Method for detecting regions of interest of time-dependent text
By using image processing units in the information acquisition system, combining grayscale conversion and Huo's conversion, accurate automatic box selection of areas of interest for text changes over time in the production machine screen is achieved, and the existing technology is solved The problem of inaccurate and time-consuming box selection improves analysis efficiency and correctness.
Patent Information
- Application Number
- CN202011423168.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-12-08
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2040-12-08
AI Technical Summary
When the existing information acquisition system automatically analyzes the screen of the production machine, it is difficult to accurately box the area of interest that changes over time, resulting in a decrease in accuracy and waste of resources, and manual box selection takes time.
By loading specific text definition files in the image processing unit of the information acquisition system, detecting text areas using grayscale conversion, vertical or horizontal projection, straight line Ho's conversion and other methods, and obtaining candidate areas through voting plane and image connection processing , realize automatic box selection.
It greatly reduces the time set by users, improves the accuracy and correctness of information, reduces the waste of computing resources, and improves the degree of automation of the system.
Smart Images

Figure CN114663641B_ABST
Abstract
Description
Technical Field
[0001] The present invention provides a method for detecting a time-dependent text region of interest, in particular, an information capture system can be used to automatically and accurately frame a text region of interest that changes with time in an operation screen, making the user's settings more convenient, greatly helping to reduce the user's setting time, and accurately detecting the required information. Background Art
[0002] The rapid development of information technology and the Internet, the significant increase in computer computing power, and the continuous improvement of data collection and storage technology have also accelerated the accumulation, recording and acquisition of big data. In terms of production line automation technology, faster equipment and open architecture must be used to meet usage needs.
[0003] In traditional factories, all parameters generated by production machines during the production process are displayed on the screen. When the parameters reach a point where a warning is required, a signal light can be used to alert the on-site staff, which means that someone needs to be near the production machine to handle the emergency. If it is necessary to count the parameters displayed on the screen, someone is also required to record them regularly. Therefore, in order to solve the above problem, some companies use machine equipment information capture systems, which can achieve customized or automated analysis or statistics and warning settings by capturing and analyzing the information on the screen.
[0004] However, there is a lot of information on the screen of the machine. In order for the information acquisition system to analyze the information on the screen effectively and quickly, the user must first select the region of interest (ROI) through the user interface of the machine software before providing it to the information acquisition system for analysis. At present, some problems often encountered when manually selecting the area to be identified are as follows:
[0005] When there are too many items to be recognized on the machine screen, many regions of interest for text that change over time (or called dynamic and static text regions, whose content changes over time) must be manually selected, which will result in a lot of time spent selecting the text region of interest when setting the mode. If the selected region is too large and contains too much unnecessary information; or the selected region is too small, so that the necessary information is not included in it, or the length of the text region is unknown (such as three or five digits, etc.), resulting in improper or incorrect selection of the region, the accuracy of recognition will decrease and the effect will be poor, and computing resources will be consumed. After the data is obtained, its correctness needs to be manually verified, and the selected region needs to be improved when necessary, which is quite time-consuming. Although there are many ready-made methods on the market that can automatically detect text regions, most algorithms on the market still have the problem of containing too much unnecessary information or the necessary information cannot be correctly detected when selecting the region of interest for text that changes over time.
[0006] When developing big data applications for industrial production machines, the first task is to collect information about the relevant machines at the production line machines (or cloud terminals). Much of the important information comes from the user interface of the machine software. If the machine screen can automatically and accurately select the area of interest with text that changes over time, it will make the user's settings more convenient. In addition to greatly helping to reduce the user's setting time, it can also correctly detect the required information. Automated detection of text areas of interest will also help system integration engineers in script editing and related machine parameter settings and other data collection and analysis applications, which is the direction that those engaged in this industry are eager to study and improve. Summary of the invention
[0007] The main purpose of the present invention is to use the image processing unit of the information capture system to detect the text interest area that changes with time in units of specific characters or character sets in the operation screen of the production machine, processing machine or other equipment. It can also be further based on the result of judging whether there are historical screens, supplemented by the detection of candidate areas of the text interest area that changes with time, and then the candidate areas obtained are processed as a union, and finally the automatically selected text interest area that changes with time is obtained. The above-mentioned detection method can be executed separately, in combination or all at once, so that the user can choose to use it according to the operation screen, and automatically select the text interest area that changes with time accurately. Since the user only needs to confirm whether the required data has been selected, the user's setting is more convenient, which will greatly help to reduce the user's setting time, and the required information can be correctly detected.
[0008] A secondary purpose of the present invention is that the detection method loads a specific text definition file, wherein the text list includes characters or word sets for the user to add, delete or modify according to the operation screen, and detects specific text on the image in the operation screen, and then expands the detection to the boundary range of each specific text interest area to generate a search window, so that the image in the search window can be converted to grayscale, and vertical or horizontal projection search is performed respectively, and a linear Hoffmann transformation processing flow can be used to detect the text and input box that changes with time in the grayscale image, and then the candidate areas of the respective text interest areas that change with time are merged, and finally the text interest area that changes with time is obtained.
[0009] Another object of the present invention is that the detection method first determines whether there is historical screen data. If there is no historical screen data, a picture of the operation screen is captured and converted into a grayscale image and regarded as a reference image. After another picture is captured and converted into a grayscale image, the pixel position image greater than a specific threshold value can be imaged in the voting plane for cumulative voting according to the difference calculation result between the grayscale image and the reference image. If there is historical screen data, an image is loaded from the historical image database as the reference image, and a grayscale image of the historical screen is captured and the reference image is differenced, and the pixel position image greater than the specific threshold value is imaged in the voting plane for cumulative voting. Then, the accumulated voting plane can be normalized and image connected to calculate the bounding box of each connected block of the image, and the overlapping bounding boxes are merged together and the boundary range of the bounding box is expanded to obtain candidate areas of interest for text that changes with time, and then the candidate areas obtained are processed by union to obtain an automatically selected text that changes with time.
[0010] Another object of the present invention is that after the information acquisition system initializes the relevant system parameters, the detection method can detect and process the regions of interest of text that change over time in units of specific characters or character sets in the above-mentioned machine screen, and after the processing is completed, the candidate regions of the regions of interest of text that change over time are obtained, and the processing flow after the above-mentioned processing is completed after the cumulative voting of the image space is completed under the premise of whether there is historical screen data, and regardless of whether there is historical screen data, a result will be obtained after the whole voting process is completed, and based on the result, the candidate regions of the regions of interest of text that change over time can be obtained, and then the respective candidate regions are processed as a union, and finally the automatically selected regions of interest of text that change over time are obtained. BRIEF DESCRIPTION OF THE DRAWINGS
[0011] Figure 1 FIG. 4 is a flow chart of the steps of the first preferred embodiment of the present invention.
[0012] Figure 2 It is a schematic diagram of the machine screen of the present invention.
[0013] Figure 3 The present invention is a schematic diagram of the action of searching the image in the search window in a projection manner and selecting the text interest area that changes with time.
[0014] Figure 4 This is a schematic diagram of the action of performing a linear Hough transformation on an image in a search window to restore a text region of interest that changes over time.
[0015] Figure 5 FIG. 1 is a flow chart of steps of a second preferred embodiment of the present invention.
[0016] Figure 6 This is a step flow chart (II) of the second preferred embodiment of the present invention.
[0017] Figure 7 This is an action diagram of the present invention performing voting union processing on the candidate areas detected based on the presence or absence of historical images, and finally automatically selecting the text interest area that changes over time.
[0018] Figure 8 FIG. 1 is a flow chart of the steps of the third preferred embodiment of the present invention.
[0019] Fig. 9 This is a step flow chart (II) of the third preferred embodiment of the present invention.
[0020] Fig.10 This is a step flow chart (three) of the third preferred embodiment of the present invention.
[0021] Explanation of the accompanying reference numerals: 100-operation screen; 101-search window; 102-text that changes with time; 103-attribute; 104-unit; 200-histogram; 201-region of interest for text that changes with time; 300-input box; 301-counting plane of linear Hough transform; 302-line restoration; 400-reference image; 401-voting plane after normalization processing; 402-image after connectivity processing; 403-candidate region of interest for text that changes with time; 404-auto-selected region of interest for text that changes with time. DETAILED DESCRIPTION
[0022] In order to achieve the above-mentioned purpose and effect, the technical means and structures adopted by the present invention are illustrated in detail with reference to the preferred embodiments of the present invention, and their structures and functions are as follows.
[0023] Please refer to Figure 1 to Figure 4As shown, there are respectively a flow chart of the steps of the first preferred embodiment of the present invention, a schematic diagram of the machine screen, a schematic diagram of the action of searching the image in the search window by projection and selecting the text interest region that changes with time, and a schematic diagram of the action of performing linear Hough transformation restoration on the image in the search window to obtain the text interest region that changes with time. It can be clearly seen from the figure that the detection method of the time-dependent text interest region of the present invention is applicable to a built-in or external information acquisition system including but not limited to a production machine, processing machine or other equipment, and the machine provides an operation screen 100 to be displayed on the screen to present various parameters or data of the machine operation and production status, wherein the information acquisition system has an image processing unit, including but not limited to a field programmable logic gate array (FPGA), an application or other intermediate software and hardware, for detecting and comparing specific characters or character sets, and automatically detecting the text area of interest by matching a period of image status that changes with time, including historical image data or real-time images. The detection method includes the following implementation steps:
[0024] (S100)Start.
[0025] (S101) Initialize system parameters, and then execute step (S201).
[0026] (S201) Load a specific character definition file.
[0027] (S202) Detection of specific characters.
[0028] (S203) Expand the boundary range of each specific text interest area and generate a search window.
[0029] (S204) Convert the image in the search window into grayscale and select the region of interest of the text that changes over time.
[0030] (S205) Obtaining a text region of interest that changes over time.
[0031] (S206)End.
[0032] It can be clearly seen from the figure and the above-mentioned implementation steps that when the information acquisition system of the present invention starts, it first initializes the relevant system parameters (Initialize parameters), including the image source selection of the host (Host), the resolution setting, etc., and loads a specific character definition file (Load specific character definition file). The specific character definition file is a text list pre-defined by the user for searching, including uppercase and lowercase letters, spaces, numbers, punctuation marks and other symbols or a character set of characters, and the text list content can be added, deleted or modified by the user according to the specific text area that may appear in the operation screen. Then, according to each loaded specific character or character set, the existing detection method is used to detect the text area, including but not limited to pattern comparison, color similarity, texture, morphology feature related methods, long short-term memory optical character recognition [Long Short-Term Memory (LSTM) OCR], deep learning for object detection [Object Detection in Deep Learning, including Single Shot Multi-Box Detector (SSD), You Only Look Although optical text recognition algorithms such as Once (YOLO) etc. cannot determine whether the detected area is a text area 102 that changes over time, they can provide detection position information of specific characters or character sets (such as specific character interest areas).
[0033] After obtaining each specific text region of interest, an extension tolerance value can be taken for the border height of each specific text region of interest to expand its border height and serve as the height of the search window 101. The leftmost or rightmost border of each region of interest can also be extended to the left or right by a width tolerance value and serve as the width of the search window 101. The height and width tolerance values can be adjusted by the user or given a preset value. This search window 101 is a region of interest or candidate region (Region proposal) where time-dependent (also known as time-varying) text may appear, so they are all explained together in the following description of this case.
[0034] In this embodiment, it is necessary to convert the image color space in the search window 101 to gray space (Convert each search window to gray space from color space) and select the region of interest of the text that changes with time, and then perform vertical or horizontal projection search and linear Hough Transform processing on the converted gray image, wherein the projection search is based on the fact that the text 102 that changes with time is generally located between the attribute 103 and the unit 104, such as AX, ACY, MX and other attributes, which are descriptions of the values of -4.598, -182.340, -8.200 and other values in the region of interest, and the unit of mm is a specific character set, and then the gray image in the search window 101 that has been obtained can be converted to the histogram 200 by vertical or horizontal projection. If vertical projection is used as an example, the horizontal axis represents the width of the search window 101, and the vertical axis represents the projection amount of the gray image. Then, the histogram 200 is searched from right to left. Search for the projection amount area (in the direction indicated by the horizontal arrow in the figure). When the length of the projection amount to the left of a certain point on the curve that is continuously non-zero (or the continuous projection amount is lower than a certain set value) reaches a certain tolerance value, it can be determined that this point is the boundary area between the attribute 103 and the area of interest, and its width is taken as the width of the time-varying text area of interest 201, and its height is equal to the height of the search window 101, so that the candidate area of the time-varying text area of interest 201 can be automatically selected. Similarly, the processing flow of the horizontal projection method search is the same, and it can be an option for the user to choose and set any projection method according to the text direction in the operation screen.
[0035] The linear Hough transform is for the input box 300 for text input in the text interest region 201 that changes with time in the operation screen. Since the input box 300 has obvious boundary information, when searching in the projection mode, the width of the text interest region 201 that changes with time may not be the width of the input box 300. However, its boundary characteristics can be used for detection of the linear Hough transform. The processing flow is to first perform edge detection (Edge Detection) on the search window 101 to obtain boundary information for calculation, and then convert it to the counting plane 301 of the linear Hough transform. In this counting plane, the intersection points with higher counts within a specific angle range (such as -90 degrees, 0 degrees, 90 degrees) can be limited to restore the line 302 through the setting of the threshold value, and the rectangular bounding box (Bounding Box) equivalent to the size of the input box 300 can be further calculated through the four restored intersection points. box) to obtain the candidate region of interest of the text changing with time, and then combine the vertical or horizontal projection search and the linear Hoffmann transformation to respectively detect the candidate region of interest of the text changing with time, and finally obtain the automatically selected region of interest of the text changing with time.
[0036] Please refer to Figure 5 to Figure 7 As shown, there are respectively the flow charts (I) and (II) of the second preferred embodiment of the present invention and the schematic diagram of the action of voting and uniting the detected candidate regions based on the existence of historical images, and finally automatically selecting the regions of interest of texts that change with time. It can be clearly seen from the figure that the method for detecting the regions of interest of texts that are time-dependent of the present invention includes the following implementation steps:
[0037] (S100)Start.
[0038] (S101) Initialize system parameters.
[0039] (S102) Determine whether there is a history screen. If yes, execute step (S401); if no, execute step (S301).
[0040] (S301) Capture a picture.
[0041] (S302) Converting the captured image into a grayscale image and treating it as a reference image.
[0042] (S303) Determine whether to stop the voting accumulation process? If yes, execute step (S501), if no, execute step (S304).
[0043] (S304) Capture a current picture.
[0044] (S305) Convert the captured image into a grayscale image.
[0045] (S306) Performing a difference operation on the grayscale image and the reference image to obtain a difference operation result.
[0046] (S307) According to the difference calculation result, the pixel positions that are greater than the specific threshold value are mapped in the voting plane to perform cumulative voting actions, and then step (S303) is repeated.
[0047] (S401) An image is loaded from a historical image database as a reference image.
[0048] (S402) Determine whether to stop the voting accumulation process? If yes, execute step (S501), if no, execute step (S403).
[0049] (S403) Extracting a grayscale image of a historical scene from a historical image database.
[0050] (S404) Performing a difference operation on the grayscale image and the reference image to obtain a difference operation result.
[0051] (S405) According to the difference calculation result, the pixel positions greater than the specific threshold value are mapped in the voting plane to perform cumulative voting actions, and then step (S402) is repeated.
[0052] (S501) Normalize the accumulated voting plane.
[0053] (S502) Image connectivity processing.
[0054] (S503) Calculate the bounding boxes of each connected block and merge the overlapping bounding boxes together.
[0055] (S504) Expanding the boundary range of the bounding box to obtain candidate regions of the text region of interest that changes over time.
[0056] (S505) The candidate regions are processed as a union to obtain an automatically selected region of interest of text that changes over time.
[0057] (S506)End.
[0058] It can be clearly seen from the figure and the above-mentioned implementation steps that the present invention can select different detection methods based on the result of determining whether there is a historical screen. If there is no historical data of the machine operation screen, the screen data must be collected for detection. The processing flow is to first retrieve a base screen for the machine screen (Retrieve basescreen) and convert the retrieved screen into a grayscale image (Color to gray). This grayscale image can be regarded as a base image (Base image) 400, and the base image 400 will be stored in a historical image database (Historical images DB) on the one hand, and continue to retrieve a current screen to perform the following voting accumulation process on the other hand.
[0059] If it is determined that the voting accumulation process needs to be continued, the processing flow is to first retrieve a current screen from the machine's operating screen (Retrieve current screen from host device), and after converting the retrieved screen into a grayscale image, the grayscale image can be stored in the historical image database as historical screen data on the one hand, and on the other hand, the following voting accumulation process is continued, and then the grayscale image and the pixels in each block of the reference image 400 are subjected to difference calculation (Difference with base image) to obtain a difference calculation result. According to the difference calculation result, the pixel position greater than a specific threshold value can be determined as a changed pixel, and mapped to the corresponding position in the voting plane (Score map) to perform an accumulation voting action (Vote to accumulation map), and then the above-mentioned voting accumulation process is repeated until it is determined to stop the entire voting process.
[0060] When it is determined that there is past historical image data, the next step is to use the image data collected in the past for detection. This step can be regarded as an offline detection of text interest areas that change over time. The processing flow is to load an image from the historical image database as the reference image 400, and capture a grayscale image of the historical image from the historical image database, and then perform a difference operation on the pixels in each block of the grayscale image and the reference image 400 to obtain a difference operation result. The pixel position greater than a specific threshold value can be determined as a changed pixel based on the difference operation result, and mapped to the corresponding position in the voting plane to perform a cumulative voting action until it is determined to stop the entire voting process.
[0061] Under the premise of whether there is past historical image data, after completing the above voting accumulation process, the accumulated voting plane can be normalized (Normalize accumulation map to score map), so that the pixel values in the grayscale image are defined in the range of [0, 1] after binary conversion, where 0 represents black and 1 represents white, to obtain the voting plane 401 after normalization processing at this stage, and then perform Blob analysis and extract high score blobs for image connectivity processing to obtain the image 402 after connectivity processing at this stage. The user can also set some additional filtering processing before connectivity processing, such as image dilation (Dilation) or closure (Closing) in morphology, so as to remove the blocks with small voting numbers, and then calculate the bounding box of each connected block, merge the overlapping bounding boxes together (Calculate the bounding box of each blob and merge the overlapping), and then extend the obtained bounding box in four directions of up, down, left and right to expand the bounding box range (Expanse the bounding box box), and the extension range can be defined by the user as a suitable range to obtain the candidate region 403 of the text interest region that changes with time, and then the respective candidate regions are processed as a union, and finally the automatically selected text interest region 404 that changes with time is obtained.
[0062] Please also refer to Figure 8 to Figure 10 As shown in FIG. 1 , the step flow charts (I) to (III) of the third preferred embodiment of the present invention are shown. It can be clearly seen from the figure that the time-dependent text region of interest detection method of the present invention includes the following implementation steps:
[0063] (S100)Start.
[0064] (S101) Initialize system parameters, and then execute steps (S102) and (S201).
[0065] (S102) Determine whether there is historical screen data. If yes, execute step (S401); if no, execute step (S301).
[0066] (S201) Load a specific character definition file.
[0067] (S202) Detection of specific characters.
[0068] (S203) Expand the boundary range of each specific text interest area and generate a search window.
[0069] (S204) Convert the image in the search window into grayscale and select the region of interest of the text that changes over time.
[0070] (S205) Obtain candidate regions of the text region of interest that changes over time, and then execute step (S601).
[0071] (S301) Capture a picture.
[0072] (S302) Converting the captured image into a grayscale image and treating it as a reference image.
[0073] (S303) Determine whether to stop the voting accumulation process? If yes, execute step (S501), if no, execute step (S304).
[0074] (S304) Capture a current picture.
[0075] (S305) Convert the captured image into a grayscale image.
[0076] (S306) Performing a difference operation on the grayscale image and the reference image to obtain a difference operation result.
[0077] (S307) According to the difference calculation result, the pixel positions that are greater than the specific threshold value are mapped in the voting plane to perform cumulative voting actions, and then step (S303) is repeated.
[0078] (S401) An image is loaded from a historical image database as a reference image.
[0079] (S402) Determine whether to stop the voting accumulation process? If yes, execute step (S501), if no, execute step (S403).
[0080] (S403) Extracting a grayscale image of a historical scene from a historical image database.
[0081] (S404) Performing a difference operation on the grayscale image and the reference image to obtain a difference operation result.
[0082] (S405) According to the difference calculation result, the pixel positions that are greater than the specific threshold value are cumulatively voted in the voting plane, and step (S402) is repeated.
[0083] (S501) Normalize the accumulated voting plane.
[0084] (S502) Image connectivity processing.
[0085] (S503) Calculate the bounding boxes of each connected block and merge the overlapping bounding boxes together.
[0086] (S504) Expand the boundary range of the bounding box to obtain candidate regions of the text region of interest that changes over time, and then execute step (S601).
[0087] (S601) The candidate regions are processed as a union to obtain an automatically selected region of interest of text that changes over time.
[0088] (S602)End.
[0089] It can be clearly seen from the figure and the above-mentioned implementation steps that in the first preferred embodiment of the present invention, steps (S201) to (S205) are used to detect and process the regions of interest of text that change over time in units of specific characters or character sets in the machine screen. After the processing is completed to obtain the candidate regions of the regions of interest of text that change over time, the second preferred embodiment will be waited for to complete steps (S501) to (S504), which is a processing flow after the machine completes the cumulative voting of the image space under the premise of whether there is historical screen data, so as to obtain the candidate regions of the regions of interest of text that change over time, and then the respective candidate regions are processed as a union, and finally the automatically selected regions of interest of text that change over time are obtained.
[0090] Therefore, the present invention mainly provides a method for detecting a time-dependent text region of interest, which is applicable to an image processing unit of an information acquisition system, and is used to detect a time-varying text region of interest 201 in a specific character or character set in an operation screen 100 of a production machine, a processing machine or other equipment, such as size (such as cm, mm), temperature (℃) or pressure (N / m2), etc., but is not limited to this. It can also be further based on the result of determining whether there is a historical screen, supplemented by the detection of candidate regions 403 of the time-varying text region of interest, and then the respective candidate regions are processed as a union, and finally the automatically selected time-varying text region of interest 404 is obtained. The above-mentioned detection method can be executed separately, in combination or all at once, so that the user can choose to use it according to the operation screen 100, and automatically select the time-varying text region of interest 404 accurately. It only needs to confirm whether the required data has been selected, so that the user's setting is more convenient, which will greatly help to reduce the user's setting time, and the required information can be correctly detected. In addition, automatic detection of text areas of interest will also help current system integration engineers in data collection and analysis applications such as script editing and related machine parameter settings.
[0091] The above detailed description is only for a preferred feasible embodiment of the present invention, but the embodiment is not used to limit the protection scope of the present invention. All other equal changes and modifications made without departing from the technical spirit disclosed by the present invention should be included in the protection scope covered by the present invention.
Claims
1. A method for detecting a time-dependent text region of interest, applicable to an information acquisition system, wherein the information acquisition system has an image processing unit for detecting an image in an operation screen, including historical screen data or real-time images, and automatically selecting a time-dependent text region of interest, characterized in that: The detection method includes the following implementation steps: (B01) Determine whether the historical screen data exists, if yes, execute step (B09), if no, execute step (B02); (B02) capturing a screen shot of the operation screen; (B03) converting the captured image into a grayscale image and treating it as a reference image; (B04) Determine whether to stop a voting accumulation process? If yes, execute step (B14); if no, execute step (B05); (B05) capturing a current screen of the operation screen; (B06) converting the captured image into a grayscale image; (B07) performing a difference operation on pixels in each block of the grayscale image and the reference image to obtain a difference operation result; (B08) mapping pixel positions greater than a specific threshold value on a voting plane according to the difference calculation result to perform cumulative voting, and then repeating step (B04); (B09) loading an image from a historical image database as a reference image; (B10) Determine whether to stop a voting accumulation process? If yes, execute step (B14); if no, execute step (B11); (B11) extracting a grayscale image of a historical scene from the historical image database; (B12) performing a difference operation on pixels in each block of the grayscale image and the reference image to obtain a difference operation result; (B13) mapping the pixel positions that are greater than the specific threshold value on a voting plane according to the difference calculation result to perform cumulative voting, and then repeating step (B10); (B14) normalizing the accumulated voting plane; (B15) performing image connectivity processing on the voting plane; (B16) calculating the bounding boxes of each connected block of the image, and merging the overlapping bounding boxes together; (B17) expanding the boundary range of the bounding box to obtain a candidate region of the text region of interest that changes over time; (B18) The candidate regions obtained are subjected to a union process to obtain the automatically selected region of interest of the text that changes with time.
2. The method for detecting a time-dependent text region of interest as claimed in claim 1, wherein: The grayscale image converted in steps (B03) and (B06) is stored in the historical image database and serves as the grayscale image of the historical frame.
3. The method for detecting a time-dependent text region of interest as claimed in claim 1, wherein: The normalization process in step (B14) is to convert the pixel values in the grayscale image into binarized values and define them in the interval [0, 1].
4. The method for detecting a time-dependent text region of interest as claimed in claim 1, wherein: The image connection processing in step (B15) is to analyze the voting plane through Blob and extract the blocks with high voting numbers.
5. The method for detecting a time-dependent text region of interest as claimed in claim 1, wherein: In step (B15), before image connectivity processing is performed, blocks with small vote numbers are removed through morphological filtering processing including image expansion and closure.
6. A method for detecting a time-dependent text region of interest, applicable to an information acquisition system, wherein the information acquisition system has an image processing unit for detecting an image in an operation screen, including historical screen data or real-time images, and automatically selecting a time-dependent text region of interest, characterized in that: The detection method includes the following implementation steps: (C01) Initialize system parameters, and then execute steps (C02) and (C03); (C02) Determine whether the historical screen data exists, if yes, execute step (C15), if no, execute step (C08); (C03) loading a specific text definition file; (C04) detecting specific text in the image in the operation screen; (C05) expanding the boundary range of each detected specific text region of interest and generating a search window; (C06) converting the image in the search window into grayscale and selecting the region of interest of the text that changes with time; (C07) obtaining the candidate region of the text region of interest that changes with time, and then executing step (C24); (C08) capturing a screen shot of the operation screen; (C09) converting the captured image into a grayscale image and treating it as a reference image; (C10) Determine whether to stop a voting accumulation process? If yes, execute step (C20); if no, execute step (C11); (C11) capturing a current screen of the operation screen; (C12) converting the captured image into a grayscale image; (C13) performing a difference operation on pixels in each block of the grayscale image and the reference image to obtain a difference operation result; (C14) mapping pixel positions greater than a specific threshold value on a voting plane according to the difference calculation result to perform cumulative voting, and then repeating step (C10); (C15) loading an image from a historical image database as a reference image; (C16) Determine whether to stop a voting accumulation process? If yes, execute step (C20); if no, execute step (C17); (C17) extracting a grayscale image of a historical scene from the historical image database; (C18) performing a difference operation on pixels in each block of the grayscale image and the reference image to obtain a difference operation result; (C19) mapping the pixel positions that are greater than the specific threshold value in a voting plane according to the difference calculation result to perform a cumulative voting action, and then repeating step (C16); (C20) normalize the accumulated voting plane; (C21) performing image connectivity processing on the voting plane; (C22) calculating the bounding boxes of each connected block of the image, and merging the overlapping bounding boxes together; (C23) expanding the boundary range of the bounding box to obtain a candidate region of the text region of interest that changes over time, and then executing step (C24); (C24) The candidate regions obtained are subjected to a union process to obtain the automatically selected region of interest of the text that changes with time.
7. The method for detecting a time-dependent text region of interest as claimed in claim 6, wherein: The search window of step (C05) performs vertical or horizontal projection search and linear Hoffmann transformation processing flow on the converted grayscale image, and detects the text and input box that change with time in the grayscale image to obtain candidate areas of interest for the text that changes with time, and then merges the respective detected candidate areas of interest for the text that changes with time.
8. The method for detecting a time-dependent text region of interest as claimed in claim 7, wherein: The projection method search is based on the fact that the time-varying text is located between the attribute and the unit, and the grayscale image is projected onto a histogram, and then the histogram is searched for areas where the projection amount is continuously non-zero or lower than a set value to determine the boundary area between the attribute and the time-varying text area of interest, and its width is taken as the width of the time-varying text area of interest, and its height is equal to the height of the search window.
9. The method for detecting a time-dependent text region of interest as claimed in claim 7, wherein: The linear Hofmeister transformation performs edge detection on the search window to obtain boundary information for calculation, and then converts it to the counting plane of the linear Hofmeister transformation. In the counting plane, the line is restored by setting a threshold value to calculate a boundary box equivalent to the input box.
10. The method for detecting a time-dependent text region of interest as claimed in claim 6, wherein: The grayscale image converted in steps (C09) and (C12) is stored in the historical image database and serves as the grayscale image of the historical frame.
11. The method for detecting a time-dependent text region of interest as claimed in claim 6, wherein: The normalization process in step (C20) is to convert the pixel values in the grayscale image into binarized values and define them in the interval [0, 1].
12. The method for detecting a time-dependent text region of interest as claimed in claim 6, wherein: The image connectivity processing in step (C21) is to analyze the voting plane through Blob and extract the blocks with high voting numbers.
13. The method for detecting a time-dependent text region of interest as claimed in claim 6, wherein: In step (C21), before image connectivity processing is performed, blocks with small votes are removed through morphological filtering processing including image expansion and closure.
