided by the embodiment of the present invention, with the filling background being black;

[0067] Figure 7 This is the fourth experimental example of edge filling provided by the embodiment of the present invention, with the filling background being black;

[0068] Figure 8 This is the experimental example 5 of edge filling provided by the embodiment of the present invention, and the filling background is black;

[0069] Fig. 9 This is the sixth experimental example of edge filling provided by the embodiment of the present invention, with the filling background being black;

[0070] In the figure: 1. Image background filling module; 2. Iteration number value range determination module; 3. Renormalized image acquisition module; 4. Extended renormalized image acquisition module; 5. Coincidence calculation module; 6. Image complexity calculation module. DETAILED DESCRIPTION

[0071] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific implementation disclosed below.

[0072] The embodiment of the present invention provides a method for calculating image complexity, which achieves a more accurate and efficient evaluation of image complexity by adjusting width and height consistency, calculating multi-scale complexity, adjusting the complexity output range, and optimizing computational efficiency. The method expands the numerical range of complexity output by comprehensively analyzing the changes in images of different sizes, and optimizes the algorithm to improve computational efficiency, which is particularly suitable for processing large-scale image data. The innovative points of the image complexity calculation method and system considering multi-scale fusion provided by the embodiment of the present invention are:

[0073] Embodiment 1: The technical improvements of the method for calculating image complexity considering multi-scale fusion provided by the embodiment of the present invention mainly include:

[0074] (1) Aspect-height consistency adjustment: Use padding or other image processing methods to adjust the original image of any aspect ratio to adapt it to the requirements of the complexity algorithm and ensure that the image width and height are consistent.

[0075] (2) Multi-scale complexity calculation: Instead of just considering the original image and the last coarse-grained image, the overall image complexity is evaluated by comprehensively analyzing the image changes at each size during the coarse-grained process.

[0076] (3) Adjustment of complexity output range: The output value range of complexity is innovatively expanded to improve the differentiation of results and make complexity evaluation more delicate and practical.

[0077] (4) Optimization of computational efficiency: By improving algorithms and simplifying the computational process, the efficiency of complex computations is improved, especially when processing large-scale image data that requires a lot of computation.

[0078] Embodiment 2, as a further detailed technical solution of the present invention, the method for calculating image complexity considering multi-scale fusion provided by the embodiment of the present invention specifically includes the following steps:

[0079] S1, determine whether the pixel values ​​of the width and height of the input image are the same. If they are not the same, fill the image background. If the larger value is an odd number, the image size is increased by 1 to become an even number;

[0080] Generally, the background color of the image filling is black, but since the present invention needs to perform normalization processing (preprocessing) on ​​the image data, the normalization processing calculation formula is:

[0081] ;

[0082] ;

[0083] ;

[0084] In the formula They are the first Row, No. The normalized values ​​of the RGB channels of the column pixels, They are the first Row, No. RGB channel values ​​of column pixels. It can be seen that when the RGB channel value of the filled background is (128, 128, 128), the normalized RGB channel value of the filled background after normalization is (0, 0, 0), so the present invention classifies and discusses the color of the channel background. In addition, the background filling method is generally divided into edge filling and positioning filling (the present invention places the original image on the left side of the background filling area for comparative analysis), so the present invention also classifies and discusses the filling method of the channel background.

[0085] S2, determines the width of two hyperparameter filters With the number of iterations , calculate the number of iterations The value range of

[0086] Specifically, determine the filter width ,generally . Iterations It is a hyperparameter in the coarse-graining process. The width and height of the renormalized image obtained by each coarse-graining will be reduced. times, and the minimum value is 1, so the invention innovatively proposes that the value range of the number of iterations N can be calculated according to the following formula:

[0087] ;

[0088] In the formula, For the The side length of the image obtained at the next coarser granularity is an integer, is the side length of the original image.

[0089] The technical function of formula (4) is to determine the value range of the number of iterations N.

[0090] S3, determine the two hyperparameters filter width With the number of iterations After that, the image is processed into coarse grains, and the Renormalized image after sub-coarse granularity ;

[0091] Perform the image The second coarse-grained processing obtains a side length of The renormalized image obtained after secondary coarse-graining ;

[0092] In the process of coarse-graining, the image is divided into multiple sizes. Then add the values ​​of each pixel in the pixel block and get the average value to get a pixel point, and use this pixel point to replace the original pixel block. Traverse each pixel block and perform this operation in turn, and finally complete a coarse-graining, and get a side length of the image before coarse-graining The renormalized image of is calculated as follows:

[0093] ;

[0094] In the formula, For the The renormalized image obtained by the second coarse-graining is Row, No. Column pixels; is the image before coarse-graining Row, No. The first pixel block Row, No. Column pixels; For the The renormalized image obtained by the second coarse-graining is Row, No. Column pixels.

[0095] No. Image before coarse-graining That is The renormalized image obtained after secondary coarse-graining , so the The renormalized image obtained after secondary coarse-graining It is The renormalized image obtained after secondary coarse-graining of ;

[0096] S4, for The renormalized image obtained after the second coarse granularity Expand and get the extended image ;

[0097] Specifically, assuming For the The renormalized image obtained after the second coarse granularity is is the original image. When calculating the complexity of the image, the present invention applies the multi-scale fusion technology, which requires calculation and The overlap between .

[0098] because The side length is of , in the calculation and When the overlap is large, the size needs to be consistent, so Expand (nearest neighbor upsampling) and get The renormalized image obtained by sub-coarse granularization Extended image of ,like Figure 1 As shown in the expanded image in , the calculation formula for the nearest neighbor upsampling expansion is:

[0099] ;

[0100] In the formula, for No. Row, No. Column pixels, for No. Row, No. Column pixels, for No. Row, No. Column pixels, for No. Row, No. Column pixels.

[0101] A simple algorithm for image expansion is Figure 2 As shown, For the The side length of the image obtained at the next coarser granularity is an integer, is the side length of the expanded image and satisfies , for The extended image of

[0102] S5, calculate the The renormalized image obtained after the second coarse granularity Extended image of With The renormalized image obtained by sub-coarse granularization The overlap between ;

[0103] Assumptions for No. Row, No. Column pixels, so it is a vector, and The product of is the vector product. The present invention innovatively proposes to calculate the The renormalized image obtained after the second coarse granularity Extended image of With The renormalized image obtained by sub-coarse granularization The overlap between , the calculation formula is as follows:

[0104] ;

[0105] In the formula, For the The renormalized image obtained by -1 times of coarse granularity is Row, No. Column pixels.

[0106] S6, Calculate image complexity , the calculation formula is as follows:

[0107] ;

[0108] ;

[0109] In the formula, For complexity The weight, For the The renormalized image obtained by sub-coarse granularization Extended image of With The renormalized image obtained by sub-coarse granularization The overlap between .

[0110] The image complexity By exponentially magnifying , the expression is:

[0111] ;

[0112] In the formula, is the complexity of the enlarged image, is the complexity of the original image, is the base of natural logarithms.

[0113] It can be seen from the above embodiments that, in case of inconsistent width and height of the original image, a filling strategy is adopted to adjust the image so that all processed images have uniform width and height, thereby ensuring the versatility and adaptability of the algorithm.

[0114] The present invention not only considers the complexity of the final coarse-grained image, but also incorporates the complexity changes of the images in various intermediate stages into the final complexity evaluation, and evaluates the overall complexity by integrating these continuous changes.

[0115] By adjusting the range of complexity output (e.g. from 0-1 to wider values ​​such as 21, 43, 72), the results are more distinguishable and easier to analyze and interpret;

[0116] The efficiency of complex calculations has been greatly improved by optimizing algorithms and simplifying calculation processes.

[0117] Embodiment 3, as Figure 3 As shown, the image complexity calculation system considering multi-scale fusion provided by the embodiment of the present invention includes:

[0118] Image background filling module 1 is used to determine whether the pixel values ​​of width and height of the input image are the same. If they are not the same, the image background is filled. If the larger value is an odd number, the image size is increased by 1 to become an even number.

[0119] Iteration number value range determination module 2 is used to determine the width of two hyperparameter filters With the number of iterations , calculate the number of iterations The value range of

[0120] Renormalized image to obtain module 3, used to determine the two hyperparameter filter widths With the number of iterations After that, the image is processed into coarse grains, and the Renormalized image after sub-coarse granularity ;

[0121] Expand the renormalized image acquisition module 4, The renormalized image obtained after the second coarse granularity Expand and get the extended image ;

[0122] Coincidence calculation module 5, calculates the The renormalized image obtained after the second coarse granularity Extended image of With The renormalized image obtained by sub-coarse granularization The overlap between ;

[0123] Image complexity calculation module 6, used to calculate image complexity , which is amplified by an exponential function.

[0124] The above technical improvements can be applied to the following areas:

[0125] Oil spill detection: Image complexity can be used to detect the severity of oil spills in seawater. By analyzing the image of the oil film in the seawater, the distribution and spread of the oil can be evaluated, so that appropriate cleanup measures can be taken.

[0126] Magnetic field detection: In magnetic field detection, image complexity can be used to analyze the strength and distribution of the magnetic field. By visualizing the magnetic field lines, the complexity of the magnetic field can be more intuitively understood, which is very important for physical experiments and engineering applications.

[0127] Medical imaging analysis: In the field of medical imaging, image complexity can help doctors analyze the complexity of the lesion area. For example, in MRI or CT scans, complexity analysis can help identify and distinguish different tissue types.

[0128] Remote sensing image analysis: In the field of remote sensing, image complexity can be used to analyze land cover types and changes, which is crucial for environmental monitoring and urban planning.

[0129] It can be seen from the above embodiments that the present invention expands the scope of algorithm application, can process more types of images, does not need to limit the original aspect ratio of the image, and improves user convenience.

[0130] The present invention provides a more sophisticated and dynamic complexity evaluation, which can more realistically reflect the changes and features of image content.

[0131] The present invention improves the sensitivity and practicability of complexity evaluation, makes the complexity comparison between different images more obvious, and facilitates further data processing and application.

[0132] The present invention reduces computing resource consumption and speeds up processing, and has significant advantages in particular when processing a large number of images or high-resolution images, thereby improving user experience.

[0133] According to the above process, a comparative experiment is conducted on the image complexity under different conditions. The initial conditions α=2, k=11 are given, and an experimental example with edge filling and a black background is given, such as Figure 4-Figure 9 The Python version used in the present invention is 3.8.0, the Pytorch-cuda version is 11.7, the GPU is GTX-1660-Super, and the CPU is i5-12400F. The comparative test analysis results are shown in Table 1:

[0134] Table 1 Comparative experimental results of image complexity (C)

[0135]

[0136] The experimental results show that the MSE of edge filling-gray is the smallest, the average running time of the present invention is 3945.44s, and the running time of the prior art is 31937.38s, which saves 27991.94s and improves the efficiency by about 87.65%.

[0137] (8) The image complexity results in Table 1 are combined with formula (9) to enlarge the results as shown in Table 2:

[0138] Table 2 Image complexity (C m ) Comparison of experimental results

[0139]

[0140] The experimental results show that the amplified data is more distinguishable, and MES has the same rules as before.

[0141] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for calculating image complexity considering multi-scale fusion, characterized in that: The method includes: S1, determine whether the pixel values ​​of the width and height of the input image are the same. If they are not the same, fill the image background. If the larger value is an odd number, the image size is increased by 1 to become an even number; S2, determines the width of two hyperparameter filters With the number of iterations , calculate the number of iterations The value range of S3, performs coarse-grained processing on the image, and obtains the Renormalized image after sub-coarse granularity ; S4, for The renormalized image obtained after the second coarse granularity Expand and get the extended image ; S5, calculate the The renormalized image obtained after the second coarse granularity Extended image of With The renormalized image obtained by sub-coarse granularization The overlap between ; S6, Calculate image complexity ; In step S2, the filter width , filter width With the number of iterations are two hyperparameters in the coarse-graining process. The width and height of the renormalized image obtained by each coarse-graining are reduced. times, and the minimum width and height is 1, the number of iterations The value range calculation formula is: ； In the formula, For the The side length of the image obtained at the second coarse granularity is an integer, is the side length of the original image; In step S3, the image is subjected to coarse-grained processing, including: In the coarse-graining process, the image is divided into multiple sizes The pixel block is formed by adding the values ​​of each pixel in the pixel block and taking the average value to obtain a pixel point, which is used to replace the original pixel block; each pixel block is traversed to complete a coarse-graining. The renormalized image obtained after the second coarse granularity , the calculation formula is: ； In the formula, For the The renormalized image obtained by the second coarse-graining is Row, No. Column pixels; For the Row, No. List; For the The renormalized image obtained by the second coarse-graining is Row, No. Column pixels; In step S4, The renormalized image obtained after the second coarse granularity Using the nearest neighbor upsampling expansion, we get The renormalized image obtained by sub-coarse granularization Extended image of ; In step S5, calculate the The renormalized image obtained after the second coarse granularity Extended image of With The renormalized image obtained by sub-coarse granularization The overlap between ,include: for No. Row, No. Column pixel points, a vector, and The product of is the vector product; the calculation formula is: ； In the formula, For the The renormalized image obtained by the second coarse-graining is Row, No. Column pixels; In step S6, the image complexity The calculation formula is: ； In the formula, is the component of complexity C, For the The renormalized image obtained by sub-coarse granularization Extended image of With The renormalized image obtained by sub-coarse granularization The overlap between .

2. The image complexity calculation method considering multi-scale fusion according to claim 1, characterized in that: In step S1, after the image background is filled, the image data is normalized. The background filling method is divided into edge filling and positioning filling. Positioning filling is to place the original image on the left side of the background filling area for comparative analysis. The normalization calculation formula is: ； In the formula, are the normalized values ​​of the RGB channels of the pixels in the row and column of the preprocessed image, are the RGB channel values ​​of the pixels in the row and column of the original image respectively.

3. The image complexity calculation method considering multi-scale fusion according to claim 1, characterized in that: Renormalized image The side length is The renormalized image obtained after the second coarse granularity of times, expand the image and The size is consistent; and meets ;in, For the The side length of the image obtained at the second coarse granularity is an integer, is the side length of the expanded image, for The extended image of .

4. The image complexity calculation method considering multi-scale fusion according to claim 1, characterized in that: The image complexity By exponentially magnifying , the expression is: ； In the formula, is the complexity of the enlarged image, C is the complexity of the original image, is the base of natural logarithms.

5. An image complexity calculation system considering multi-scale fusion, characterized in that: The system implements the image complexity calculation method considering multi-scale fusion as claimed in any one of claims 1 to 4, and the system comprises: The image background filling module (1) is used to determine whether the pixel values ​​of the width and height of the input image are the same. If they are not the same, the image background is filled. If the larger value is an odd number, the image size is increased by 1 to become an even number. Iteration number range determination module (2) is used to determine the width of two hyperparameter filters With the number of iterations , calculate the number of iterations The value range of filter width , filter width With the number of iterations are two hyperparameters in the coarse-graining process. The width and height of the renormalized image obtained by each coarse-graining are reduced. times, and the minimum width and height is 1, the number of iterations The value range calculation formula is: ； In the formula, For the The side length of the image obtained at the second coarse granularity is an integer, is the side length of the original image; The renormalized image is obtained by module (3), which is used to determine the two hyperparameter filter widths. With the number of iterations After that, the image is processed into coarse grains, and the first Renormalized image after sub-coarse granularity ; Perform coarse-grained processing on the image, including: In the coarse-graining process, the image is divided into multiple sizes The pixel block is formed by adding the values ​​of each pixel in the pixel block and taking the average value to obtain a pixel point, which is used to replace the original pixel block; each pixel block is traversed to complete a coarse-graining. The renormalized image obtained after the second coarse granularity , the calculation formula is: ； In the formula, For the The renormalized image obtained by the second coarse-graining is Row, No. Column pixels; For the Row, No. List; For the The renormalized image obtained by the second coarse-graining is Row, No. Column pixels; Expand the renormalized image acquisition module (4) to The renormalized image obtained after the second coarse granularity Expand and get the extended image ; for The renormalized image obtained after the second coarse granularity Using the nearest neighbor upsampling expansion, we get The renormalized image obtained by sub-coarse granularization Extended image of ; The overlap calculation module (5) calculates the The renormalized image obtained after the second coarse granularity Extended image of With The renormalized image obtained by sub-coarse granularization The overlap between ; Calculate the The renormalized image obtained after the second coarse granularity Extended image of With The renormalized image obtained by sub-coarse granularization The overlap between ,include: for No. Row, No. Column pixel points, a vector, and The product of is the vector product; the calculation formula is: ； In the formula, For the The renormalized image obtained by the second coarse-graining is Row, No. Column pixels; Image complexity calculation module (6), used to calculate image complexity , magnified by exponential function; image complexity The calculation formula is: ； ； In the formula, is the component of complexity C, For the The renormalized image obtained by sub-coarse granularization Extended image of With The renormalized image obtained by sub-coarse granularization The overlap between .

Images