Sensitive information masking system for archive digitization image optimization integrated with image recognition

Abstract

The present invention discloses an archive digitization image optimization sensitive information masking system that integrates image recognition, which relates to the technical field of image sensitive information masking, including an image masking analysis module, a masking effect analysis module and a final masking effect evaluation module. Sensitive information is accurately identified through OCR technology and cosine similarity matching, and a masking strategy is dynamically selected based on the area ratio of sensitive areas: a chaotic mapping algorithm is used for global masking when dense, and a fine masking matrix is ​​generated through grid division when sparse, so as to retain non-sensitive information to the maximum extent; redundant areas are located based on a masking matrix accumulation algorithm, and combined with sensitive type priority and attribution weight analysis, not only the image quality degradation caused by overlapping of multiple areas is solved; it also solves the problems that traditional methods cannot dynamically select masking methods according to the distribution of sensitive areas and lack normalized decision rules when multiple masking matrices conflict, thereby improving the automation efficiency and masking accuracy of archive digitization processing.

Description

Technical Field

[0001] The present invention belongs to the technical field of image sensitive information masking, and specifically relates to an archive digitization image optimization sensitive information masking system integrated with image recognition. Background Art

[0002] In the process of archive digitization, although sensitive information masking technology has achieved basic functions through OCR recognition and rule matching, its core flaws limit its practical application effectiveness: existing technologies use static masking strategies and cannot dynamically select global masking (such as chaotic mapping) or local fine masking based on the area ratio of sensitive areas. This leads to an imbalance between masking efficiency and integrity. When sensitive areas are densely populated, global masking causes the loss of non-sensitive information, while local masking is computationally complex and difficult to cope with large-scale processing requirements.

[0003] At the same time, multiple masking matrices lack a quantitative analysis and correction mechanism for repeated masking of the same area. Redundant operations cause image structure damage (such as mosaic granulation and excessive blurring), seriously reducing the readability of the archive. More importantly, the existing scheme does not establish a priority rule for sensitive types (such as ID card number > bank card number > IP address), and uses a homogenized masking strategy for information of different protection levels, resulting in the risk of insufficient protection for high-priority sensitive information.

[0004] Furthermore, masking effectiveness assessment relies on manual spot checks or a single coverage metric, failing to achieve closed-loop optimization through a multidimensional quantitative model of masking integrity, nor supporting iterative tuning of masking strategies. These technical bottlenecks severely restrict the efficiency of data security governance during the archive digitization process, necessitating systematic breakthroughs through dynamic masking strategy adaptation, redundant conflict resolution, and multidimensional evaluation mechanisms. Based on this, a sensitive information masking system for archive digitization image optimization, integrated with image recognition, is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide an archival digitized image optimization sensitive information masking system that integrates image recognition, which solves the technical problems of being unable to dynamically select a masking method based on the distribution of sensitive areas and lacking a normalized decision rule when multiple masking matrices conflict.

[0006] The image masking analysis module is used to identify sensitive information and divide regions of each digital image of the archives. Through data analysis, the location distribution and area ratio of the corresponding sensitive regions of each digital image of the archives are obtained. A masking method for sensitive information of each digital image of the archives is selected, and the image to be masked is obtained through masking operation to obtain the initial fine masking effect. The masking method includes: full masking based on chaos mapping algorithm and fine masking through fine masking analysis.

[0007] The shielding effect analysis module includes a shielding matrix analysis unit and a shielding matrix correction unit, wherein: the shielding matrix analysis unit is used to perform shielding redundancy analysis and redundancy correction on the initial fine shielding effect of each image to be shielded;

[0008] The mask matrix correction unit is used to perform correlation analysis between the masked redundant areas corresponding to each to-be-masked image and each standard mask matrix, and perform correction judgment based on the attribution degree of each redundant area and the priority of each standard mask matrix;

[0009] The final shielding effect evaluation module is used to perform secondary shielding based on the correction judgment result to obtain the secondary shielding result, design the iteration termination condition based on the shielding integrity and the structural similarity of the shielding image, and form the redundant shielding iterative analysis according to the iterative analysis condition.

[0010] As a further solution of the present invention: sensitive information is identified on each digital image of the archives to obtain each sensitive information, including: S101: Based on each digital image of the archives collected by the system, the digital image of the archives is collected based on the OCR technology. As input, a single archival digitized image Perform OCR recognition and extraction to obtain a text information set ;in, represents the kth archival digitized image in the archival digitized image set, and , represents the text information set, k represents the digital image identifier of each archive and , m represents the number of text information;

[0011] Text messages and preset sensitive information database Perform cosine similarity matching calculation on text information to obtain digital images of archives Sensitive matching degree of corresponding text information ;in, Representing digital images of archives The highest sensitive matching degree of the i-th text information in Representing digital images of archives The i-th text information extracted by OCR in , represents the jth sensitive information in the preset sensitive information database, and a represents the number of sensitive information contained in the preset sensitive information database;

[0012] S102: Digitalize archive images Sensitive matching degree of corresponding text information Matches the preset threshold Make comparisons;

[0013] When the text information corresponding to the digital image of the archive Sensitive matching Greater than the preset matching threshold When , the corresponding text information is marked as sensitive information, and its corresponding coordinate information is recorded; otherwise, it is ordinary information and no processing is done;

[0014] S103: Repeat the above steps S101 to S102 to obtain the sensitive information corresponding to each digital image of the archive, and The text information marked as sensitive information is associated with the coordinate information extracted by OCR to generate a sensitive area information set covering sensitive information , where p represents the number of sensitive areas, h represents the index of each sensitive area and , for each sensitive area Contains the following content: sensitive area coordinate frame range ), sensitive type label, sensitive matching .

[0015] As a further solution of the present invention: the method for shielding sensitive information is selected, comprising the following steps: for each sensitive area Calculate its area ratio ;in, Indicates the area proportion corresponding to each sensitive area h, Respectively represent the area coordinates corresponding to each sensitive area h, Indicates the total image area corresponding to the digital image of the archive where each sensitive area is located;

[0016] For each sensitive area The total area ratio of all sensitive areas corresponding to each digital image of the archive is obtained by traversing and summing , according to the formula:

[0017] ; Where, p represents the number of sensitive areas;

[0018] The total area ratio of all sensitive areas corresponding to the digital images of each archive Determine the threshold value with the preset masking method Make comparisons;

[0019] when When , the whole image is masked by using the chaos mapping algorithm;

[0020] when When the corresponding digital archive image is marked as the image to be masked , fine masking is performed using fine masking analysis; where u is the identifier of the image to be masked.

[0021] As a further solution of the present invention: Sensitive information matrix construction process: Obtaining the initial fine shielding effect through shielding operation, including the following steps: Obtaining each image to be shielded through traversal analysis , and divide all the images to be masked into a H×W grid according to the preset width grid_width of each grid unit and the preset height grid_height of each grid unit, where H is the number of rows and W is the number of columns; and represents the grid cell row index of the matrix, , represents the grid cell column index of the matrix;

[0022] Collect all sensitive areas corresponding to the images to be masked Coordinate frame range );in, Identifies the sensitive area corresponding to the image to be masked and , Indicates the number of sensitive areas corresponding to the image to be masked;

[0023] The corresponding sensitive areas Coordinate frame range ), mapped to the grid to get the column range and row range of the corresponding grid ), specifically:

[0024] and Corresponding grid row range: , ;

[0025] and Corresponding grid column range: , ;

[0026] Combine column range and row range ), fill in the H×W matrix to get the sensitive area information matrix ;

[0027] Repeat the sensitive information matrix construction process to obtain each fine masking image Corresponding A sensitive information matrix;

[0028] Generate the corresponding sensitive information matrix of the same structure according to each image to be masked Standard masking matrix , and perform fine masking operations on each image to be masked to obtain the initial fine masking effect.

[0029] As a further solution of the present invention: in the mask matrix analysis unit, the mask redundancy analysis is performed on the initial fine masking effect of each image to be masked, including: collecting each image to be masked The corresponding standard shielding matrices are obtained to obtain the standard shielding matrix set ;in, represents the number of standard masking matrices;

[0030] Accumulate all standard masking matrices to generate the total masking matrix corresponding to the image to be masked , according to the formula: ;in, Characterizing Grid Cells The amount covered by the standard masking matrix;

[0031] like , then the grid cell The location of the area marked as masked redundant.

[0032] As a further solution of the present invention: in the masking matrix analysis unit, the redundancy correction process includes: traversing the total masking matrix of the image to be masked , record all The location of the shielding redundant area , forming a set of masked redundant regions According to the formula:

[0033] ;in, Represents the set of redundant regions corresponding to each image to be masked, It is an indicator function, the value is 1 when the condition is met, otherwise it is 0;

[0034] when 0, indicating that there is a masked redundant area, which needs to be corrected and enters the masking matrix correction unit;

[0035] when 0, no redundancy in representation, directly output the first fine masking effect image As the final masking effect image corresponding to the image to be masked.

[0036] As a further solution of the present invention: in the mask matrix correction unit, the mask redundant area corresponding to each to-be-masked image is subjected to correlation analysis with each standard mask matrix, including: , for each masked redundant region position , calculate the standard shielding matrix of its attribution The degree of belonging is based on the formula:

[0037] ;in, Represents the standard masking matrix In the grid cell the proportion of responsibility in

[0038] Based on the pre-set priority table of sensitive types and the standard masking matrix based on their attribution The corresponding sensitive information types are assigned priorities; the preset sensitive type priority table includes priority sorting information of various types of sensitive information from high to low.

[0039] As a further solution of the present invention: in the masking matrix correction unit, the correction judgment is made according to the attribution degree of each redundant area and the priority of each standard masking matrix, including: the standard masking matrix to which each masked redundant area belongs obtained according to the correlation analysis The redundancy correction judgment function is defined by combining the degree of belonging and its corresponding priority. The function objective of the redundancy correction judgment function is:

[0040] For masking redundant area locations , determine the masking matrix marks that need to be retained and delete other redundant marks; including: based on the priority corresponding to each standard masking matrix, filter out the standard masking matrix set with the highest priority ;in, To shield A set of standard masking matrices;

[0041] exist In the example, the standard masking matrix with the largest degree of attribution is selected and recorded as the retained standard masking matrix ;

[0042] When there are multiple If they have the same degree of belonging, the standard masking matrix with the smallest index is selected;

[0043] To retain the standard masking matrix , retaining its position in the masked redundant area The corresponding matrix area;

[0044] For all other standard masking matrices marked as standard masking matrices to be deleted, obtain the position of the standard masking matrix to be deleted in the masking redundant area. The matrix area information of each standard masking matrix to be deleted is placed in the masking redundant area The matrix area information is set to zero.

[0045] As a further solution of the present invention: according to the judgment result of the redundancy correction judgment function in the shielding matrix correction unit, the standard shielding matrix set Correction is performed, and based on the corrected standard masking matrix set, each image to be masked is masked to obtain a secondary fine masking effect;

[0046] The masking area of ​​all standard masking matrices on their corresponding sensitive areas is marked as masking integrity ZB.

[0047] And based on the secondary masking result corresponding image and the image to be masked Structural similarity ;

[0048] The mask integrity ZB corresponding to the secondary masking result and the image corresponding to the secondary masking result and the image to be masked are calculated. The corresponding preset threshold and Make comparisons;

[0049] when and When , the iterative correction stops, otherwise it returns to the shielding matrix correction unit for iterative analysis.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] (1) This invention dynamically selects a masking method based on the proportion of sensitive areas: when the sensitive areas are dense, a chaotic mapping algorithm is used for rapid global masking; when the sensitive areas are sparse, a fine masking matrix is ​​generated through grid division to maximize the retention of non-sensitive area information. At the same time, redundant areas are cumulatively located based on the masking matrix, and repeated masking is intelligently corrected by combining priority and attribution, avoiding the image quality degradation caused by multiple area overlap in traditional methods.

[0052] (2) This invention introduces a dual-index evaluation system of masking integrity and structural similarity, and uses a threshold to control the iterative termination condition to ensure a balance between privacy protection and image usability. Furthermore, it uses OCR technology and cosine similarity matching to accurately identify sensitive information. This modular architecture (masking analysis → effect correction → evaluation feedback) allows for flexible expansion, overcoming the technical bottleneck of traditional methods that rely on manual judgment and have poor semantic adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 Schematic diagram of the system framework structure of the present invention;

[0054] Figure 2 Schematic diagram of the logic implementation of the image shading analysis module of the present invention;

[0055] Figure 3 Schematic diagram of the logic implementation of the shielding matrix analysis unit of the present invention;

[0056] Figure 4 Schematic diagram of the logic implementation of the shielding matrix correction unit of the present invention. DETAILED DESCRIPTION

[0057] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0058] Example 1: Please refer to Figure 1 The present application provides a sensitive information masking system for optimized archival digitized images integrated with image recognition, comprising: an image masking analysis module, a masking effect analysis module, and a final masking effect evaluation module; the masking effect analysis module comprises a masking matrix analysis unit and a masking matrix correction unit; specifically:

[0059] in accordance with Figure 2 The image masking analysis module is used to identify sensitive information of each digital image of the archives to obtain each sensitive information, obtain the position distribution and area ratio of the corresponding sensitive areas of each digital image of the archives through data analysis, select a masking method for the sensitive information of each digital image of the archives, and obtain the image to be masked through the masking operation to obtain the initial fine masking effect;

[0060] The masking method includes: performing whole-image masking based on a chaotic mapping algorithm and performing fine masking through fine masking analysis;

[0061] The sensitive information identification and regional division of each digital image of the archives is specifically as follows:

[0062] S101: Based on the digital images of each archive collected by the system, the digital images of the archive are collected and combined based on the OCR technology. As input, a single archival digitized image Perform OCR recognition and extraction to obtain a text information set ;in, represents the kth archival digitized image in the archival digitized image set, and , represents the text information set, k represents the digital image identifier of each archive and , m represents the number of text messages; and preset sensitive information database Perform cosine similarity matching calculation on text information to obtain digital images of archives Sensitive matching degree of corresponding text information ;in, Representing digital images of archives The highest sensitive matching degree of the i-th text information in Representing digital images of archives The i-th text information extracted by OCR in , represents the jth sensitive information in the preset sensitive information database, and a represents the number of sensitive information contained in the preset sensitive information database;

[0063] S102: Digital images based on archives Sensitive matching degree of corresponding text information Matches the preset threshold Make comparisons;

[0064] When the text information corresponding to the digital image of the archive Sensitive matching Greater than the preset matching threshold When , the corresponding text information is marked as sensitive information, and its corresponding coordinate information is recorded; otherwise, it is ordinary information and no processing is done;

[0065] S103: Repeat the above steps S101 to S102 to obtain the sensitive information corresponding to each digital image of the archive; including:

[0066] Digitize images of various archives The text information marked as sensitive information and its coordinate information extracted by OCR include the starting point ( ) and the end point ( ) to associate; merge adjacent sensitive text information regions through the bounding box fitting algorithm to generate a sensitive region information set covering sensitive information , where p represents the number of sensitive areas, h represents the index of each sensitive area and For each sensitive area Contains the following:

[0067] Sensitive area coordinate frame range ); represents the pixel range of the sensitive area in the image;

[0068] Sensitive type labels (such as ID number, IP address, etc.), sensitive matching ;

[0069] The preset sensitive information database of the embodiment of the present disclosure may include key IP addresses, personnel ID numbers, personnel mobile phone numbers, unit names, personnel names, etc. The corresponding text information is matched with the preset sensitive information database, and the sensitive information detection results corresponding to the digital images of each archive are obtained;

[0070] S104: Sensitive area data analysis and masking method selection; including:

[0071] For each sensitive area Calculate its area ratio ;in, Indicates the area proportion corresponding to each sensitive area h, Respectively represent the area coordinates corresponding to each sensitive area h, Indicates the total image area corresponding to the digital image of each sensitive area, which can be directly obtained through OCR technology; The total area ratio of all sensitive areas corresponding to each digital image of the archive is obtained by traversing and summing , according to the formula:

[0072] ;

[0073] The total area ratio of all sensitive areas corresponding to the digital images of each archive Determine the threshold value with the preset masking method Make comparisons;

[0074] when When the sensitive information is considered to be widely distributed, the whole image is masked using a chaos mapping algorithm;

[0075] when When the sensitive information is considered to be concentrated locally, the corresponding digital image of the archive is marked as an image to be masked. , perform fine masking through fine masking analysis; where u is the identifier of the image to be masked;

[0076] S105: Constructing a sensitive information matrix corresponding to the image to be masked; including:

[0077] Obtain each image to be shielded by traversal analysis , and divide all the images to be masked into a H×W grid according to the preset width grid_width of each grid unit and the preset height grid_height of each grid unit, where H is the number of rows and W is the number of columns; and Represents the row index of the matrix, corresponding to the grid cell index of the image in the height direction, , represents the column index of the matrix, corresponding to the grid cell index of the image in the width direction;

[0078] Collect all sensitive areas corresponding to the images to be masked Coordinate frame range );in, is the sensitive area identifier corresponding to the image to be masked, and , Indicates the number of sensitive areas corresponding to the image to be masked;

[0079] The corresponding sensitive areas Coordinate frame range ), mapped to the grid to get the column range and row range of the corresponding grid ), specifically:

[0080] and Corresponding grid row range: , ;

[0081] and Corresponding grid column range: , ;

[0082] Combine column range and row range ), fill in the H×W matrix to get the sensitive area information matrix ;

[0083] Repeat step S105 to obtain each image to be shielded Corresponding A sensitive information matrix;

[0084] Generate the corresponding sensitive information matrix of the same structure according to each image to be masked Standard masking matrix , and perform fine masking operation on each image to be masked to obtain the initial fine masking effect;

[0085] It should be noted that the fine masking operation is performed on each image to be masked to obtain the initial fine masking effect. The masking matrix is ​​applied to the corresponding image to be masked, and each standard masking matrix is ​​matched with its corresponding sensitive information, thereby masking the sensitive area and obtaining the initial fine masking effect. .

[0086] The shielding effect analysis module includes a shielding matrix analysis unit and a shielding matrix correction unit;

[0087] in accordance with Figure 3 The masking matrix analysis unit is used to perform masking redundancy analysis on the initial fine masking effect of each image to be masked, and determine whether it is necessary to perform redundancy correction on each standard masking matrix corresponding to each image to be masked based on the masking redundancy analysis result; if not, directly output the secondary masking result; if necessary, enter the masking matrix correction unit to perform redundancy correction analysis on the standard masking matrix; specifically:

[0088] The redundancy is that there are two or more standard shielding matrices shielding the same area;

[0089] Collect each image to be masked The corresponding standard shielding matrices are obtained to obtain the standard shielding matrix set ;in, represents the number of standard masking matrices;

[0090] Accumulate all standard masking matrices to generate the total masking matrix corresponding to the image to be masked , according to the formula: ;in, Characterizing Grid Cells The amount covered by the standard masking matrix;

[0091] like , then the grid cell The location of the area marked as masked redundant;

[0092] Traverse the total masking matrix of the image to be masked , record all The location of the shielding redundant area , forming a set of masked redundant regions , according to the formula:

[0093] ;in, Represents the set of redundant regions corresponding to each image to be masked, It is an indicator function, the value is 1 when the condition is met, otherwise it is 0;

[0094] when 0, indicating that there is a masked redundant area, which needs to be corrected and enters the masking matrix correction unit;

[0095] when 0, no redundancy in representation, directly output the first fine masking effect image As the final masking effect image corresponding to the image to be masked.

[0096] in accordance with Figure 4 The mask matrix correction unit is used to perform correlation analysis between the masked redundant areas corresponding to each to-be-masked image and each standard mask matrix. The correlation analysis is to determine the degree of belonging of each redundant area in each standard mask matrix, and to establish a redundancy correction judgment function based on the priority of each standard mask matrix to perform correction judgment.

[0097] According to the masked redundant area collection , for each masked redundant region position , calculate the standard shielding matrix of its attribution The degree of belonging is based on the formula:

[0098] ;in, Represents the standard masking matrix In the grid cell the proportion of responsibility in

[0099] A standard masking matrix for its attribution based on a preset priority table of sensitive types Perform priority assignment, that is, assign the preset sensitive type priority to the corresponding sensitive information type in the preset sensitive type priority table;

[0100] It should be noted that the preset sensitive type priority can be directly obtained through a preset sensitive type priority table. For example, in this embodiment, the preset sensitive type priority table may be identity card number>bank card number>IP address, etc.;

[0101] The standard masking matrix of each masked redundant area obtained by correlation analysis The redundancy correction judgment function is defined by combining the degree of belonging and its corresponding priority. The function objective of the redundancy correction judgment function is:

[0102] For masking redundant area locations , determine the masking matrix marks that need to be retained and delete other redundant marks; including: based on the priority corresponding to each standard masking matrix, filter out the standard masking matrix set with the highest priority ;in, To shield A set of standard masking matrices;

[0103] exist In the example, the standard masking matrix with the largest degree of attribution is selected and recorded as the retained standard masking matrix ;

[0104] in, Indicates the location of the masked redundant area The set of all standard masking matrices;

[0105] When there are multiple With the same degree of belonging, the standard masking matrix with the smallest index is selected as the retained standard masking matrix, for example Takes precedence over ;

[0106] To retain the standard masking matrix , retaining its position in the masked redundant area The corresponding matrix area;

[0107] For all other standard masking matrices marked as standard masking matrices to be deleted, obtain the position of the standard masking matrix to be deleted in the masking redundant area. The matrix area information of each standard masking matrix to be deleted is placed in the masking redundant area The matrix area information is set to zero.

[0108] The final shielding effect evaluation module is used to perform secondary shielding based on the correction judgment result to obtain the secondary shielding result, design the iteration termination condition based on the shielding integrity and the structural similarity of the shielding image, and form a redundant shielding iterative analysis according to the iterative analysis condition;

[0109] According to the judgment result of the redundancy correction judgment function in the masking matrix correction unit, the standard masking matrix set Correction is performed, and based on the corrected standard masking matrix set, each image to be masked is masked to obtain a secondary fine masking effect;

[0110] Obtain the shielding integrity ZB corresponding to the secondary shielding result, that is, count the shielding areas of all standard shielding matrices of the secondary shielding result on the corresponding sensitive areas and mark them as shielding integrity ZB;

[0111] Based on the secondary masking result, the image corresponding to the image to be masked (Structural Similarity Index);

[0112] The mask integrity ZB corresponding to the secondary masking result and the image corresponding to the secondary masking result and the image to be masked are calculated. The corresponding preset threshold and Make comparisons;

[0113] when and When , the iterative correction stops, otherwise it returns to the masking matrix correction unit for the next round of iteration; wherein, the preset threshold and The specific value is set by professional researchers based on actual needs.

[0114] In this example, the masking method is dynamically selected based on the proportion of sensitive areas: when the sensitive areas are dense, a chaotic mapping algorithm is used for rapid global masking; when the sensitive areas are sparse, a fine masking matrix is ​​generated through grid division to maximize the retention of non-sensitive area information. At the same time, redundant areas are located based on the cumulative masking matrix, and repeated masking is intelligently corrected by combining priority and attribution, avoiding the image quality degradation caused by multiple area overlaps in traditional methods. By introducing a dual-index evaluation system of masking integrity and structural similarity, the iterative termination condition is controlled by threshold to ensure a balance between privacy protection and image availability. In addition, OCR technology and cosine similarity matching are used to accurately identify sensitive information, and a modular architecture is combined to achieve flexible expansion. This solves the problems of traditional methods that cannot dynamically select masking methods based on the distribution of sensitive areas and lack normalized decision rules when multiple masking matrices conflict, thereby improving the automation efficiency and masking accuracy of archive digitization processing.

[0115] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0116] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A sensitive information masking system for digital archive images optimized by integrating image recognition, characterized by: It includes image shading analysis module, shading effect analysis module and final shading effect evaluation module, among which: The image masking analysis module is used to identify sensitive information and divide regions of each digital image of the archives. Through data analysis, the location distribution and area ratio of the corresponding sensitive regions of each digital image of the archives are obtained. A masking method for sensitive information of each digital image of the archives is selected, and the image to be masked is obtained through masking operation to obtain the initial fine masking effect. The masking methods include: full masking based on chaos mapping algorithm and fine masking through fine masking analysis. The shielding effect analysis module includes a shielding matrix analysis unit and a shielding matrix correction unit, wherein: the shielding matrix analysis unit is used to perform shielding redundancy analysis and redundancy correction on the initial fine shielding effect of each image to be shielded; The mask matrix correction unit is used to perform correlation analysis between the masked redundant areas corresponding to each to-be-masked image and each standard mask matrix, and perform correction judgment based on the attribution degree of each redundant area and the priority of each standard mask matrix; The final shielding effect evaluation module is used to perform secondary shielding based on the correction judgment result to obtain the secondary shielding result, design the iteration termination condition based on the shielding integrity and the structural similarity of the shielding image, and form a redundant shielding iterative analysis according to the iterative analysis condition; The redundancy correction process includes: Traverse the total masking matrix of the image to be masked , record all The location of the shielding redundant area , forming a set of masked redundant regions According to the formula: ;in, represents the set of redundant regions corresponding to each image to be masked, u is the identifier of the image to be masked, H is the number of grid rows, W is the number of grid columns, It is an indicator function, the value is 1 when the condition is met, otherwise it is 0; when 0, indicating that there is a masked redundant area, which needs to be corrected and enters the masking matrix correction unit; when 0, no redundancy in representation, directly output the first fine masking effect image As the final masking effect image corresponding to the image to be masked; The shielding matrix analysis unit includes: a standard shielding matrix according to which each shielding redundant area belongs The redundancy correction judgment function is defined by combining the degree of belonging and its corresponding priority. The function objective of the redundancy correction judgment function is: For masking redundant area locations , filter out the standard masking matrix set with the highest priority ;in, To shield A set of standard masking matrices; exist In the example, the standard masking matrix with the largest degree of attribution is selected and recorded as the retained standard masking matrix ;in, Indicates the location of the masked redundant area The set of all standard masking matrices; When there are multiple When they have the same degree of belonging, the standard masking matrix with the smallest index is selected as the retained standard masking matrix, where b is the index of the retained standard masking matrix; To retain the standard masking matrix , retaining its position in the masked redundant area The corresponding matrix area; Mark all other standard mask matrices as standard mask matrices to be deleted and place them in the mask redundant area. The matrix area information is set to zero.

2. The system for optimizing sensitive information masking of digital archive images by integrating image recognition according to claim 1 is characterized in that: Sensitive information is identified on each digital image of the archives to obtain each sensitive information, including: S101: Based on the digital images of the archives collected by the system, the digital images of the archives are collected based on the OCR technology. As input, a single archival digitized image Perform OCR recognition and extraction to obtain a text information set ;in, represents the kth archival digitized image in the archival digitized image set, and , represents the text information set, k represents the digital image identifier of each archive and , m represents the number of text information; Text messages and preset sensitive information database Perform cosine similarity matching calculation on text information to obtain digital images of archives Sensitive matching degree of corresponding text information ;in, Representing digital images of archives The highest sensitive matching degree of the i-th text information in Representing digital images of archives The i-th text information extracted by OCR in , represents the jth sensitive information in the preset sensitive information database, and a represents the number of sensitive information contained in the preset sensitive information database; S102: Digitalize archive images Sensitive matching degree of corresponding text information Matches the preset threshold Make comparisons; When the text information corresponding to the digital image of the archive Sensitive matching Greater than the preset matching threshold When , the corresponding text information is marked as sensitive information, and its corresponding coordinate information is recorded; otherwise, it is ordinary information and no processing is done; S103: Repeat the above steps S101 to S102 to obtain the sensitive information corresponding to each digital image of the archive, and The text information marked as sensitive information is associated with the coordinate information extracted by OCR to generate a sensitive area information set covering sensitive information , where p represents the number of sensitive areas, h represents the index of each sensitive area and , for each sensitive area Contains the following content: sensitive area coordinate frame range ), sensitive type label, sensitive matching .

3. The system for optimizing sensitive information masking of digital archive images by integrating image recognition according to claim 1 is characterized in that: The selection of masking method for sensitive information includes the following steps: for each sensitive area Calculate its area ratio ;in, Indicates the area proportion corresponding to each sensitive area h, Respectively represent the area coordinates corresponding to each sensitive area h, Indicates the total image area corresponding to the digital image of the archive where each sensitive area is located; For each sensitive area The total area ratio of all sensitive areas corresponding to each digital image of the archive is obtained by traversing and summing , according to the formula: ; Where, p represents the number of sensitive areas; The total area ratio of all sensitive areas corresponding to the digital images of each archive Determine the threshold value with the preset masking method Make comparisons; when When , the whole image is masked by using the chaos mapping algorithm; when When the corresponding digital archive image is marked as the image to be masked , fine masking is performed using fine masking analysis; where u is the identifier of the image to be masked.

4. The system for optimizing sensitive information masking of digital archive images by integrating image recognition according to claim 1 is characterized in that: Sensitive information matrix construction process: The initial fine masking effect is obtained through masking operation, including the following steps: Obtain each image to be masked through traversal analysis , and divide all the images to be masked into a H×W grid according to the preset width grid_width of each grid unit and the preset height grid_height of each grid unit, where H is the number of rows and W is the number of columns; and , represents the grid cell row index of the matrix, , represents the grid cell column index of the matrix; Collect all sensitive areas corresponding to the images to be masked Coordinate frame range );in, Identifies the sensitive area corresponding to the image to be masked and , Indicates the number of sensitive areas corresponding to the image to be masked; The corresponding sensitive areas Coordinate frame range ), mapped to the grid to get the column range and row range of the corresponding grid ), specifically: and Corresponding grid row range: , ; and Corresponding grid column range: , ; Combine column range and row range ), fill in the H×W matrix to get the sensitive area information matrix ; Repeat the sensitive information matrix construction process to obtain each fine masking image Corresponding A sensitive information matrix; Generate the corresponding sensitive information matrix of the same structure according to each image to be masked Standard masking matrix , and perform fine masking operations on each image to be masked to obtain the initial fine masking effect.

5. The system for optimizing sensitive information masking of digital archive images integrated with image recognition according to claim 1 is characterized in that: In the mask matrix analysis unit, the mask redundancy analysis is performed on the initial fine masking effect of each image to be masked, including: collecting each image to be masked The corresponding standard shielding matrices are obtained to obtain the standard shielding matrix set ;in, represents the number of standard masking matrices; Accumulate all standard masking matrices to generate the total masking matrix corresponding to the image to be masked , according to the formula: ;in, Characterizing Grid Cells The amount covered by the standard masking matrix; like , then the grid cell It is marked as the location of the masked redundant area; the 2 represents the number of times the grid cell is covered by multiple standard masking matrices.

6. The system for optimizing sensitive information masking of digital archive images integrated with image recognition according to claim 1 is characterized in that: In the mask matrix correction unit, the mask redundant areas corresponding to each to-be-masked image are analyzed in association with each standard mask matrix, including: , for each masked redundant region position , calculate the standard shielding matrix of its attribution The degree of belonging is based on the formula: ;in, Represents the standard masking matrix In the grid cell the proportion of responsibility in Priority is assigned to the sensitive information type corresponding to the preset sensitive type priority table and based on the standard masking matrix to which it belongs; the preset sensitive type priority table includes priority sorting information of various types of sensitive information from high to low.

7. The system for optimizing sensitive information masking of digital archive images integrated with image recognition according to claim 1 is characterized in that: The redundant shielding iterative analysis is formed according to the iterative analysis conditions, including: according to the judgment result of the redundant correction judgment function in the shielding matrix correction unit, the standard shielding matrix set is Correction is performed, and based on the corrected standard masking matrix set, each image to be masked is masked to obtain a secondary fine masking effect; The masking area of ​​all standard masking matrices on their corresponding sensitive areas is marked as masking integrity ZB. And based on the secondary masking result corresponding image and the image to be masked Structural similarity ; The mask integrity ZB corresponding to the secondary masking result and the image corresponding to the secondary masking result and the image to be masked are calculated. The corresponding preset threshold and Make comparisons; when and When , the iterative correction stops, otherwise it returns to the shielding matrix correction unit for iterative analysis.

Images