Image recognition and text restoration method and device for information security transmission

Abstract

The invention provides an image recognition and text restoration method and device for information security transmission, and the method comprises the steps: dividing text data needing to be transmitted into a plurality of fixed-length data segments, and generating reference points at the periphery of each data segment; the data segments and the reference points are automatically switched and played in sequence, and a video is recorded; safely transmitting the video to an information receiving end, and segmenting the video into single pictures; identifying a reference point in the picture, and extracting an effective data area in the picture according to the reference point; and identifying characters in the effective data region by using a region detection model based on a convolutional neural network, and restoring the effective data region into an original text. According to the invention, the security of information transmission can be improved, and the efficiency and accuracy of information processing can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of information transmission, and in particular relates to an image recognition and text restoration method and device for secure information transmission. Background Art

[0002] In the field of information transmission, especially in environments that require high security, traditional text transmission methods are vulnerable to various forms of attacks and interception, especially to hacker attacks or theft during network transmission. Although existing encryption technologies are widely used in data transmission, encrypted text may still face the risk of being decrypted. For example, traditional encryption algorithms cannot completely prevent data leakage and tampering, and require additional key management and encryption and decryption processes, which increases the complexity and cost of the system. In addition, although the image transmission method used in the prior art improves security to a certain extent, the information in the image needs to be manually identified, which is inefficient and prone to errors. Therefore, the disadvantage of the prior art is that it cannot provide a safe and efficient information transmission method. Summary of the invention

[0003] The present invention provides an image recognition and text restoration method and device for secure information transmission, which improves the security of information transmission and the efficiency and accuracy of information processing.

[0004] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0005] An image recognition and text restoration method for secure information transmission, comprising:

[0006] S1. The information sending end divides the text data to be transmitted into multiple fixed-length data segments, displays the data segments on the WEB page, and generates reference points around each data segment;

[0007] S2, automatically switching to play each data segment and reference point in sequence, and recording the video of all data segments and their reference points through a video recording device;

[0008] S3, securely transmitting the video to an information receiving end, and the information receiving end divides the video into individual pictures by frame;

[0009] S4, the information receiving end identifies the reference point in the picture, and extracts the valid data area in the picture according to the reference point;

[0010] S5, using a region detection model based on a convolutional neural network to detect characters in the valid data region;

[0011] S6. Restore the recognized characters to the original text.

[0012] Furthermore, the method for generating the reference point in step S1 includes:

[0013] S101, determine the reference point position, and define a structure including a data segment and the reference point position;

[0014] S102, designing a reference point style, using a style that is different from the data segment and other page contents;

[0015] S103: After the WEB page generates and displays the data segment, reference points are generated at corresponding positions around the data segment according to the reference point positions and reference point styles.

[0016] Furthermore, step S4 specifically includes:

[0017] S401, converting the image into a grayscale image by grayscaling, and performing binarization;

[0018] S402, calculating the gradient Ix and Iy in the x and y directions, as well as the product IxIy, the second-order derivatives Ixx and Iyy of each pixel in the image; constructing a corner point response function, and calculating the corner point response value R of each pixel;

[0019] R = det(M) - k·trace(M) 2 ;

[0020] Where M is the structure tensor, which consists of IxIy, Ixx, Iyy; k is a constant;

[0021] Set a threshold and regard points with response values ​​R greater than the threshold as candidate corner points;

[0022] S403, for each candidate corner point, check whether there are other candidate corner points in its neighborhood, and calculate the distance to other candidate corner points; based on the distance and angle relationship, identify the corner point that is most likely to belong to the reference point by analyzing the relative position, angle and distance between the corner points;

[0023] S404: Remove redundant corner points from the identified corner points, and extract the content within the range of the corner points as a valid data area.

[0024] Furthermore, in step S5, the region detection model uses a target detection algorithm to identify the character region in the valid data region, and extracts the identified character region; and identifies the characters in the character region through a character recognition model.

[0025] Furthermore, a cyclic redundancy check (CRC) code is added to the data segment, and step S5 performs a CRC check after identifying the characters in the area.

[0026] On the other hand, the present invention also proposes an image recognition and text restoration device for secure information transmission, comprising:

[0027] Splitting module: The information sender splits the text data to be transmitted into multiple fixed-length data segments, displays the data segments on the WEB page, and generates reference points around each data segment;

[0028] Recording module: automatically switches and plays each data segment and reference point in sequence, and records the video played by all data segments and their reference points through the video recording device;

[0029] Segmentation module: securely transmit the video to the information receiving end, and the information receiving end segments the video into individual pictures by frame;

[0030] Extraction module: The information receiving end identifies the reference points in the image and extracts the valid data area in the image according to the reference points;

[0031] Recognition module: using a region detection model based on a convolutional neural network to identify characters in the valid data region;

[0032] Restoration module: restores the recognized characters to the original text.

[0033] Furthermore, the split modules include:

[0034] Reference point location unit: determines the reference point location and defines the structure containing the data segment and the reference point location;

[0035] Reference point style unit: Design the reference point style, using a style that is different from the data segment and other page content;

[0036] Generating unit: after the WEB page generates and displays the data segment, the reference point is generated at the corresponding position around the data segment according to the reference point position and reference point style.

[0037] Furthermore, the extraction module includes:

[0038] Grayscale unit: Use grayscale to convert the image into a grayscale image and perform binarization;

[0039] Candidate corner point unit: calculate the gradient Ix and Iy in the x and y directions of each pixel in the image, as well as the product IxIy, the second-order derivatives Ixx and Iyy; construct the corner point response function and calculate the corner point response value R of each pixel;

[0040] R = det(M) - k·trace(M) 2 ;

[0041] Where M is the structure tensor, which consists of IxIy, Ixx, Iyy; k is a constant;

[0042] Set a threshold and regard points with response values ​​R greater than the threshold as candidate corner points;

[0043] Corner point identification unit: For each candidate corner point, check whether there are other candidate corner points in its neighborhood and calculate the distance to other candidate corner points; based on the distance and angle relationship, identify the corner point that is most likely to belong to the reference point by analyzing the relative position, angle and distance between the corner points;

[0044] Area unit: Remove redundant corner points from the identified corner points and extract the content within the corner point range as the valid data area.

[0045] Furthermore, in the recognition module, the region detection model uses a target detection algorithm to recognize the character region in the valid data region, and extracts the recognized character region; and recognizes the characters in the character region through a character recognition model.

[0046] Furthermore, the splitting module adds a cyclic redundancy check (CRC) check code to the data segment, and the recognition module performs a CRC check after recognizing the characters in the area.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] 1. The present invention improves the security of information transmission and reduces the risk of being attacked and intercepted;

[0049] 2. The present invention can effectively protect the secure transmission of information and avoid the loss and error of information during the transmission process;

[0050] 3. The present invention automates the processing flow and improves the efficiency and accuracy of information processing;

[0051] 4. The text file restored by the present invention has high readability and is easy for users to understand and use. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a flowchart of an embodiment of the present invention. DETAILED DESCRIPTION

[0053] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0054] The present invention is further described below in conjunction with specific embodiments and drawings.

[0055] Embodiment 1:

[0056] The method of this embodiment is as follows Figure 1 As shown, including:

[0057] 1. Split and reference points:

[0058] The scenario of this embodiment is to transmit text data in a high information security area.

[0059] The server in the high information security area splits the text data to be transmitted into multiple fixed-length data segments, displays the data segments in the WEB page, and generates reference points around the data segments;

[0060] The steps to generate a data segment are as follows:

[0061] 1. Data segmentation: First, the data to be transmitted (such as text, binary files, etc.) needs to be converted into hexadecimal, and then segmented according to a certain size (which can be set according to the screen resolution);

[0062] 2. Add redundancy and checksum information: In order to improve the reliability and integrity of data transmission, a cyclic redundancy check (CRC) checksum is added to each data segment;

[0063] 3. Display data in the WEB page: The WEB page reads the data segment with the CRC checksum added and displays it.

[0064] The steps to generate reference points are as follows:

[0065] 1. Determine the reference point position: In this embodiment, the reference point position is designed to be placed at the four corners of the data segment in the WEB page, so that it is easy to find in the subsequent image recognition process, and an HTML structure including the data segment and the four reference point positions is defined.

[0066] 2. Design the reference point style: To facilitate the distinction between the reference point and other image contents during the image recognition process, the reference point in this embodiment is a black rectangular box. Use CSS to set the style for the data segment and the reference point to ensure that the reference point appears in the form of a black rectangular box and surrounds the data segment.

[0067] 3. Generate reference points: After determining the position and style of the reference points, the WEB page will generate reference points at the corresponding positions around the data segment content. The position and size of the reference points can be adjusted as needed to adapt to different data segment sizes and page layouts.

[0068] After the data segments and reference points are generated, the user performs a browsing operation on the WEB page (for example, clicks a browse data button), and the system automatically switches the images of the data segments and the corresponding reference points in sequence.

[0069] 2. Data segment video recording:

[0070] In order to facilitate users to record a complete video containing all data segments, this embodiment provides a countdown playback function. After the user clicks a button to start the countdown function, the system starts the countdown, for example, a countdown of 3 seconds. After the countdown ends, the WEB page automatically switches the playback data segments and the images of the reference points in sequence. During the countdown, the user can start the video recording device to start recording the video played on the WEB page. The video device setting parameters are a frame rate of not less than 30fps, that is, 30 frames per second, and a video resolution of not less than 1080p.

[0071] The image playback duration of each data segment and its reference point is 200 milliseconds. If it is divided into 100 data segments, 5 data segments and their reference points can be played per second during automatic playback, and all of them can be played in 20 seconds. Users can record a video of about 20 seconds.

[0072] 3. Video upload and segmentation:

[0073] The user securely uploads the recorded video to the server that needs to receive the text data. The server divides the video into individual pictures by frame based on the frame rate and playback duration of the video. Due to the continuity of the video frames, there will be a certain number of redundant pictures in each data segment to ensure the integrity and accuracy of the data. In this embodiment, the video frame rate is 30fps, and the image duration of each data segment and its reference point during recording is 200 milliseconds, so there will be 6 identical pictures in each data segment.

[0074] 4. Image processing:

[0075] In order to extract the valid data part of the image, that is, the area surrounded by the four black rectangular blocks, the server performs a series of processing on the image.

[0076] The specific steps are as follows:

[0077] 1. First, use grayscale to convert the segmented color image into a grayscale image to reduce the amount of calculation; use an appropriate threshold to binarize the grayscale image to highlight the black rectangular blocks.

[0078] 2. Calculate the gradients Ix and Iy in the x and y directions of each pixel in the image, as well as their products IxIy, Ixx and Iyy (second-order derivatives). Construct corner response function: Use the Harris corner detection formula to calculate the corner response value R of each pixel.

[0079] R = det(M) - k·trace(M) 2 ;

[0080] Among them, M is the structure tensor, which consists of IxIy, Ixx, Iyy; k is a constant, usually between 0.04 and 0.06.

[0081] Set a threshold and consider points with response values ​​R greater than the threshold as candidate corner points.

[0082] 3. For each candidate corner point, check whether there are other candidate corner points in its neighborhood and calculate the distance between them.

[0083] Based on the distance information and angle relationship, the corner points most likely to belong to the black rectangular block are identified by analyzing the relative positions, angles and distances between the corner points.

[0084] Among the identified rectangular corner points, redundant corner points are removed and only the content within the corner point range is retained.

[0085] 5. Character recognition and post-processing:

[0086] Use the region detection model based on convolutional neural network to identify the characters in the region. The specific steps are as follows:

[0087] 1) Data preparation: collect image datasets containing hexadecimal character data segments and annotate them; then divide them into training sets and test sets;

[0088] 2) Model training: Use the labeled data set to train the target detection model and character recognition model respectively. The target detection model uses the Faster R-CNN target detection algorithm to identify the area where the characters are located; the character recognition model uses the CRNN (Convolutional Recurrent Neural Network) to identify the characters in the area where the characters are located and output the data segment.

[0089] 3) Model evaluation: Evaluate the performance of the model on the test set, including indicators such as accuracy and recall.

[0090] 4) Model deployment: Deploy the trained model to actual applications for real-time detection and recognition of hexadecimal data segments.

[0091] To ensure the accuracy of the data, the system performs a CRC check on the recognized character data segment. If the check fails, the system will automatically replace it with a redundant image (i.e., another 5 of the 6 identical images) until the data check passes.

[0092] 6. File restoration and display:

[0093] Since the original data segments are switched for playback and recording in sequence, the recognized data segments can be restored to the original text data in sequence for display.

[0094] This embodiment combines image transmission and OCR technology to achieve secure, accurate and efficient information transmission. Reference points are used to determine the layout and format of the image to improve the accuracy of character recognition. Spelling error correction, punctuation addition and special symbol recognition in post-processing operations improve text readability.

[0095] Embodiment 2:

[0096] In this embodiment 2, an image recognition and text restoration device for secure information transmission is proposed, including:

[0097] Splitting module: The information sender splits the text data to be transmitted into multiple fixed-length data segments, displays the data segments on the WEB page, and generates reference points around each data segment;

[0098] Recording module: automatically switches and plays each data segment and reference point in sequence, and records the video played by all data segments and their reference points through the video recording device;

[0099] Segmentation module: securely transmit the video to the information receiving end, and the information receiving end segments the video into individual pictures by frame;

[0100] Extraction module: The information receiving end identifies the reference points in the image and extracts the valid data area in the image according to the reference points;

[0101] Recognition module: using a region detection model based on a convolutional neural network to identify characters in the valid data region;

[0102] Restoration module: restores the recognized characters to the original text.

[0103] The split modules include:

[0104] Reference point location unit: determines the reference point location and defines the structure containing the data segment and the reference point location;

[0105] Reference point style unit: Design the reference point style, using a style that is different from the data segment and other page content;

[0106] Generating unit: after the WEB page generates and displays the data segment, the reference point is generated at the corresponding position around the data segment according to the reference point position and reference point style.

[0107] The extraction module includes:

[0108] Grayscale unit: Use grayscale to convert the image into a grayscale image and perform binarization;

[0109] Candidate corner point unit: calculate the gradient Ix and Iy in the x and y directions of each pixel in the image, as well as the product IxIy, the second-order derivatives Ixx and Iyy; construct the corner point response function and calculate the corner point response value R of each pixel;

[0110] R = det(M) - k·trace(M) 2 ;

[0111] Where M is the structure tensor, which consists of IxIy, Ixx, Iyy; k is a constant;

[0112] Set a threshold and regard points with response values ​​R greater than the threshold as candidate corner points;

[0113] Corner point identification unit: For each candidate corner point, check whether there are other candidate corner points in its neighborhood and calculate the distance to other candidate corner points; based on the distance and angle relationship, identify the corner point that is most likely to belong to the reference point by analyzing the relative position, angle and distance between the corner points;

[0114] Area unit: Remove redundant corner points from the identified corner points and extract the content within the corner point range as the valid data area.

[0115] In the recognition module, the region detection model uses a target detection algorithm to recognize the character region in the valid data region, and extracts the recognized character region; and recognizes the characters in the character region through a character recognition model.

[0116] In addition, the splitting module adds a cyclic redundancy check (CRC) check code to the data segment, and the recognition module performs a CRC check after recognizing the characters in the area.

[0117] The device described in this embodiment 2 can implement the image recognition and text restoration method for secure information transmission described in embodiment 1, and achieve the same technical effect as embodiment 1.

[0118] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An image recognition and text restoration method for secure information transmission, characterized in that: include: S1. The information sending end divides the text data to be transmitted into multiple fixed-length data segments, displays the data segments on the WEB page, and generates reference points around each data segment; S2, automatically switching to play each data segment and reference point in sequence, and recording the video of all data segments and their reference points through a video recording device; S3, securely transmitting the video to an information receiving end, and the information receiving end divides the video into individual pictures by frame; S4, the information receiving end identifies the reference point in the picture, and extracts the valid data area in the picture according to the reference point; S5, using a region detection model based on a convolutional neural network to identify characters in the valid data region; S6. Restore the recognized characters to the original text.

2. The image recognition and text restoration method for secure information transmission according to claim 1, characterized in that: The method for generating the reference point in step S1 includes: S101, determine the reference point position, and define a structure including a data segment and the reference point position; S102, designing a reference point style, using a style that is different from the data segment and other page contents; S103: After the WEB page generates and displays the data segment, reference points are generated at corresponding positions around the data segment according to the reference point positions and reference point styles.

3. The image recognition and text restoration method for secure information transmission according to claim 1, characterized in that: Step S4 specifically includes: S401, converting the image into a grayscale image by grayscaling, and performing binarization; S402, calculating the gradient Ix and Iy in the x and y directions, as well as the product IxIy, the second-order derivatives Ixx and Iyy of each pixel in the image; constructing a corner point response function, and calculating the corner point response value R of each pixel; R=det(M)-k·trace(M) 2 ; Where M is the structure tensor, which consists of IxIy, Ixx, Iyy; k is a constant; Set a threshold and regard points with response values ​​R greater than the threshold as candidate corner points; S403, for each candidate corner point, check whether there are other candidate corner points in its neighborhood, and calculate the distance to other candidate corner points; based on the distance and angle relationship, identify the corner point that is most likely to belong to the reference point by analyzing the relative position, angle and distance between the corner points; S404: Remove redundant corner points from the identified corner points, and extract the content within the range of the corner points as a valid data area.

4. The image recognition and text restoration method for secure information transmission according to claim 1, characterized in that: In step S5, the region detection model uses a target detection algorithm to identify the character region in the valid data region, and extracts the identified character region; and recognizes the characters in the character region through a character recognition model.

5. The image recognition and text restoration method for secure information transmission according to any one of claims 1 or 4, characterized in that: A cyclic redundancy check (CRC) check code is added to the data segment, and step S5 performs a CRC check after identifying the characters in the area.

6. An image recognition and text restoration device for secure information transmission, characterized in that: include: Splitting module: The information sender splits the text data to be transmitted into multiple fixed-length data segments, displays the data segments on the WEB page, and generates reference points around each data segment; Recording module: automatically switches and plays each data segment and reference point in sequence, and records the video played by all data segments and their reference points through the video recording device; Segmentation module: securely transmit the video to the information receiving end, and the information receiving end segments the video into individual pictures by frame; Extraction module: The information receiving end identifies the reference points in the image and extracts the valid data area in the image according to the reference points; Recognition module: using a region detection model based on a convolutional neural network to identify characters in the valid data region; Restoration module: restores the recognized characters to the original text.

7. The image recognition and text restoration device for secure information transmission according to claim 6, characterized in that: The split modules include: Reference point location unit: determines the reference point location and defines the structure containing the data segment and the reference point location; Reference point style unit: Design the reference point style, using a style that is different from the data segment and other page content; Generating unit: after the WEB page generates and displays the data segment, the reference point is generated at the corresponding position around the data segment according to the reference point position and reference point style.

8. The image recognition and text restoration device for secure information transmission according to claim 6, characterized in that: The extraction module includes: Grayscale unit: Use grayscale to convert the image into a grayscale image and perform binarization; Candidate corner point unit: calculate the gradient Ix and Iy in the x and y directions of each pixel in the image, as well as the product IxIy, the second-order derivatives Ixx and Iyy; construct the corner point response function and calculate the corner point response value R of each pixel; R=det(M)-k·trace(M) 2 ; Where M is the structure tensor, which consists of IxIy, Ixx, Iyy; k is a constant; Set a threshold and regard points with response values ​​R greater than the threshold as candidate corner points; Corner point identification unit: For each candidate corner point, check whether there are other candidate corner points in its neighborhood and calculate the distance to other candidate corner points; based on the distance and angle relationship, identify the corner point that is most likely to belong to the reference point by analyzing the relative position, angle and distance between the corner points; Area unit: Remove redundant corner points from the identified corner points and extract the content within the corner point range as the valid data area.

9. The image recognition and text restoration device for secure information transmission according to claim 6, characterized in that: In the recognition module, the region detection model uses a target detection algorithm to recognize the character region in the valid data region, and extracts the recognized character region; and recognizes the characters in the character region through a character recognition model.

10. The image recognition and text restoration device for secure information transmission according to claim 6 or 9, characterized in that: The splitting module adds a cyclic redundancy check (CRC) check code to the data segment, and the recognition module performs a CRC check after recognizing the characters in the area.

Images