Automatic image cleaning method based on image quality evaluation

Through an automated cleaning method based on image quality evaluation, the problems of image duplication and blurring in endoscopic technology are solved, efficient screening and utilization of image data are achieved, and the diagnostic accuracy and efficiency of endoscopic examinations are improved.

CN120747553APending Publication Date: 2025-10-03XUZHOU MEDICAL UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510816313.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-06-18
Publication Date
2025-10-03

AI Technical Summary

Technical Problem

Existing endoscopic technology has problems such as image duplication, image blur, and little differentiation between multiple images, resulting in a large number of images with poor usability, affecting data analysis and storage efficiency.

Method used

An automated cleaning method based on image quality evaluation is adopted, images are classified and screened through a redundant recognition algorithm, an image quality evaluation strategy is constructed, image quality thresholds are set, and high-quality images are screened and retained.

Benefits of technology

It achieves dimensionality reduction of image data, improves the efficiency of image quality evaluation and data mining, maximizes the use of image data, and improves the diagnostic accuracy and efficiency of endoscopic examinations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120747553A_ABST
    Figure CN120747553A_ABST
Patent Text Reader

Abstract

The invention discloses an automatic image cleaning method based on image quality evaluation, and relates to the technical field of image processing, and the method comprises the steps: carrying out the classification and screening of collected graphic data based on a redundancy recognition algorithm, achieving the dimension reduction of image data, and providing high-value data for the subsequent image quality evaluation; an image quality evaluation strategy is constructed, quality evaluation is performed on the screened images, and an image-quality score data set is formed, so that image data with low availability is removed, and subsequent cleaning and data mining efficiency is improved; setting an image quality threshold based on a task demand, comparing the quality score in the data set with the image quality threshold, and screening and retaining a corresponding image; personalized image screening is achieved, and image data are utilized to the maximum extent.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an automatic image cleaning method based on image quality evaluation. Background Art

[0002] Endoscopic technology is increasingly used in medical diagnosis. However, the image acquisition process often results in a large number of similar or duplicate images, posing challenges to data analysis and storage. To optimize data quality and improve analysis efficiency, the key lies in the introduction of efficient image cleaning technology. Image cleaning is more than just a simple deduplication process; it involves leveraging advanced computer vision and deep learning technologies to automatically identify and filter high-quality images. Through precise image feature extraction and intelligent comparison, the system can quickly identify the optimal acquisition for the same scenario, reducing data redundancy and storage requirements. The application of image cleaning technology enables endoscopy to more effectively address the challenges posed by data quality and duplicate images. This not only improves the diagnostic accuracy and efficiency of endoscopic examinations but also lays a solid foundation for the intelligent development of medical services.

[0003] Existing methods mostly focus on image preprocessing, performing denoising, enhancement, and segmentation to improve endoscopic image quality and, consequently, the efficiency of endoscopic image mining. However, massive amounts of image data often suffer from issues such as image duplication, image blur that cannot be processed, and limited differentiation between multiple images. This creates further challenges for subsequent image preprocessing and mining, resulting in a large number of images with poor usability. Summary of the Invention

[0004] The purpose of the present invention is to provide an automatic image cleaning method based on image quality evaluation, which is used to solve the problems of image duplication, image blur that cannot be processed, and small differences between multiple images in the above-mentioned background technology.

[0005] To solve the above technical problems, the present invention adopts the following technical solution: an automatic image cleaning method based on image quality evaluation, comprising the following steps:

[0006] S1. Classify and filter the collected graphic data based on the redundancy recognition algorithm;

[0007] S2. Build an image quality evaluation strategy to evaluate the quality of the screened images and form a data set of image-quality scores;

[0008] S3. Set an image quality threshold based on task requirements, compare the quality scores in the data set with the image quality threshold, and filter and retain the corresponding images.

[0009] Preferably, in step S1, the redundancy identification algorithm includes:

[0010] S11. Pre-classification based on the same subject in the same scenario;

[0011] S12. Measure the blurriness of images in the same category and calculate the blurriness of the images.

[0012] S13. In each category, select the image with the lowest blur as the reference image and perform redundancy evaluation. The evaluation formula is:

[0013]

[0014] Among them, cov(x,y) represents the covariance of x and y, σ x ,σ y Represents the standard deviation of x, y, where x, y represent the pixel coordinates in the image.

[0015] S14. Filter and retain image parts with similarity differences.

[0016] Preferably, the ambiguity measurement method is:

[0017]

[0018] Where I(i) is the grayscale value of the i-th pixel in the image, is the average grayscale value of all pixels in the image, and N is the total number of pixels.

[0019] Preferably, in step S2, the image quality evaluation strategy is:

[0020] S21, dividing the image into several small blocks of equal size;

[0021] S22, calculate and summarize the local statistical information of each small block;

[0022] S23. Calculate global statistical information based on the summary information;

[0023] S24. Calculate quality scores using local and global statistical information as inputs to multiple regression equations The calculation formula is as follows:

[0024]

[0025] Among them, K represents the total number of regression equations, k represents the kth regression equation, α k represents the attention weight of the k-th regression equation, Represents the input value of the kth regression equation.

[0026] Preferably, the attention weight α of the kth regression equation is k for:

[0027]

[0028] e k =a(x) T w k ;

[0029] Among them, e k represents the unnormalized attention score, w k represents the weight vector of the k-th regression equation, a(x) T A vector representation of the input features.

[0030] Preferably, in step S3, the image quality threshold is dynamically adjusted based on different task requirements, and images with a quality score greater than the corresponding image quality threshold under the task requirements are retained.

[0031] Beneficial effects: The present invention uses a redundant identification algorithm to classify, measure fuzziness and perform redundant processing on the collected massive image data, thereby achieving dimensionality reduction of image data and providing high-value data for subsequent image quality evaluation. By constructing an image quality evaluation strategy, a data set of image-quality scores is obtained, thereby removing image data with low availability and improving the efficiency of subsequent cleaning and data mining. In addition, dynamic thresholds are set according to different task requirements to achieve personalized image screening and maximize the use of image data. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0033] In the attached figure:

[0034] Figure 1 is a flow chart of the automatic image cleaning method of the present invention;

[0035] Figure 2 Schematic diagram of the redundancy identification algorithm process of the present invention;

[0036] Figure 3 Schematic diagram of the image quality evaluation process of the present invention;

[0037] Figure 4 Schematic diagram of the image cleaning process of the present invention. DETAILED DESCRIPTION

[0038] The following describes the embodiments of the present invention in conjunction with the accompanying drawings. The terms used in the embodiments of the present invention are only used to explain the specific embodiments of the present invention and are not intended to limit the present invention. The following describes the embodiments of the present application in conjunction with the accompanying drawings.

[0039] refer to Figure 1As shown, an automatic image cleaning method based on image quality evaluation includes the following steps:

[0040] S1. The designed redundancy identification algorithm achieves image data dimensionality reduction by classifying, measuring fuzziness, and processing redundancy in the massive amount of collected image data, providing high-value data for subsequent image quality evaluation;

[0041] S2. Designed image quality evaluation method, i.e., constructing an image quality evaluation strategy to further remove image data with low usability by evaluating the quality of the images to be screened;

[0042] S3. The designed image cleaning method sets the image quality threshold according to different task requirements, realizes personalized image screening, maximizes the use of image data, and improves the efficiency of subsequent cleaning and data mining.

[0043] In a specific embodiment, if Figure 2 As shown in Figure 2, the redundancy identification algorithm process is shown, including:

[0044] First, classify the same subject in the same scenario;

[0045] Next, the blurriness of images in the same category is measured using the following formula:

[0046]

[0047] Where I(i) is the grayscale value of the i-th pixel in the image, is the average gray value of all pixels in the image, and N is the total number of pixels;

[0048] Then, in each category, the image with the lowest blur is selected as the reference image for redundancy evaluation. The evaluation formula is:

[0049]

[0050] Among them, cov(x,y) represents the covariance of x and y, σ x ,σ y Represents the standard deviation of x, y, where x, y represent the pixel coordinates in the image.

[0051] Finally, the image parts with similarity differences are filtered and retained.

[0052] In a specific embodiment, if Figure 3 As shown in Figure 1, the image quality evaluation process is shown, where the image quality evaluation strategy is:

[0053] First, the image is divided into many small blocks, each of which has the same size;

[0054] Then, local statistical properties such as mean, variance, local contrast, etc. are calculated for each small block;

[0055] Next, the local features of each small block are summarized and global statistical information such as the mean and variance of the entire image are calculated;

[0056] Finally, a comprehensive quality score is calculated using local and global statistical information as input to multiple regression equations. Used to express the quality level of an image, the formula is as follows:

[0057]

[0058] In the above formula, K represents the total number of regression equations, k represents the kth regression equation, α k represents the attention weight of the k-th regression equation, represents the input value of the kth regression equation; where:

[0059]

[0060] e k =a(x) T w k ;

[0061] In the above formula, e k represents the unnormalized attention score, w k represents the weight vector of the k-th regression equation, a(x) T A vector representation of the input features.

[0062] In one embodiment, reference Figure 4 As shown in the figure, the image cleaning process is demonstrated. The threshold is dynamically adjusted according to the above image quality score and the quality requirements of different tasks. Through comparison and judgment, if it is greater than or equal to the threshold, the image is retained; if it is less than the threshold, the image is discarded.

[0063] The above describes the embodiments of the present invention in detail with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. After knowing the contents described in the present invention, ordinary technicians in this technical field can make several equivalent changes and substitutions without departing from the principles of the present invention. These equivalent changes and substitutions should also be regarded as falling within the scope of protection of the present invention.

Claims

1. An automated image cleaning method based on image quality evaluation, characterized by: The steps include: S1. Classify and filter the collected graphic data based on the redundancy recognition algorithm; S2. Build an image quality evaluation strategy to evaluate the quality of the screened images and form a data set of image-quality scores; S3. Set an image quality threshold based on task requirements, compare the quality scores in the data set with the image quality threshold, and filter and retain the corresponding images.

2. The automatic image cleaning method based on image quality evaluation according to claim 1, characterized in that: In step S1, the redundancy identification algorithm includes: S11. Pre-classification based on the same subject in the same scenario; S12. Measure the blurriness of images in the same category and calculate the blurriness of the images. S13. In each category, select the image with the lowest blur as the reference image and perform redundancy evaluation. The evaluation formula is: Among them, cov(x,y) represents the covariance of x and y, σ x ,σ y Represents the standard deviation of x, y, where x, y represent the pixel coordinates in the image. S14. Filter and retain image parts with similarity differences.

3. The automatic image cleaning method based on image quality evaluation according to claim 2, characterized in that: The ambiguity measurement method is: Where I(i) is the grayscale value of the i-th pixel in the image, is the average grayscale value of all pixels in the image, and N is the total number of pixels.

4. The automatic image cleaning method based on image quality evaluation according to claim 1, characterized in that: In step S2, the image quality evaluation strategy is: S21, dividing the image into several small blocks of equal size; S22, calculate and summarize the local statistical information of each small block; S23. Calculate global statistical information based on the summary information; S24. Calculate quality scores using local and global statistical information as inputs to multiple regression equations The calculation formula is as follows: Among them, K represents the total number of regression equations, k represents the kth regression equation, α k represents the attention weight of the k-th regression equation, Represents the input value of the kth regression equation.

5. The automatic image cleaning method based on image quality evaluation according to claim 4, characterized in that: The attention weight α of the k-th regression equation k for: have been k =a(x) T w k ; Among them, e k represents the unnormalized attention score, w k represents the weight vector of the k-th regression equation, a(x) T A vector representation of the input features.

6. The automatic image cleaning method based on image quality evaluation according to claim 1, characterized in that: In step S3, the image quality threshold is dynamically adjusted based on different task requirements, and images with a quality score greater than the corresponding image quality threshold under the task requirements are retained.