Image screening method and device based on Matlab

Abstract

The invention discloses an image screening method and device based on Matlab, and belongs to the technical field of image processing, and the method comprises the following steps: carrying out crawler collection on related images on a webpage; data cleaning is carried out on the collected images, common image format files are filtered out, and an increasing number is distributed to each cleaned image file; performing nonlinear dimension reduction processing on the cleaned image data; performing quality evaluation on the image data after dimension reduction; and analyzing the correlation between the selected most representative images by using a Pearson correlation coefficient, and verifying the representativeness and consistency of the selected images. According to the method, the high-quality image with characteristics can be automatically screened out, subjective deviation in manual evaluation is avoided, and the image screening efficiency, the result objectivity and the quality control capability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and more specifically, to an image screening method and device based on Matlab. Background Art

[0002] In the process of product design, designers in the fields of visual design, packaging design, advertising design, etc. often face a common challenge: how to efficiently and accurately select image elements that meet specific design requirements from a large amount of image resources. These design elements require not only high quality of the image itself, such as excellent performance in clarity and color reproduction, but also specific design features that match the project requirements, such as specific color matching, texture, shape and outline, etc.

[0003] At present, the main method of image screening in design still relies on manual operation by designers. Although this method can ensure the professionalism and pertinence of the screening results to a certain extent, it also exposes a series of limitations and challenges that cannot be ignored.

[0004] In the absence of specific references, screening image samples is a complex task. When faced with a large image library, manual screening will involve highly repetitive tasks, which can easily lead to fatigue and reduced efficiency, affecting the screening effect. At the same time, different design experts often have differences in aesthetic concepts, design styles, and evaluation criteria. This difference can lead to subjective bias in the image screening process, so that the same batch of images may produce completely different results when screened by different designers. This inconsistency will not only affect the overall effect of the design project, but also bring unnecessary troubles and troubles to the subsequent design work. Moreover, as time goes by, the designer's aesthetic fatigue and judgment criteria may change, and the judgment criteria for image quality and features will also be adjusted accordingly, making it difficult to maintain high standards of quality control, and it is easy to miss some high-quality images or select low-quality images. In addition, the process of manual screening of images often lacks scientific basis and quantitative standards to guide the designer's judgment. This unscientific screening method not only makes it difficult to ensure the consistency and reliability of the screening results, but also makes the designer lack clear goals and guidance in the screening process, thereby increasing the difficulty and uncertainty of the screening work. Summary of the invention

[0005] An object of the present invention is to provide an image screening method and device based on Matlab to solve the problems of low efficiency, strong subjectivity, difficult quality control and complex feature recognition in the above-mentioned manual image sample screening process.

[0006] According to a first aspect of the present invention, there is provided an image screening method based on Matlab, comprising the following steps:

[0007] Step 1: Crawling and collecting relevant images on web pages;

[0008] Step 2: Clean the collected images, filter out common image format files, and assign an increasing number to each cleaned image file;

[0009] Step 3: Perform nonlinear dimensionality reduction on the cleaned image data;

[0010] Step 4: Perform quality assessment on the image data after dimensionality reduction;

[0011] Step 5: Use the Pearson correlation coefficient to analyze the correlation between the most representative images selected to verify the representativeness and consistency of the selected images.

[0012] Optionally, in the step of performing crawler collection on the relevant images on the web page, data conversion is performed on the image, and the image data is adjusted to a size of 512x512 pixels and then saved in a target directory in PNG format.

[0013] Optionally, in the step of data cleaning the collected images, use Python to adjust the naming of the image data, specify a folder path, then obtain a list of all files in the folder, and sort the file list; filter out all files in the list that have common image formats.

[0014] Optionally, use a column order function to assign an incremental number starting from 1 to each file, and rename each file to this incremental number while keeping the original extension unchanged; while renaming each file, if the new file name does not exist, open each image file, convert it to PNG format, and save it to the new file path.

[0015] Optionally, the step of performing nonlinear dimensionality reduction processing on the cleaned image data includes:

[0016] Blend all image data;

[0017] Based on Matlab, the T-SNE (t-Distributed Stochastic Neighbor Embedding, a machine learning algorithm for high-dimensional data visualization) method is used to directly perform nonlinear dimensionality reduction on image data to two-dimensional space;

[0018] Calculate the similarity and probability distribution between data points in two-dimensional space;

[0019] Minimize the difference between the probability distribution of high-dimensional data and the probability distribution of low-dimensional data, optimize the position of data points in low-dimensional space, and visualize image data.

[0020] Optionally, the difference between the probability distribution of high-dimensional data and the probability distribution of low-dimensional data is measured by the KL divergence (Kullback-Leibler Divergence) metric, and a gradient descent optimization algorithm is used to iteratively adjust the positions of data points to obtain a low-dimensional representation that reflects the similarity relationship in the high-dimensional space, so as to perform effective image data visualization.

[0021] Optionally, in the step of performing quality assessment on the image data after dimensionality reduction, a NIQE (Naturalness Image Quality Evaluator, a non-reference image quality assessment algorithm) evaluation model in Matlab is used to perform quality assessment on the image data after dimensionality reduction, including:

[0022] Image features are extracted from the original image library, and the multivariate Gaussian (MVG) model is used to fit these features, and the relationship between image features and image quality is trained. Finally, NIQE is calculated based on the parameters obtained from the multivariate Gaussian model:

[0023] ;

[0024] in and Represents the mean of the MVG model for natural and generated images, and The covariance matrix of the multivariate Gaussian model representing natural and generated images;

[0025] After evaluating the NIQE index of the collected image samples, the obtained data is quantitatively analyzed to obtain the results.

[0026] Optionally, the three images with the smallest values ​​using the NIQE evaluation model are selected as the most representative images, and the image with the largest value using the NIQE evaluation model is selected as the least representative image.

[0027] Optionally, the step of analyzing the correlation between the selected most representative images using the Pearson correlation coefficient comprises:

[0028] Test whether there is a statistically significant relationship between X and Y;

[0029] Analyze the positive and negative direction of the correlation coefficient and the degree of correlation.

[0030] According to a second aspect of the present invention, there is provided an image screening device based on Matlab, comprising:

[0031] A data collection module is used to crawl and collect relevant images on web pages;

[0032] The data cleaning module is used to clean the collected images, filter out common image format files, and assign an increasing number to each cleaned image file;

[0033] Data dimension reduction module, used to perform nonlinear dimension reduction processing on the cleaned image data;

[0034] Data evaluation module, used to evaluate the quality of image data after dimensionality reduction;

[0035] The data verification module uses the Pearson correlation coefficient to analyze the correlation between the most representative images selected to verify the representativeness and consistency of the selected images.

[0036] The Matlab-based image screening method and device disclosed in the present invention have the following technical effects:

[0037] Images of different sizes and formats are uniformly converted to 512x512 pixels and saved in PNG format to ensure the consistency and comparability of image data;

[0038] Using the column order function to assign an incremental number to each file facilitates the tracking and management of image data, so that each image has a unique identifier;

[0039] By minimizing the difference between high-dimensional and low-dimensional probability distributions, T-SNE optimizes the location of data points so that low-dimensional representation can more accurately reflect the characteristics of high-dimensional data, achieving effective image visualization, allowing researchers to intuitively observe and analyze the relationship and distribution characteristics between image data;

[0040] The NIQE function generates an objective quality score, which provides a reliable basis for the quantitative evaluation of image quality. The Pearson correlation coefficient analysis can further screen out image subsets with high correlation, thereby improving the accuracy and reliability of subsequent analysis.

[0041] The present invention uses Matlab's NIQE function and T-SNE technology to establish an image screening model, which can automatically screen out high-quality images with features, and can significantly improve the drawbacks of featureless image screening in current designs. Not only can the image quality be evaluated without reference, but also targeted screening can be performed in combination with image features. The automated image screening model improves the efficiency of image screening, the objectivity of the results, and the quality control capability, and can be widely used in product design, image database management, computer vision and other fields.

[0042] Further features and advantages of the present invention will become apparent from the following detailed description of exemplary embodiments of the present invention with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.

[0044] Figure 1 The overall flow chart of the image screening method based on Matlab provided by the embodiment of the present invention;

[0045] Figure 2 Taking a cigarette box image as an example, the T-SNE clustering dimension reduction visualization image data result diagram is shown;

[0046] Figure 3 Taking the cigarette box image as an example, after evaluating it through the NIQE index, the obtained data is quantitatively analyzed to obtain the result curve distribution diagram;

[0047] Figure 4 This is a data graph that uses the Pearson correlation coefficient to analyze the correlation between representative images and to test whether there is a statistically significant relationship (P<0.05) between XY;

[0048] Figure 5 It is a coefficient heat map that reflects the positive and negative direction of the correlation coefficient and the degree of correlation. DETAILED DESCRIPTION

[0049] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention unless otherwise specifically stated.

[0050] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0051] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

[0052] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0053] The present invention proposes an embodiment of an image screening method based on Matlab. Specifically, Figure 1 As shown, the following steps are included:

[0054] Step 1: Crawling and collecting relevant images on web pages, reducing the tediousness and time-consuming process of manually collecting images;

[0055] Step 2: Clean the collected images, filter out common image format files, and assign an incremental number to each cleaned image file; specifically, filter out non-image files (such as PDF, text files, etc.) and uncommon or unsupported image formats (such as BMP, TIFF, etc., unless there is a specific requirement), and retain common image formats (such as JPEG, PNG, GIF, etc.);

[0056] Step 3: Perform nonlinear dimensionality reduction processing on the cleaned image data; this is helpful to extract key information from a large number of images and reduce computational complexity;

[0057] Step 4: Perform quality assessment on the image data after dimensionality reduction. By screening, the number of images to be processed is reduced, and the overall processing efficiency is improved;

[0058] Step 5: Use the Pearson correlation coefficient to analyze the correlation between the most representative images selected to verify the representativeness and consistency of the selected images.

[0059] The image screening method based on Matlab in the embodiment of the present invention can efficiently screen out high-quality and representative images through a series of steps such as crawler collection, data cleaning, nonlinear dimensionality reduction, quality assessment and correlation analysis, which not only improves the efficiency and accuracy of image processing, but also provides a reliable foundation for subsequent image analysis, recognition and other applications.

[0060] In an embodiment of the present invention, in the step of crawling and collecting relevant images on a web page, since the initially collected images have different sizes and image formats, Python is used to define the source directory and the target directory, and then the source directory and all its subdirectories are traversed to find all image files; then, the script creates the target directory if it does not exist. For each image file, the script converts its path to a path relative to the source directory, and then creates the same subdirectory structure in the target directory; performs data conversion on the image, adjusts the image data to a size of 512x512 pixels, and then saves it in the target directory in PNG format. Among them, for each image file found, the script converts its path to a path relative to the source directory. The use of relative paths makes the location of image files more flexible and not limited by absolute paths. The unified image size and format enhance the compatibility of images, so that processed images can be more easily shared and exchanged between different platforms, software and tools.

[0061] In the embodiment of the present invention, in the step of data cleaning of the collected images, Python is used to adjust the naming of the image data, specify a folder path, and then obtain a list of all files in the folder and sort the file list; filter out all files in the list with common image formats. Using Python to adjust the naming of image data, obtain a file list, sort, and filter common image formats can significantly improve the readability, manageability, compatibility, and processing efficiency of image data.

[0062] In an embodiment of the present invention, a column order function, specifically an enumerate() function, is used to assign an incremental number starting from 1 to each file, and each file is renamed to this incremental number while keeping the original extension unchanged; while renaming each file, if the new file name does not exist, each image file is opened, converted to PNG format, and saved to a new file path. In the process of renaming and converting files, by checking whether the new file name exists, it is determined whether the file has been processed, thereby avoiding repeated operations on files that have been processed, ensuring that each file is processed only once, and avoiding data corruption or inconsistency caused by repeated processing. The entire renaming and conversion process is completed automatically through Python scripts without manual intervention, reducing errors and omissions caused by human operations, and can quickly process a large number of files, improving work efficiency.

[0063] Since there is a large sample imbalance between the image data, mixing all samples and then selecting representative cigarettes will result in a smaller probability of cigarettes with a smaller data volume being selected. Therefore, nonlinear dimensionality reduction processing is performed on the cleaned image data. The specific steps include:

[0064] All image data are mixed to avoid neglecting samples with small data volumes;

[0065] Based on Matlab, the T-SNE method is used to directly perform nonlinear dimensionality reduction on image data to two-dimensional space;

[0066] Calculate the similarity and probability distribution between data points in two-dimensional space;

[0067] Minimize the difference between the probability distribution of high-dimensional data and the probability distribution of low-dimensional data, optimize the position of data points in low-dimensional space, and visualize image data.

[0068] Specifically, the process of performing nonlinear dimensionality reduction on image data is as follows:

[0069] a) In the T-SNE algorithm, the similarity between data points in the high-dimensional space is first calculated, measured by the distance in the high-dimensional space. However, T-SNE does not use these distances directly, but calculates a probability distribution for each data point, which represents the "proximity relationship" of the data point with other data points. When two data points are close in the high-dimensional space, they are more likely to be selected as neighbors of each other; conversely, if the two data points are far apart, they are less likely to be selected as neighbors. This probability distribution is calculated by the softmax function, and its smoothness is controlled by a parameter called "perplexity".

[0070] b) Secondly, map the points in the high-dimensional space to the low-dimensional space and find a new position for each data point, and calculate the similarity between the data points in the low-dimensional space. T-SNE will also calculate a probability distribution for each data point in the low-dimensional space.

[0071] c) Finally, the T-SNE algorithm optimizes the position of data points in low-dimensional space by minimizing the difference between the probability distributions in high-dimensional space and low-dimensional space. The difference between the probability distribution of high-dimensional data and the probability distribution of low-dimensional data is measured by KL divergence (Kullback-Leibler Divergence). The purpose of T-SNE is to find a layout of data points in a low-dimensional space so that the probability distribution in the high-dimensional space is as close as possible to the probability distribution in the low-dimensional space. To achieve this goal, T-SNE uses optimization algorithms such as gradient descent to adjust the position of data points in the low-dimensional space. By iterating this process, the final low-dimensional representation can better reflect the similarity relationship between data points in the high-dimensional space, thereby achieving effective data visualization. The Euclidean distance is used, and the maximum number of iterations is 1000 to obtain the visualization result, as shown in the figure. Figure 2 As shown in the figure, the chart represents different cigarette brands with points in multiple colors such as red, yellow and blue. These points are distributed in a two-dimensional space, representing their clustering in T-SNE. Each brand point has a label showing which category they belong to. The chart is a two-dimensional coordinate system. The horizontal and vertical axes represent the two dimensions calculated by the T-SNE algorithm. By observing the distribution of points, the similarity and distance between different types of cigarettes can be analyzed. For example, if two points are very close, then the types of cigarettes they represent are very similar in some features; if the two points are far apart, then the types of cigarettes they represent have very different features.

[0072] Based on the T-SNE clustering visualization results and the fact that the eight brands of cigarettes have a very wide audience in the current society, the eight brands of cigarettes are analyzed separately to observe the commonalities and the least representative features between the most representative packages in the collected images. Since there is no "perfect sample" for comparative analysis in the current market, the objective evaluation index NIQE evaluation model without reference images is selected to quantitatively analyze the results. The NIQE evaluation model quantifies the quality of the image by extracting the statistical features of the image and comparing it with the known statistical model of natural images.

[0073] In an embodiment of the present invention, in the step of performing quality assessment on the image data after dimensionality reduction, the NIQE evaluation model in Matlab is used to perform quality assessment on the image data after dimensionality reduction, including:

[0074] Image features are extracted from the original image library, and the multivariate Gaussian (MVG) model is used to fit these features. The relationship between image features and image quality is trained, and finally NIQE is calculated based on the parameters obtained from the multivariate Gaussian model. The smaller the NIQE, the higher the image quality:

[0075] ;

[0076] in and Represents the mean of the MVG model for natural and generated images, and Represents the covariance matrix of the multivariate Gaussian model of natural images and generated images; this indicator is mainly used to evaluate the naturalness of an image. The smaller its value, the more natural the image is and the more it conforms to the visual aesthetics of the human eye. The obtained image is evaluated by the NIQE indicator and then quantitatively analyzed to obtain the data NIQE result distribution curve. If Figure 3 The figure shows the distribution curve of the results obtained by quantitative analysis of the data obtained after the image samples of eight brands of cigarettes were evaluated by the NIQE index. The figure intuitively shows the distribution of scores of the eight brands of cigarettes on the NIQE evaluation index in the form of a line graph, providing consumers with reference information on the satisfaction of different brands of cigarettes.

[0077] After evaluating the NIQE index of the collected image samples, the obtained data is quantitatively analyzed to obtain the results.

[0078] In the embodiment of the present invention, the three images with the smallest NIQE evaluation model value are selected as the most representative images, and the image with the largest NIQE evaluation model value is selected as the least representative image. By selecting the image with the best quality as the representative, it is possible to avoid introducing a large number of poor quality and redundant images in subsequent processing or analysis; by analyzing the least representative image, the reasons for the decline in image quality, such as noise, blur, distortion, etc., can be understood.

[0079] In an embodiment of the present invention, the step of analyzing the correlation between the selected most representative images using the Pearson correlation coefficient includes:

[0080] Test whether there is a statistically significant relationship between XY, where X and Y represent the eigenvalues ​​of any two of the three most representative images (these eigenvalues ​​are calculated based on the NIQE index). The Pearson correlation coefficient is calculated as:

[0081] ;

[0082] Where n is the number of data points, x i and i are the i-th observation values ​​of X and Y respectively, and are the means of X and Y respectively. Set a significance level (such as 0.05). If the calculated P value is less than this significance level, it is considered that there is a statistically significant relationship between X and Y. Otherwise, there is no correlation between the two variables. Figure 4 The numbers in the cells (such as 0.997, 0.964, etc.) represent the degree of association or similarity between the two brands. The closer the value is to 1, the higher the degree of association or similarity between the two brands. For example, the number in the cell corresponding to the first brand and the second brand is 0.997, indicating that the first brand and the second brand have a high degree of similarity or association under this specific measurement.

[0083] Analyze the positive and negative direction of the correlation coefficient and the degree of correlation. If the correlation coefficient r is positive, it means that there is a positive correlation between X and Y. If the correlation coefficient r is negative, it means that there is a negative correlation between X and Y. The closer the absolute value of the correlation coefficient r is to 1, the stronger the linear relationship between X and Y. Figure 5 As shown, most of the data values ​​are greater than 0.8, reflecting that the stronger the correlation between the corresponding images, the stronger the similarity or consistency of the images in terms of features, and the selected representative images meet the requirements.

[0084] The present invention also provides an embodiment of an image screening device based on Matlab, comprising:

[0085] A data collection module is used to crawl and collect relevant images on web pages;

[0086] The data cleaning module is used to clean the collected images, filter out common image format files, and assign an increasing number to each cleaned image file;

[0087] Data dimension reduction module, used to perform nonlinear dimension reduction processing on the cleaned image data;

[0088] Data evaluation module, used to evaluate the quality of image data after dimensionality reduction;

[0089] The data verification module uses the Pearson correlation coefficient to analyze the correlation between the most representative images selected to verify the representativeness and consistency of the selected images.

[0090] It should be noted that the information interaction, execution process and other contents between the above-mentioned modules are based on the same concept as the embodiment of the method of the present application, and are devices corresponding to the above-mentioned Matlab-based image screening method. All implementation methods in the above-mentioned method embodiment are applicable to the embodiments of the device. Its specific functions and technical effects brought about can be found in the method embodiment part, which will not be repeated here.

[0091] The embodiment of the present invention converts complex high-dimensional image feature data into easy-to-understand low-dimensional graphics through T-SNE. Through visualization, it is easier to identify patterns and features in the image, assist screening and classification, and intuitively understand the distribution of image features; NIQE is based on the natural scene statistics (NSS) model, and evaluates the quality of the image by calculating the statistical characteristics of the image. It can evaluate its quality without a reference image and generate an objective quality score. Matlab and T-SNE are used to perform clustering and classification analysis in the space after dimensionality reduction. Combined with the NIQE score, high-quality and low-quality images can be clearly distinguished in the space after dimensionality reduction, the screening process can be optimized, and subjective bias in manual evaluation can be avoided, thereby improving the efficiency of image screening, the objectivity of the results, and the quality control ability.

[0092] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings, but the above is only a preferred embodiment of the present invention. It should be noted that the technical features involved in the above embodiments and their preferred methods can be reasonably combined and matched into a variety of equivalent schemes by those skilled in the art without departing from or changing the design ideas and technical effects of the present invention; therefore, the present invention is not limited to the scope of implementation shown in the drawings, and all changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which still do not exceed the spirit covered by the specification and drawings, should be within the protection scope of the present invention.

Claims

1. A Matlab-based image screening method, characterized in that: The following steps are involved: Step 1: Crawling and collecting relevant images on web pages; Step 2: Clean the collected images, filter out common image format files, and assign an increasing number to each cleaned image file; Step 3: Perform nonlinear dimensionality reduction processing on the cleaned image data; Step 4: Perform quality assessment on the image data after dimensionality reduction; Step 5: Use the Pearson correlation coefficient to analyze the correlation between the most representative images selected to verify the representativeness and consistency of the selected images.

2. The image screening method based on Matlab according to claim 1, characterized in that: In the step of crawling and collecting relevant images on the web page, the image data is converted and the image data is adjusted to a size of 512x512 pixels and then saved in a target directory in PNG format.

3. The image screening method based on Matlab according to claim 1, characterized in that, In the step of data cleaning of the collected images, Python is used to adjust the naming of the image data, specify a folder path, then obtain a list of all files in this folder and sort the file list; filter out all files in the list with common image formats.

4. The image screening method based on Matlab according to claim 3, characterized in that: Use the column sequence function to assign an incremental number starting from 1 to each file, and rename each file to this incremental number while keeping the original extension unchanged; while renaming each file, if the new file name does not exist, open each image file, convert it to PNG format, and save it to the new file path.

5. The image screening method based on Matlab according to claim 1, characterized in that: The steps of performing nonlinear dimensionality reduction on the cleaned image data include: Blend all image data; Based on Matlab, the T-SNE method is used to directly perform nonlinear dimensionality reduction on image data to two-dimensional space; Calculate the similarity and probability distribution between data points in two-dimensional space; Minimize the difference between the probability distribution of high-dimensional data and the probability distribution of low-dimensional data, optimize the position of data points in low-dimensional space, and visualize image data.

6. The image screening method based on Matlab according to claim 5, characterized in that: The KL divergence (Kullback-Leibler Divergence) metric is used to measure the difference between the probability distribution of high-dimensional data and the probability distribution of low-dimensional data. The gradient descent optimization algorithm is used to iteratively adjust the positions of data points to obtain a low-dimensional representation that reflects the similarity relationship in the high-dimensional space, thereby performing effective image data visualization.

7. The image screening method based on Matlab according to claim 1, characterized in that: In the step of evaluating the quality of the image data after dimensionality reduction, the NIQE evaluation model in Matlab is used to evaluate the quality of the image data after dimensionality reduction, including: Image features are extracted from the original image library, and the multivariate Gaussian (MVG) model is used to fit these features, and the relationship between image features and image quality is trained. Finally, NIQE is calculated based on the parameters obtained from the multivariate Gaussian model: ； in and Represents the mean of the MVG model for natural and generated images, and The covariance matrix of the multivariate Gaussian model representing natural and generated images; After evaluating the NIQE index of the collected image samples, the obtained data is quantitatively analyzed to obtain the results.

8. The image screening method based on Matlab according to claim 7, characterized in that: The three images with the smallest values ​​using the NIQE evaluation model were selected as the most representative images, and the image with the largest value using the NIQE evaluation model was selected as the least representative image.

9. The image screening method based on Matlab according to claim 8, characterized in that: The steps to analyze the correlation between the selected most representative images using the Pearson correlation coefficient include: Test whether there is a statistically significant relationship between X and Y; Analyze the positive and negative direction of the correlation coefficient and the degree of correlation.

10. An image screening device based on Matlab, characterized in that: include: A data collection module is used to crawl and collect relevant images on web pages; The data cleaning module is used to clean the collected images, filter out common image format files, and assign an increasing number to each cleaned image file; Data dimension reduction module, used to perform nonlinear dimension reduction processing on the cleaned image data; Data evaluation module, used to evaluate the quality of image data after dimensionality reduction; The data verification module uses the Pearson correlation coefficient to analyze the correlation between the most representative images selected to verify the representativeness and consistency of the selected images.

Images