A picture duplicate detection method, database, device, readable storage medium and system

Abstract

The application discloses a picture duplicate checking method, a database, an apparatus, a readable storage medium and a system, and a picture duplicate checking method based on Python, which comprises the following steps: 1) obtaining a plurality of to-be-checked and existing color pictures; 2) scaling the color pictures; the scaling obtains a 32*32 pixel RGB three-channel image; 3) combining the images after DCT operation of each dimension of the RGB three-channel image to obtain a 32*32 three-dimensional matrix of the image; 4) calculating the arithmetic mean of different 8*8 regions to obtain an 8*8 format image feature matrix; 5) converting into a 64-bit list to obtain an average value, a binary number sequence and a decimal number sequence; and the like; and a duplicate checking method for gray-scale pictures; the application finds the matrix of each channel with more information through induction, thereby obtaining a more real hash value, and the duplicate picture checking is more complete and accurate, and the efficiency is higher.

Description

Technical Field

[0001] This invention relates to the field of general defect-type algorithms, and in particular to an image deduplication method based on Python. Background Technology

[0002] In practical projects involving general-purpose defect-type algorithms, the provided image data often suffers from problems such as low resolution, unclear categories, and partial duplication, which poses considerable difficulties for preprocessing work such as data cleaning.

[0003] In addition, to hide image information, customers will uniformly set the names of images from different batches and categories to serial numbers from 0 to N, which will lead to the following situations: (1) Images that are completely identical in the same batch of data are named with different serial numbers.

[0004] (2) Images that are partially identical in different batches of data are named with different serial numbers and stored in different folders.

[0005] Furthermore, with a large number of image samples, it is difficult to manually filter out the duplicates. Directly labeling duplicate images without intervention will indirectly reduce sample diversity, thereby reducing labeling efficiency.

[0006] To address the aforementioned issues, this case study presents an image plagiarism detection technology based on Python.

[0007] The information disclosed in this section is only for understanding the background of the inventive concept of this disclosure, and therefore may include information that does not constitute prior art. Summary of the Invention

[0008] The purpose of this invention is to provide a Python-based image deduplication method that improves efficiency, accuracy, and the effectiveness of filtering duplicate images.

[0009] To achieve the above objectives, the solution of this invention is: an image deduplication method based on Python, comprising the following steps: 1) Obtain multiple images to be checked for plagiarism and existing color images; 2) The color image is scaled using the resize function built into the OpenCV library; the scaling results in a 32*32 pixel RGB three-channel image. 3) Perform DCT operations on the image according to each dimension of the RGB three channels and combine them to obtain a 32*32 three-dimensional matrix of the image, and save it as a color image; 4) In the 32*32 three-dimensional matrix of the image, the R channel takes the upper left 8*8, middle 8*8, and lower right 8*8 regions, and the G... The R channel takes the middle 8*8 and lower left 8*8 regions, and the B channel takes the middle 8*8 and upper right 8*8 regions. The image feature matrix in 8*8 format is obtained by calculating the arithmetic mean of the above R, G and B channel regions. 5) Convert the image feature matrix into a 64-bit list and calculate the average; compare the 64-bit list with the average value. If the value is less than the average, assign 0; if the value is greater than the average, assign 1 to obtain the binary sequence of the image mean. Then, loop through the binary sequence of the image mean, using three numbers as the step size, and convert it to decimal to obtain the decimal sequence of the image mean. 6) Pad the tens digit of any number less than 10 in the middle of the decimal sequence of the image's mean with zeros to obtain the duplicate to be checked and the existing color. Color image hash value; 7) Let dictionary A contain the key and value of the color image to be checked for duplicates, and let dictionary B contain the existing color images... Image key and existing color image value; the color image value to be checked and the existing color image value include the hash value of the color image to be checked and the original path of the color image to be checked and the existing color image; the color image key to be checked and the existing color image key are respectively marked with a sequence. For each "key" in dictionary A, a binary search is performed in dictionary B; comparing dictionary A and dictionary B, let the sample of the color image value to be checked be Ax, and let the sample of the existing color image value be B(0-n); if Ax is less than B0, the color image to be checked and the existing color image are not repeated; if Ax is greater than Bn, the color image to be checked and the existing color image are not repeated; if Ax is in the interval of B(0-n), it is compared with the middle value of B, and the range is narrowed by binary search to determine whether they are completely equal. If they are completely equal, the color image key to be checked and the existing color image key are output; otherwise, the color image to be checked and the existing color image are not repeated.

[0010] The solution of this invention also includes a Python-based image deduplication method, comprising the following steps: 1) Obtain multiple images to be checked for plagiarism and existing grayscale images; 2) The grayscale image is scaled using the resize function built into the OpenCV library; the scaling results in a 32*32 pixel grayscale image. 3) The grayscale image is combined after performing a single-channel DCT operation on the image to obtain the 32*32 image matrix; 4) In the 8x8 region in the middle of the image matrix, obtain an 8x8 format image by taking the arithmetic mean of the 8x8 region. Like the feature matrix; 5) Convert the image feature matrix into a 64-bit list and calculate the average; compare the 64-bit list with the average value. If the value is less than the average, assign 0; if the value is greater than the average, assign 1 to obtain the binary sequence of the image mean. Then, loop through the binary sequence of the image mean, using three numbers as the step size, and convert it to decimal to obtain the decimal sequence of the image mean. 6) Pad the tens digit of any number less than 10 in the middle of the decimal sequence of the image's mean with zeros to obtain the duplicates to be checked and the existing gray numbers. Image hash value; 7) Let dictionary A contain the key and value of the grayscale image to be checked for duplicates, and let dictionary B contain the existing grayscale images. Image key and existing grayscale image value; the grayscale image value to be checked and the existing grayscale image value include the hash value of the grayscale image to be checked and the original path of the grayscale image to be checked and the existing grayscale image; the grayscale image key to be checked and the existing grayscale image key are respectively sequence marked, and for each "key" in dictionary A, a binary search is performed in dictionary B; comparing dictionary A and dictionary B, let the sample of the grayscale image value to be checked be Ax, and let the sample of the existing grayscale image value be B(0-n); if Ax is less than B0, the grayscale image to be checked and the existing grayscale image are not duplicated; if Ax is greater than Bn, the grayscale image to be checked and the existing grayscale image are not duplicated; if Ax is in the interval of B(0-n), it is compared with the middle value of B, and the range is narrowed by binary search to determine whether they are completely equal. If they are completely equal, the grayscale image key to be checked and the existing grayscale image key are output; otherwise, the grayscale image to be checked and the existing grayscale image are not duplicated.

[0011] The solution of the present invention also includes a database for an image deduplication method based on Python, comprising storage space, wherein the storage space distributes the aforementioned image deduplication method based on Python and related information.

[0012] The solution of the present invention also includes an image plagiarism detection device, comprising a memory and a processor; the memory is used to store a computer program; the processor is used to implement the above-described Python-based image plagiarism detection method when the computer program is executed.

[0013] The solution of the present invention also includes a computer-readable storage medium, characterized in that the storage medium stores a computer program, which, when executed by a processor, implements the above-described image deduplication method based on Python.

[0014] The solution of this invention also includes an image plagiarism detection system, comprising the image plagiarism detection device as described above, an internet interface, and peripherals, wherein the peripherals include a monitor, a keyboard, and a mouse. The image plagiarism detection device is connected to the Internet interface and peripheral devices in sequence; The monitor is used to display the above-mentioned Python-based image deduplication method; The keyboard and mouse are used to control and modify the above-mentioned Python-based image deduplication method; The internet interface is used to connect the above-mentioned Python-based image deduplication method to the internet.

[0015] After adopting the above scheme, the gain effect of the present invention is as follows: 1. This invention provides a general image processing method in the field of defect-based algorithms. It uses Direct Current Transform (DCT) to obtain low-frequency information, transforming the image from the pixel domain to the frequency domain. Images generally contain a lot of redundancy and correlation. After transformation to the frequency domain, only a small portion of the frequency components have coefficients that are not zero; most coefficients are zero (or close to zero). The coefficient matrix after DCT transformation shows increasingly higher frequencies from the top left to the bottom right corner. Therefore, the image's energy is mainly retained in the low-frequency coefficients in the top left corner. The method provided by this invention retains more information and has higher fidelity.

[0016] 2. The sampling method derived from the inventor's innovative experiments finds a more informative matrix for each channel through induction, thereby obtaining a more realistic hash value. It has a better effect on finding all duplicate images accurately and efficiently. Attached Figure Description

[0017] Figure 1 This is an illustration of the steps of the image plagiarism detection method of the present invention; Figure 2 This is a schematic diagram of the interface of an image plagiarism detection system based on Python according to the present invention; Figure 3 This is a schematic diagram of the steps in Embodiment 1 of the present invention. Figure 1 ; Figure 4 This is a schematic diagram of the steps in Embodiment 1 of the present invention. Figure 2 ; Figure 5 This is a schematic diagram of the steps in Embodiment 1 of the present invention. Figure 3 ; Figure 6 This is a schematic diagram of the steps in Embodiment 1 of the present invention. Figure 4 ; Figure 7 This is a schematic diagram of the steps in Embodiment 1 of the present invention. Figure 5 ; Figure 8 This is a schematic diagram of the steps in Embodiment 1 of the present invention. Figure 6 ; Figure 9 This is a schematic diagram of the steps in Embodiment 1 of the present invention. Figure 7 ; Figure 10 This is a schematic diagram of the steps in Embodiment 1 of the present invention. Figure 8 ; Figure 11 This is a schematic diagram of the steps in Embodiment 1 of the present invention. Figure 9 ; Figure 12 This is a schematic diagram of the steps in Embodiment 1 of the present invention. Figure 10 ; Figure 13 This is a schematic diagram of the steps in Embodiment 1 of the present invention. Figure 10 one; Figure 14 This is a schematic diagram of the steps in Embodiment 1 of the present invention. Figure 10 two. Detailed Implementation

[0018] In the following description, numerous specific details are set forth for illustrative purposes to provide a comprehensive understanding of various exemplary embodiments. However, it will be apparent that various exemplary embodiments may be implemented without these specific details or with one or more equivalent arrangements. Furthermore, the various exemplary embodiments may be different, but not necessarily exclusive. For example, specific configurations and features of exemplary embodiments may be used or implemented in another exemplary embodiment without departing from the inventive concept.

[0019] Those skilled in the art will know that Python was designed in the early 1990s by Guido van Rossum of the Netherlands Organization for Mathematical and Computer Science as an alternative to a language called ABC. Python provides efficient high-level data structures and enables simple and efficient object-oriented programming. Python's syntax, dynamic typing, and interpreted language nature make it a popular programming language for scripting and rapid application development on most platforms. With continuous updates and the addition of new features, it has gradually been used for independent, large-scale project development.

[0020] Those skilled in the art will know that OpenCV is a cross-platform computer vision and machine learning software library released under the Apache 2.0 license (open source), which can run on Linux, Windows, Android and Mac OS operating systems.

[0021] Those skilled in the art will know that the Discrete Cosine Transform (DCT) is a transform related to the Fourier Transform. It is similar to the Discrete Fourier Transform (DFT), but uses only real numbers. The Discrete Cosine Transform is equivalent to a Discrete Fourier Transform approximately twice its length, performed on a real even function (because the Fourier transform of a real even function is still a real even function). In some variations, the input or output positions need to be shifted by half a unit (there are 8 standard types of DCT, 4 of which are common).

[0022] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] This invention provides a Python-based image deduplication method, such as... Figure 1 As shown, an image plagiarism detection system based on Python includes the following steps: 1) Obtain multiple images to be checked for plagiarism and existing images; 2) Use the resize function built into the OpenCV library to scale the image; the scaling will result in a 32*32 pixel, RGB three-channel image or a grayscale image; 3) Perform DCT operations on the RGB three channels or the grayscale single channel of the image separately, and then combine them to obtain a 32*32 three-dimensional matrix or a 32*32 matrix of the image; the DCT formula is as follows:

[0024] 4) In the 32*32 three-dimensional matrix of the color image, the R channel takes the upper left 8*8, middle 8*8 and lower right 8*8 regions, the G channel takes the middle 8*8 and lower left 8*8 regions, and the B channel takes the middle 8*8 and upper right 8*8 regions. The color image feature matrix in 8*8 format is obtained by calculating the arithmetic mean of the above R channel, G channel and B channel regions. The grayscale image feature matrix in 32*32 matrix is ​​obtained by taking the arithmetic mean of the middle 8*8 region. 5) Convert the image feature matrix into a 64-bit list and calculate the average value; compare the 64-bit list with the average value, assign 0 if it is less than the average value, and assign 1 if it is greater than the average value to obtain the binary sequence of the image mean value; 6) Loop through the binary sequence of the image mean, using three numbers as the step size, and convert it to decimal to obtain the decimal sequence of the image mean; pad the tens digit of any number less than 10 in the middle of the decimal sequence of the image mean with zeros to obtain the image hash value; 7) Let dictionary A contain the key and value of the image to be checked for duplication, and let dictionary B contain the key and value of existing images. The image to be checked and the existing image value contain the hash value of the image to be checked and the original path of the image to be checked and the existing image. The key of the image to be checked and the existing image key are marked respectively. Compare dictionary A and dictionary B. Let the sample of the image value to be checked be Ax, and the sample of the existing image value be B(0-n). If Ax is less than B0, the image to be checked and the existing image are not duplicated. If Ax is greater than Bn, the image to be checked and the existing image are not duplicated. If Ax is in the interval of B(0-n), compare it with the middle value of B. Use binary search to narrow down the range and determine whether they are completely equal. If they are completely equal, output the key of the image to be checked and the existing image key. Otherwise, the image to be checked and the existing image are not duplicated.

[0025] Specifically, according to the inventor's experiments, step 4) follows a trend: most of the RGB three-channel information is concentrated in the low-frequency region at the upper left. Based on this trend, the following 8*8 regions are selected: the upper left 8*8 region of the R channel, the lower right 8*8 region of the R channel, the lower left 8*8 region of the G channel, the upper right 8*8 region of the B channel, and the middle 8*8 region of each RGB channel (the grayscale value is selected once in the middle). An 8*8 matrix is ​​obtained by calculating their arithmetic mean.

[0026] It also includes a database for an image deduplication method based on Python, comprising storage space, wherein the storage space is distributed to store the image deduplication method and information based on Python as described above.

[0027] It also includes an image plagiarism detection device, comprising a memory and a processor; the memory is used to store a computer program; the processor is used to implement, when executing the computer program, a Python-based image plagiarism detection method as described above.

[0028] It also includes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the Python-based image deduplication method described above.

[0029] It also includes a Python-based image plagiarism detection system, comprising the image plagiarism detection device, internet interface, and peripherals as described above. The peripherals include a monitor, keyboard, and mouse. The image plagiarism detection device is connected to the internet interface and peripherals in sequence. The monitor is used to display the Python-based image plagiarism detection method. The keyboard and mouse are used to control and modify the Python-based image plagiarism detection method. Example 1 Run the above-mentioned Python-based image plagiarism detection system; refer to Figure 2This embodiment uses two batches of samples, each containing 10,000 data points, both randomly selected and named in different formats: "APPLE (1-10440) / BANANA (1-10016)". Drag the two target folders to their corresponding positions in the image; the text boxes below will display the original addresses of the two folders. (Note: To check for duplicate images within a folder, check "Single Folder Matching," then click "File Comparison" to display the paths of duplicate images within the same folder; looping compares all images from two folders for duplicate checking.) Click "File Comparison" to start the process. The tool will print the names of duplicate data in the text boxes below and finally display the number of duplicates and the time taken. In this demonstration, a total of 8936 duplicate images were found, with an average time of 8.3480 ms.

[0030] We can directly use the resize function built into the OpenCV library to perform a scaling operation on the image.

[0031] img = cv2.resize(img, (32, 32), interpolation=cv2.INTER_CUBIC) Interpolation: This specifies the interpolation method. After image scaling, pixels are recalculated, and this parameter specifies how the pixels are recalculated. INTER_CUBIC: Bicubic interpolation within a 4x4 pixel neighborhood. Only the scaling function is used; no formula is involved.

[0032] The image is reduced to a value of (32, 32, 3), where 32 represents pixels and 3 represents the RGB channels. Since DCT calculations use a two-dimensional matrix, directly inputting a three-dimensional image can easily lead to errors. Therefore, DCT is performed separately for each of the RGB dimensions, and the final results are combined into a single three-dimensional matrix and saved as a color image. For grayscale images, only one DCT operation is required. The DCT formula is shown above.

[0033] like Figure 3 The image shown contains pixel values ​​for a 32*32 pixel RGB three-channel image (some values ​​omitted). like Figure 4 The image shows a 32x32 three-dimensional matrix (partially omitted) resulting from the combination of three DCTs. like Figure 5-8 The new pixel table shown is sampled, and here we show 2, 11, 19, and 26.

[0034] As shown in the above figure, the trend follows a rule that most of the information of the three RGB channels is concentrated in the low-frequency region on the upper left side. Based on this rule, the following regions are selected: the 8*8 region at the upper left of the R channel, the 8*8 region at the lower right of the R channel, the 8*8 region at the lower left of the G channel, the 8*8 region at the upper right of the B channel, and the respective middle 8*8 regions of RGB (the grayscale value is taken once in the middle) As Figure 9 shown, an 8*8 matrix is obtained by calculating its arithmetic mean.

[0035] As Figure 10 shown, it is converted into a 64-bit list.

[0036] As Figure 11 shown, its average value is calculated.

[0037] As Figure 12 shown, the 64 bits are compared with the average value, where a value less than the average is assigned 0 and a value greater than the average is assigned 1 (binarization).

[0038] As Figure 13-14 shown, the sequence is cycled with a step size of 3 numbers, converted from binary to decimal, and a zero is added to the tens place for numbers less than 10 in the middle to obtain the hash value of a single image.

[0039] Two dictionaries are established through the above steps, where dictionary A contains the keys of images to be checked for duplication and the values of images to be checked for duplication, and dictionary B contains the keys of existing images and the values of existing images; the keys of the images to be checked for duplication and the keys of the existing images are marked respectively; dictionary A and dictionary B are compared, where the value sample of the image to be checked for duplication is set as Ax, and the value sample of the existing images is set as B(0-n).

[0040] The "values" of dictionary A and dictionary B include different hash values and original paths of all images in the file, and the "keys" include sequence marks; duplicate "keys" will mark two different original paths of images in the corresponding "values", as Figure 2 described, and are printed out at the same time (for self-checking of duplicate images in a single folder). The "keys" of the dictionary are arranged according to their numerical values, and for each "key" in the first dictionary, a binary search is performed in the second dictionary for comparison.

[0041] The first dictionary is set as A, Ax is the image to be checked for duplication each time, the second dictionary is set as B, and the "keys" are B(0-n).

[0042] If Ax < B0, there is no duplication; If Ax > Bn, there is no duplication; If Ax is within the interval of B(0-n), it is compared with the middle value of B each time to narrow the scope, so as to judge whether they are completely equal. If yes, the "value" of the duplicate image, that is, the original path it contains, is output.

[0043] Color images of the relevant steps would be more intuitive, but according to the detailed requirements, no examples will be provided here. Instead, the images will be provided in the form of partial grayscale images.

[0044] In summary, this invention uses Direct Current Transformation (DCT) to acquire low-frequency information, transforming the image from the pixel domain to the frequency domain. Images generally contain a lot of redundancy and correlation; after transformation to the frequency domain, only a small portion of the frequency components have coefficients that are not zero, while most coefficients are zero (or close to zero). The coefficient matrix after DCT transformation shows increasingly higher frequencies from the top left to the bottom right corner. Therefore, the image's energy is mainly retained in the low-frequency coefficients in the top left corner. The method provided by this invention retains more information and has higher fidelity.

[0045] The above description is only a preferred embodiment of the present invention and is not intended to limit the design of this case. All equivalent changes made based on the key design features of this case shall fall within the protection scope of this case.

Claims

1. A Python-based image deduplication method, characterized in that, Including the following steps: 1) Obtain multiple images to be checked for plagiarism and existing color images; 2) The color image is scaled using the resize function built into the OpenCV library; the scaling results in a 32*32 pixel RGB three-channel image. 3) Perform DCT operations on the image according to each dimension of the RGB three channels and combine them to obtain a 32*32 three-dimensional matrix of the image, and save it as a color image; 4) In the 32*32 three-dimensional matrix of the image, the R channel takes the upper left 8*8, middle 8*8, and lower right 8*8 regions, and the G... The R channel takes the middle 8*8 and lower left 8*8 regions, and the B channel takes the middle 8*8 and upper right 8*8 regions. The image feature matrix in 8*8 format is obtained by calculating the arithmetic mean of the above R, G and B channel regions. 5) Convert the image feature matrix into a 64-bit list and calculate the average; compare the 64-bit list with the average. If the value is less than the average, assign 0; if the value is greater than the average, assign 1 to obtain the binary sequence of the image mean. Then, loop through the binary sequence of the image mean, using three numbers as the step size, and convert it to decimal to obtain the decimal sequence of the image mean. 6) Pad the tens digit of any number less than 10 in the middle of the decimal sequence of the image's mean with zeros to obtain the duplicate to be checked and the existing color. Color image hash value; 7) Let dictionary A contain the key and value of the color image to be checked for duplicates, and let dictionary B contain the existing color images... Image key and existing color image value; the color image value to be checked and the existing color image value include the hash value of the color image to be checked and the original path of the color image to be checked and the existing color image; the color image key to be checked and the existing color image key are respectively marked with a sequence. For each "key" in dictionary A, a binary search is performed in dictionary B; comparing dictionary A and dictionary B, let the sample of the color image value to be checked be Ax, and let the sample of the existing color image value be B(0-n); if Ax is less than B0, the color image to be checked and the existing color image are not repeated; if Ax is greater than Bn, the color image to be checked and the existing color image are not repeated; if Ax is in the interval of B(0-n), it is compared with the middle value of B, and the range is narrowed by binary search to determine whether they are completely equal. If they are completely equal, the color image key to be checked and the existing color image key are output; otherwise, the color image to be checked and the existing color image are not repeated.

2. A Python-based image plagiarism detection method, characterized in that, Including the following steps: 1) Obtain multiple images to be checked for plagiarism and existing grayscale images; 2) The grayscale image is scaled using the resize function built into the OpenCV library; the scaling results in a 32*32 pixel grayscale image. 3) The grayscale image is combined after performing a single-channel DCT operation on the image to obtain the 32*32 image matrix; 4) In the 8x8 region in the middle of the image matrix, obtain an 8x8 format image by taking the arithmetic mean of the 8x8 region. Like the feature matrix; 5) Convert the image feature matrix into a 64-bit list and calculate the average; compare the 64-bit list with the average value. If the value is less than the average, assign 0; if the value is greater than the average, assign 1 to obtain the binary sequence of the image mean. Then, loop through the binary sequence of the image mean, using three numbers as the step size, and convert it to decimal to obtain the decimal sequence of the image mean. 6) Pad the tens digit of any number less than 10 in the middle of the decimal sequence of the image's mean with zeros to obtain the duplicates to be checked and the existing gray numbers. Image hash value; 7) Let dictionary A contain the key and value of the grayscale image to be checked for duplicates, and let dictionary B contain the existing grayscale images. The image key and existing grayscale image value; the grayscale image value to be checked and the existing grayscale image value include the hash value of the grayscale image to be checked and the original path of the grayscale image to be checked and the existing grayscale image; the grayscale image key to be checked and the existing grayscale image key are respectively sequence-marked, and for each "key" in dictionary A, a binary search is performed in dictionary B; comparing dictionary A and dictionary B, let the sample of the grayscale image value to be checked be Ax, and let the sample of the existing grayscale image value be B(0-n); if Ax is less than B0, the grayscale image to be checked and the existing grayscale image are not duplicated; if Ax is greater than Bn, the grayscale image to be checked and the existing grayscale image are not duplicated; if Ax is in the interval of B(0-n), it is compared with the middle value of B, and the range is narrowed by binary search to determine whether they are completely equal. If they are completely equal, the grayscale image key to be checked and the existing grayscale image key are output; otherwise, the grayscale image to be checked and the existing grayscale image are not duplicated.

3. A database for an image plagiarism detection method based on Python, characterized in that... It includes storage space, which is used to distribute the image deduplication method and information based on Python as described in claim 1.

4. A database for an image plagiarism detection method based on Python, characterized in that, It includes storage space, which is distributed to store the image deduplication method and information based on Python as described in claim 2.

5. An image plagiarism detection device, characterized in that, It includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement, when executing the computer program, a Python-based image deduplication method as described in claim 1.

6. An image plagiarism detection device, characterized in that, It includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement, when executing the computer program, a Python-based image deduplication method as described in claim 2.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the image deduplication method based on Python as described in claim 1.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the Python-based image deduplication method as described in claim 2.

9. An image plagiarism detection system, characterized in that, Includes the image plagiarism detection device as described in claim 5, an internet interface, and peripherals, wherein the peripherals include a display, a keyboard, and a mouse. The image plagiarism detection device is connected to the Internet interface and peripheral devices in sequence; The display is used to show the image deduplication method based on Python as described in claim 1; The keyboard and mouse are used to control and modify the image deduplication method based on Python as described in claim 1; The Internet interface is used to connect the image deduplication method based on Python as described in claim 1 to the Internet.

10. An image plagiarism detection system, characterized in that, Includes the image plagiarism detection device as described in claim 6, an internet interface, and peripherals, wherein the peripherals include a display, a keyboard, and a mouse. The image plagiarism detection device is connected to the Internet interface and peripheral devices in sequence; The display is used to show the image deduplication method based on Python as described in claim 2; The keyboard and mouse are used to operate and modify the image deduplication method based on Python as described in claim 2; The internet interface is used to connect the image deduplication method based on Python as described in claim 2 to the network.

Images