Character recognition system, character recognition method, and location map creation system

Abstract

To provide a character recognition system and a character recognition method that can not only accurately recognize handwritten characters to obtain accurate character data but also shorten the time required for recognition, and a location map creation system using the character recognition system. [Solution] The system comprises an image input means 1 for inputting an image G containing handwritten characters, an OCR means 2 for recognizing the characters contained in said image G to obtain character data T·T..., an extraction means 3 for extracting partial images g·g... containing handwritten characters from said image G, and a matching means 5 for comparing said partial images g·g... with said character data T·T... and matching those with similar features, and is configured so that character data T corresponding to any partial image g can be obtained by said matching means 5.

Description

[Technical Field]

[0001] The present invention relates to a character recognition system and a character recognition method for recognizing handwritten characters from an image containing handwritten characters and converting them into character data, and also to a location map creation system for creating a clean-up location map from a handwritten site map at a construction site. [Background technology]

[0002] In the field of information transmission, advances in digitalization have promoted paperless operations in various fields. In the industrial world, manufacturing and construction work is carried out based on blueprints, and while traditionally these were hand-drawn using drafters, the spread of CAD (Computer Aided Design) systems has led to an increasing number of companies creating drawings on computers. However, at manufacturing and construction sites, it is often necessary to take notes and record the situation on-site, and as a result, many people still keep printed out paper drawings on hand and write letters and figures on them with a pen.

[0003] One example is soil investigations for house construction. When building a house, the ground is investigated to see if it is suitable for the planned building specifications, and the foundation to be constructed is designed based on the results of the investigation. One of the investigations carried out at this time is the screw weight penetration test. The screw weight penetration test is used particularly for simple investigations of soft ground, and involves applying a certain load to a rod equipped with a screw-equipped penetration body, which is then pushed vertically into the ground while rotating. The rotational force and force required to penetrate to a certain depth are then measured. This measurement is carried out at multiple locations on the construction site.

[0004] The measurements obtained from the above tests are written by hand on paper drawings by workers on the construction site. The reason for using this analog method is that it is simpler and easier to rewrite than carrying around a digital display device and using software to display drawing data or input text information. However, handwritten drawings are often dirty and the writing is difficult to read, so when the results of ground surveys based on these drawings are handed over to construction companies, etc., they must be cleaned up using CAD or other tools to create neat, easy-to-read drawings. Traditionally, this work has been done by hand.

[0005] Therefore, it is considered to automatically recognize and clean up handwritten drawings. Character recognition technology is an important technology for this. In particular, a high recognition rate for handwritten characters is required, and it is desirable to have a system that can accurately recognize various character styles. Various methods have been developed to improve the recognition rate. For example, Patent Document 1 discloses a mechanism for recognizing handwritten characters and converting them into character data, particularly a technology for normalizing images showing words.

[0006] The technology in Patent Document 1 employs a method of inputting an image containing handwritten characters, dividing the image into individual characters, and normalizing each character individually. Even if the handwritten characters are a series of sentences, the size and aspect ratio of each character may differ. Predetermined image processing is then applied to the partial images divided into each character. For example, characters that appear vertically squashed are stretched vertically, and widely spaced characters are compressed. Applying image processing is said to make it easier to recognize handwritten characters because each character is normalized to a predetermined size and aspect ratio. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Japanese Patent Application Laid-Open No. 2001-344565 Summary of the Invention [Problem to be solved by the invention]

[0008] However, the technology of Patent Document 1 requires dividing each character string and performing image processing on it. When characters are written all over a drawing, such as the above-mentioned ground survey drawing, there is a problem that it takes a long time to recognize all the characters. Furthermore, even though the characters are normalized, they are still transformed through image processing, which can actually diminish the characteristics of the original characters, potentially resulting in misrecognition. As described above, the conventional technology has a problem in that it is not possible to improve both the recognition rate and the overall processing speed.

[0009] The present invention has been made in view of the above-mentioned problems, and its object is to provide a character recognition system and a character recognition method that can not only accurately recognize handwritten characters to obtain accurate character data, but also reduce the time required for recognition, as well as a location map creation system that uses this character recognition system. [Means for solving the problem]

[0010] The means adopted by the present inventors to solve the above problems will be described below. The character recognition system of the present invention comprises an image input means for inputting an image containing handwritten characters, an OCR means for recognizing the characters contained in the image to obtain character data, an extraction means for extracting a partial image containing handwritten characters from the image, and a matching means for comparing the partial image with the character data and matching those with similar features, and is characterized in that character data corresponding to any partial image can be obtained by the matching means.

[0011] OCR (Optical Character Recognition) is a method for recognizing and reading multiple character data, including handwritten characters as well as typed characters, from the entire input image. Character recognition is not limited to reading character strings, but also includes reading individual characters. By performing character recognition processing on the entire image, processing time can be reduced compared to when performing character recognition processing on individual character image portions one by one.

[0012] The character data read by the OCR means includes not only handwritten characters but also typed characters. The extraction means extracts the part containing handwritten characters as a partial image from the entire input image. This means does not recognize the characters, but simply extracts them as images.

[0013] The matching means then compares each character data recognized by the OCR means with the partial image extracted by the extraction means, and matches those with similar characteristics. Similar characteristics include, for example, coordinate values ​​for each image as a whole, but are not limited to these and include other characteristics as well. This makes it possible to quickly select appropriate character data from multiple character data according to the characteristics of the partial image, unlike conventional methods that directly recognize characters from images and turn them into character data.

[0014] A means that can be employed to solve the problem includes an image processing means that applies different image processing to the image to obtain a plurality of processed images, and an auxiliary OCR means that recognizes characters contained in the image from the plurality of processed images to obtain a plurality of auxiliary character data, and the matching means compares the partial image with character data including the plurality of auxiliary character data to match those with similar features, and can also include a ranking means that adopts the character data with the largest number of characters among the character data matched to any partial image as the character data corresponding to that partial image.

[0015] Even if the OCR means misrecognizes characters, the image processing means, auxiliary OCR means, and ranking means allow for appropriate matching of character data. Specifically, the image processing means performs different types of image processing on the original image to obtain multiple processed images, and the auxiliary OCR means recognizes characters from these processed images to obtain auxiliary character data. The number of auxiliary character data obtained using this series of means is equal to the number of types of image processing. The ranking means then selects the character data that appears most frequently among the character data matched to any partial image as the character data corresponding to that partial image. The character data to be ranked may include character data matched by the original matching means in addition to the auxiliary character data. The auxiliary character data obtained from the image processing means and auxiliary OCR means is character data obtained by variously emphasizing or suppressing the characteristics of handwritten characters, so even if the data has been transformed in this way, if there are a large number of characters that are recognized, it can be considered to be appropriate character data corresponding to the partial image.

[0016] In the above configuration, the image processing can include any of color conversion, color removal, and size expansion / contraction. Color conversion can remove noise components or highlight features that should be emphasized. Size expansion / contraction can also correct and normalize the size and tilt of handwritten characters. These features can improve the recognition rate of handwritten characters in any partial image.

[0017] As yet another means that can be adopted to solve the problem, in addition to the above-mentioned configuration, the matching means can be provided with a size determination means that excludes from the matching, among the character data matched to the partial image, character data whose area occupying the entire image is smaller than a predetermined size.

[0018] When writing characters by hand, it is difficult to make the characters extremely small because the characters are written by moving the hand using a writing implement. Therefore, the size determination means determines that, among the character data matched by the matching means, character data smaller than a predetermined size is not a handwritten character and excludes it from the matching. This reduces the risk of misrecognition.

[0019] As yet another means that can be adopted to solve the problem, in addition to the above configuration, it is also possible to provide a library in which predetermined character strings are recorded, and when character data corresponding to any partial image obtained by the matching means is similar to a character string in the library, the character data can be matched with the character string in the library.

[0020] Even if the matched character data accurately recognizes handwritten characters, if the handwritten characters are written in a style unique to the writer or contain typos or omissions, the data will be accurately recognized as is, resulting in inconsistent writing. Therefore, by preparing a library in which predetermined character strings are recorded and configuring the system to match character data with character strings in the library, it is possible to standardize the notation of recognized character data regardless of the writer.

[0021] Yet another means that can be adopted to solve the problem is a character recognition method that includes an image input step of inputting an image containing handwritten characters, an OCR step of recognizing the characters contained in the image to obtain character data, an extraction step of extracting a partial image containing handwritten characters from the image, and a matching step of comparing the partial image with the character data and matching those with similar features, and character data corresponding to any partial image can be obtained by the matching step.

[0022] The character recognition method is configured as follows: an image containing handwritten characters is input in an image input step, and the characters are recognized in an OCR step to obtain character data; a partial image containing handwritten characters is extracted in an extraction step, and the character data and partial image are compared, and characters with similar features are matched together, thereby selecting appropriate character data from multiple character data that corresponds to the features of the partial image.

[0023] Still another means that can be adopted to solve the problem includes an image processing step of applying different image processing to the image to obtain a plurality of processed images, and an auxiliary OCR step of recognizing characters contained in the image from the plurality of processed images to obtain a plurality of auxiliary character data, wherein the matching step compares the partial image with character data including the plurality of auxiliary character data to match those with similar features, and a ranking step of adopting the character data with the largest number of characters among the character data matched to any partial image as the character data corresponding to that partial image.

[0024] Characters are recognized in an auxiliary OCR step for a plurality of processed images to obtain auxiliary character data, and the character that is recognized most frequently in the ranking step is adopted as the character data corresponding to that partial image, thereby enabling accurate selection of character data corresponding to handwritten characters in any partial image.

[0025] As a means that can be adopted to solve the problem, the above-mentioned means can be adopted as a location map creation system for creating an investigation location map based on a ground investigation. The means for this purpose is equipped with the above-mentioned character recognition system, and the image is a site map with a blank layout plan and handwritten measurement information, the OCR means is a means for recognizing characters contained in the site map to obtain character data, the extraction means is a means for extracting a partial image containing handwritten characters from the site map, and the matching means is a means for comparing the partial image with the character data and matching those with similar features, and the character data matched by the matching means can be configured to be displayed near the position corresponding to the handwritten character portion on the layout plan.

[0026] In the ground survey of a detached house, etc., a survey location map is created from a site map drawn by hand on-site. At this time, the original blank layout map and the site map with handwritten measurement information are input, and the character data obtained by the OCR means and the partial image obtained by the extraction means are matched by the matching means, so that accurate character data corresponding to the handwritten characters can be displayed near the position corresponding to the handwritten character portion on the layout map. [Effects of the Invention]

[0027] The character recognition system, character recognition method, and location map creation system of the present invention recognize character data from an image containing handwritten characters, while also employing a means for extracting handwritten character portions from the image and comparing the character data with the partial image to match those with similar features. This not only makes it possible to accurately recognize handwritten characters and obtain accurate character data, but also has the effect of shortening the time required for recognition. [Brief explanation of the drawings]

[0028] [Figure 1] 1 is an explanatory diagram illustrating the overall configuration of a character recognition system according to the present invention; [Figure 2] 1 is a flowchart showing a processing flow of the character recognition system of the present invention. [Figure 3]FIG. 2 is an explanatory diagram showing the contents of an OCR means in the character recognition system of the present invention. [Figure 4] FIG. 2 is an explanatory diagram illustrating the contents of an extraction unit in the character recognition system of the present invention. [Figure 5] FIG. 2 is an explanatory diagram illustrating the contents of a classification unit in the character recognition system of the present invention. [Figure 6] FIG. 2 is an explanatory diagram illustrating the contents of a matching means in the character recognition system of the present invention. [Figure 7] 1 is an explanatory diagram showing a location map creation system using a character recognition system of the present invention; [Figure 8] FIG. 10 is an explanatory diagram illustrating a modified example of the character recognition system of the present invention. [Figure 9] FIG. 10 is an explanatory diagram showing the contents of an auxiliary OCR means, an image processing means, and a ranking means in a modified example of the present invention. [Figure 10] 10A and 10B are explanatory diagrams showing types of image processing in a modified example of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0029] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to Figures 1 to 7. In the following description, each figure is depicted schematically for the sake of simplicity. The character recognition system (hereinafter simply referred to as the system) 100 of the present invention is a system for recognizing handwritten characters from an image containing handwritten characters and converting the characters into character data. There are various types of images containing handwritten characters, such as drawings with notes written in pen and documents such as specifications. Furthermore, the processing order of each means and step described below is not limited to the order described, and they may be rearranged or other means or steps may be added.

[0030] First, the overall configuration and processing flow of the present system 100 will be described with reference to Figures 1 and 2. Image input means 1 is a block for performing image input step S1, and is a block for reading an image G including handwritten characters based on an operation by a user U and displaying the image G on a terminal P. These documents are handwritten on paper and are therefore scanned using a scanner S or the like. The file format of the scanned image G may be an image format such as JPEG (Joint Photographic Experts Group) or PNG (Portable Network Graphics), or may be a document file such as PDF (Portable Document Format) that can contain image data. An image G read by an image input means 1 is sent to a server C on the cloud via a network N, for example. A program that performs the following processes of an OCR means 2, an extraction means 3, a classification means 4, and a matching means 5 runs on the server C on the cloud. The results of processing by this program, such as the read image G, are displayed on a terminal P in a timely manner via the network N. This allows a user U to visually confirm the results of each process.

[0031] The OCR means 2 is a block that performs the OCR step S2, and is a block that recognizes all characters, including handwritten characters, from the image G read by the image input means 1 to obtain character data T·T . . . Image G contains handwritten characters as well as pre-written typed characters and graphics. As shown in Fig. 3, OCR means 2 performs OCR processing on the entire image G using a predetermined OCR engine. To improve the overall processing speed of system 100, it is desirable to perform this OCR processing on the entire image G only once, but it may also be performed multiple times.

[0032] The extraction means 3 is a block that performs the extraction step S3, and is a block that extracts only the handwritten character portion from the image G to obtain partial images g·g . . . separately from the OCR means 2. Methods for extracting partial images g·g... of handwritten characters from image G include extraction based on image features such as edge smoothness and line fluctuation, as well as utilizing a deep learning model with a neural network trained using large amounts of data on handwritten and typed characters. The method in Figure 4 uses a deep learning model to extract partial images g·g... of handwritten characters.

[0033] The classification means 4 is a block that performs classification step S4, and is a block that makes preparations to enable character normalization in the matching means 5, which will be described later, based on the results of classifying the partial images g·g... extracted by the extraction means 3 by feature. In other words, character normalization is performed in advance in order to perform more appropriate matching in the matching means 5, which will be described later. For example, as shown in Figure 5, items that share common features such as the same color of written characters, a string of characters, a number, a circle, or a symbol with a number next to it are classified. Items that share a single feature may be classified, or items that share multiple features may be further subdivided and classified.

[0034] Incidentally, by providing the classification means 4, it is possible to improve the accuracy of matching in the matching means 5, which will be described later, but if matching can be performed without problems, it is also possible to configure without providing the classification means 4. If the classification means 4 is not provided, it is possible to improve the overall processing speed.

[0035] The matching means 5 is a block that performs the matching step S5, and matches the character data T·T... obtained by the OCR means 2 with the partial images g·g..., and associates the appropriate character data T with each handwritten character. The simplest method, as shown in Figure 6, is to match the coordinate values ​​of partial images g·g... containing arbitrary handwritten characters with the coordinate values ​​of the entire image G held by character data T·T... obtained by OCR means 2 that are close to each other.

[0036] As an example of matching, for the circled handwritten character "-510," first search for partial image g with coordinate values ​​close to the coordinate values ​​where the character was read. At this time, because of the circled feature in partial images g·g..., it can be determined that the circle is additional and that the "-510" inside is the target to be output as character data. Therefore, from the character data T·T... obtained by OCR means 2, it can be assumed that "-510" is the appropriate character data, and these can be linked.

[0037] As another example, for handwritten characters consisting of "Kuusen" (airplane) and "H=6.0m" arranged vertically in two lines, first, a partial image g with coordinate values ​​close to the coordinate values ​​at which the characters were read is searched for. Since partial images g·g... are character strings and have multiple lines, it can be determined that each character string is to be displayed in two lines. Therefore, from the character data T·T... obtained by OCR means 2, the character data "Kuusen" and the character data "H=6.0m" are selected, and the combination of these in two lines vertically is considered to be the appropriate character data, and these are linked.

[0038] As another example, in the case of a symbol "●" with the number "1,700" to the upper right and the number "3" to the lower right, when the handwritten character "3" is extracted as partial image g, it is extracted including the symbol "●" and the typed characters "1,700." In this example, there is no character data corresponding to "●" in the character data T·T... obtained by OCR means 2, so "●" is not matched. On the other hand, the character data T·T... obtained by OCR means 2 includes the character data "1,700", so it is matched with the "1,700" part in partial image g.

[0039] Therefore, the size determination means 51 executes the size determination step S51. When the area (number of pixels) occupied per character in the "1,700" part of the partial image g is equal to or less than a predetermined size with respect to the entire image G, the size determination means 51 is a block that excludes this from matching on the assumption that it is not a handwritten character. As a result, even when the partial image g including handwritten characters and typed characters is matched with the character data T·T..., only the handwritten characters can be matched.

[0040] In addition to the method of matching by coordinates, as an auxiliary method, normalization of characters can be performed using the result of the classification means 4 described above and then matching can be performed. For example, when matching coordinate values, even if a certain character data T is determined to match the partial image g representing a measurement point, the character data T may be a full-width character based on the result of character normalization. In this case, it is determined that this character data T should not be matched with the partial image g of the measurement, and then it is matched with the partial image g determined to be appropriate next.

[0041] As a specific example, for those classified as measurement points or dimension lines, even if characters other than numbers are matched by coordinate values, they are removed as having no relation to the measurement points or dimension lines. Also, for those classified as GL (level), even if the symbol at the beginning of the word is unclear, examples include correcting so that the beginning of the word is always one of +, -, ±, correcting to +10 if it is f10 or t10, or correcting to ±10 if it is 土10 (10 in the Chinese character ツチ). As a result, even when characters and symbols are dense in the image G, matching can be performed with high accuracy.

[0042] By using the character recognition system 100 of the present invention configured as described above, a position map creation system 1000 as shown in FIG. 7 can be configured. In the location map creation system 1000 of the present invention, the image G read by the image input means 1 is a site map G1 on which handwritten notes are written showing the locations where screw weight penetration tests were conducted and the results of those tests in a preliminary ground survey conducted at a house construction site.

[0043] In the location map creation system 1000 of the present invention, the image input means 1 may be configured to read in the site map G1 and also read in the original layout map G2 before handwriting. Here, by loading the layout drawing G2 together with the site drawing G1, the reference in the layout drawing G2 can be aligned with the reference in the site drawing G1, and the size and inclination of the loaded site drawing G1 can be calibrated. In this calibration process, the user U may manually align the reference points, or the feature amounts of both images may be automatically recognized and aligned.

[0044] When the size and inclination of the site map G1 are adjusted by the calibration, character data T·T... are obtained by the OCR means 2. At this time, in addition to the character data T·T..., symbols for gutters, antennas, slopes, etc. may also be read. The character data T·T... obtained here includes character data resulting from reading handwritten characters such as handwritten measurements written at the ground survey site and notes indicating gutter locations, but also typewritten characters such as dimensions displayed on the original layout drawing G2. Therefore, all characters, including handwritten characters and typewritten characters, are obtained as character data T·T....

[0045] For the calibrated site map G1, the extraction means 3 and classification means 4 are applied to extract the portions containing the handwritten characters and obtain partial images g·g... that are classified by their features. In the classification means 4, a partial image g containing a "●" symbol and a number adjacent to it is classified as indicating a ground investigation position. Also, a circled number preceded by a symbol such as "-" or "±" is classified as indicating a height relative to a reference position. In this way, partial images g·g... with similar features are classified as partial images g·g... of the same category.

[0046] Then, the matching means 5 matches the optimal character data T·T... obtained by the OCR means 2 for each of the partial images g·g.... At this time, matching is performed taking into consideration the common characteristics of the items classified by the classification means 4. For example, when a "●" is adjacent to a number, the "●" is a symbol that simply indicates the survey location, and character data T·T... is not matched to this. However, it is processed so that it is displayed adjacent to the number as the symbol "●". These pieces of character data T·T... are displayed in the vicinity of the positions of the partial images g·g... of the handwritten character portion on the site map G1.

[0047] Furthermore, numbers enclosed in circles with a "-" or "±" at the beginning of the number represent height relative to a reference point, and therefore an indicator line is provided from the circle to the target position. In this way, the matching means 5 may not only simply match the partial images g·g . . . with the character data T·T . . . but may also arrange symbols and figures according to the classified features.

[0048] The character data T·T... matched as described above is displayed near the position corresponding to the handwritten character portion on the layout drawing G2. For example, the character data T of "+90" is displayed on the layout drawing G2 at the same coordinates as the part where "+90" was handwritten on the site drawing G1. Once matching has been performed on all partial images g·g..., the appropriate character data T·T... corresponding to the handwritten characters is placed in the appropriate position on the layout map G2, and can be output as the survey location map A. The matching means 5 can reduce mismatches by taking into account the compatibility of the characters in addition to the coordinate values ​​when matching. If certain character data T contains a high proportion of full-width characters, it is determined that the compatibility with the GL (level) partial image g is low and it is excluded from matching (because GL is often half-width).

[0049] When the survey location map A has character data T·T... placed on it, it is possible to add or modify the content of the character data T·T..., or manually adjust the position of the placed symbols and character data T·T.... At this time, it is preferable to display the blank layout map G2 on the left side of the display screen of the terminal P where the work is being performed, and the survey location map A created by the location map creation system 1000 of the present invention on the right side, so that the correction work can be performed while checking the differences from the state before it was handwritten (not shown).

[0050] After the correction work was completed, the output survey location map A was a drawing showing the results of the ground survey in comparison with the layout map G2, with the measurement results shown in easily visible typed characters. In this way, a drawing can be created in a so-called clean-up state without having to be manually input by a human being one by one.

[0051] As described above, with the character recognition system 100 and survey location map creation system 1000 of the present invention, the character data T·T... obtained by the OCR means 2 can be accurately matched by the matching means 5 with the partial images g·g... of handwritten character portions extracted by the extraction means 3 and classified by characteristics by the classification means 4. By performing the OCR process, which requires a lot of processing time, in a fewer number of times while performing the process of matching handwritten character images with character data, it is possible to achieve both a high recognition rate for handwritten characters and a reduction in the time required for recognition.

[0052] "Variation 1" Next, a character recognition system 101 according to a modified example of the present invention will be described with reference to Figures 8 to 10. In the following description, the same parts will be denoted by the same reference numerals, and duplicated descriptions will be omitted.

[0053] This modified example differs from the configurations of FIGS. 1 to 7 in that it includes an image processing means 6, an auxiliary OCR means 7, and a ranking means 8. As shown in FIG. 9, the image processing means 6 is a block for performing a plurality of different image processes on the image G read by the image input means 1 to obtain a plurality of processed images p·p . . . The auxiliary OCR means 7 is a block for recognizing characters from a plurality of processed images p·p . . . to obtain auxiliary character data t·t . The ranking means 8 is a block that determines the largest number of matched character data T·T... and auxiliary character data t·t... among a plurality of auxiliary character data t·t....

[0054] As mentioned above, the OCR means 2 recognizes all characters as character data T·T.... In this case, for example, adjacent numbers may be recognized as individual character data, or the adjacent numbers may be recognized as a series of characters, and either of these character data will be obtained as the result of the OCR means 2. In such a case, the matching means 5 may match the wrong character data with any handwritten character.

[0055] Therefore, new image processing is applied to the original image G by the image processing means 6. The image processing involves multiple separate processes. An example of image processing is shown in Figure 10. For example, color conversion processing may be performed to convert a color image G into grayscale, or into any of Red / Green / Blue or other color scales. Also, color removal processing may be performed, such as removing only the black portion or only the red portion. Furthermore, the aspect ratio of the entire image G may be changed, and processing may be performed to stretch the image vertically or horizontally. The types of image processing are not limited to these, and other processing may be used. These processing may be performed alone or in combination.

[0056] Next, the auxiliary OCR means 7 performs OCR processing on the processed images p·p... that have been subjected to image processing by the image processing means 6 in this manner, thereby obtaining auxiliary character data t·t.... The number of pieces of auxiliary character data t·t... is equal to the number of types of image processing. For example, if there are nine types of image processing, at least nine pieces of auxiliary character data t·t... are obtained for one character (or character string). In this case, for example, the auxiliary character data t obtained by the auxiliary OCR means from the processed image p that has been converted to grayscale may be different from the auxiliary character data t obtained by the auxiliary OCR means from the processed image p that has been vertically stretched. In this way, the multiple pieces of auxiliary character data t·t... obtained by the auxiliary OCR means from the multiple processed images p·p... may each be different character data.

[0057] Therefore, the ranking means 8 adopts the character data with the largest number among the different auxiliary character data t·t... that match with the partial images g·g... and the character data T·T... obtained from the original image G by the OCR means as the appropriate character data for that partial image g·g.... For example, as shown in Figure 9, if the results of the OCR processing means 2 for the original image P and the results of the auxiliary OCR means 8 based on multiple processed images p·p... are "10", "2", "1002", "2", "02", "2", "02", "2", the most common character data is "2" with 4 votes. Therefore, the appropriate character data for this partial image g is "2". If the number of votes is the same, the one with the higher average reliability in the OCR process should be used. Reliability can be calculated using various methods, but for example, in the case of a model that predicts a character string at once, it can be calculated using the likelihood (probability) for the entire character string.

[0058] As described above, the system 101 can recognize handwritten characters more accurately without significantly increasing processing time, even when the image contains characters that may cause symbols to be recognized as characters or may be recognized as different characters.

[0059] "Variation 2" A character recognition system 102 according to yet another modification of the present invention may be configured to include a library L in which predetermined character strings are recorded in advance. The library L in this modified example is a so-called name dictionary, and stores, for example, the following character strings: "north retaining wall," "west retaining wall," "east retaining wall," and so on. Taking the example of "north retaining wall," in Library L, the string "north retaining wall" after name matching is linked to strings such as "northern you" and "kita you."

[0060] When the matching means 5 performs matching, if the matched character data T is identical to or similar to the character string before the name identification, the library L performs a process of replacing it with the character string after the name identification. In the above example, if "Kitayou" is handwritten on image G, after processing by the matching means 5, "Kitayou" is replaced with "north retaining wall." In this way, provision of library L not only unifies terminology and makes documents more accurate, but also improves visibility. Name matching using library L can be performed not only by the matching means 5, but also immediately after obtaining character data by the OCR means 2, and can be performed at any step in the implementation steps.

[0061] The present invention is not limited to the above-described embodiment, and can be modified as appropriate within the scope of the claims of the present invention. For example, since soil surveys are often carried out at the four corners of a house and their centers, the four corners of the symbol indicating the survey location may be automatically aligned in terms of vertical and horizontal coordinates. In addition, the numbers adjacent to the symbol indicating the investigation position are considered to represent the investigation order, which is often performed clockwise. Therefore, even if the OCR means misrecognizes the numbers adjacent to the investigation position symbol, it is possible to automatically correct them so that they start from 1 and proceed clockwise in ascending order. Furthermore, the character recognition system of the present invention can be applied not only to site drawings at construction sites, but also to any other images that contain handwritten characters. [Explanation of symbols]

[0062] 100,101 Character Recognition System 1000 Location Map Creation System 1. Image input means 2 OCR means 3 Extraction means 4 Classification means 5 Matching Methods 51 Size determination method 6 Image processing methods 7 Auxiliary OCR means 8 Ranking Methods C. Cloud server G Image G1 Site Map G2 layout diagram g Partial image N Network P terminal p processed image U User S scanner T character data t Supplementary character data

Claims

1. an image input means for inputting an image including handwritten characters; an OCR means for recognizing characters contained in the image to obtain character data; an extraction means for extracting a partial image including handwritten characters from the image; a matching means for comparing the partial image with the character data and matching those having similar features with each other; A character recognition system characterized in that character data corresponding to any partial image is obtained by said matching means.

2. image processing means for applying different image processing to the image to obtain a plurality of processed images; an auxiliary OCR means for recognizing characters contained in the plurality of processed images and obtaining a plurality of auxiliary character data; the matching means compares the partial image with the character data including the plurality of auxiliary character data and matches those having similar characteristics with each other; 2. The character recognition system according to claim 1, further comprising a ranking means for selecting the character data that is most numerous among the character data matched to an arbitrary partial image as the character data corresponding to that partial image.

3. 3. The character recognition system according to claim 2, wherein the image processing includes any one of color conversion, color removal, and size expansion / contraction.

4. 3. The character recognition system according to claim 1, further comprising a size determination unit configured to exclude, from the character data matched to the partial image, character data whose area occupying the entire image is smaller than a predetermined size.

5. It has a library of predetermined character strings, 3. The character recognition system according to claim 1, wherein when character data corresponding to any partial image obtained by said matching means is similar to a character string in said library, said character data is matched with the character string in said library.

6. an image input step of inputting an image including handwritten characters; an OCR step of recognizing characters contained in the image to obtain character data; an extraction step of extracting a partial image including handwritten characters from the image; a matching step of comparing the partial image with the character data and matching those having similar features, A character recognition method characterized in that character data corresponding to an arbitrary partial image is obtained by the matching step.

7. an image processing step of obtaining a plurality of processed images by applying different image processing to the image; an auxiliary OCR step for recognizing characters contained in the plurality of processed images to obtain a plurality of auxiliary character data; the matching step compares the partial image with character data including the plurality of auxiliary character data and matches those having similar characteristics with each other; 7. The character recognition method according to claim 6, further comprising a ranking step of adopting the character data that is most numerous among the character data matched to any partial image as the character data corresponding to that partial image.

8. A location map creation system for creating a survey location map based on a ground survey, A character recognition system according to claim 1 or claim 2, The image is a blank layout plan and a site plan with handwritten measurement information; the OCR means is a means for recognizing characters included in the site map to obtain character data; the extraction means is means for extracting a partial image including handwritten characters from the site map, the matching means is means for comparing the partial image with the character data and matching those having similar features with each other; A location map creation system characterized in that the character data matched by said matching means is displayed near a position on said layout map corresponding to the handwritten character portion.

Images