Character recognition method and ballot reader
The character recognition method addresses the challenge of reading multi-line handwritten strings on ballot papers by combining characters across lines, enhancing accuracy and preventing misinterpretation in vote counting.
Patent Information
- Application Number
- JP2024110889
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-07-10
- Publication Date
- 2026-01-23
AI Technical Summary
Conventional character recognition methods and ballot reading devices struggle with reading handwritten character strings spanning multiple lines on ballot papers, requiring labor-intensive registration and potentially misinterpreting the content due to line breaks, leading to incorrect vote apportionment.
A character recognition method that identifies whether the handwritten string is one or multiple lines and combines characters across lines to form a single string, using photoelectric conversion and image processing to accurately read and recognize multi-line inputs.
Enables easy reading of multi-line handwritten characters on ballot papers, preventing misinterpretation and ensuring accurate vote counting by connecting and recognizing characters across lines.
Smart Images

Figure 2026010852000001_ABST
Abstract
Description
[Technical Field]
[0001] The present invention relates to a character recognition method and a ballot reading device, and more particularly to a character recognition method for handwritten characters and a ballot reading device that uses the character recognition method. [Background technology]
[0002] A known conventional character recognition method is, for example, the character recognition method described in Patent Document 1. In this character recognition method, a handwritten character string written on a ballot paper is read using a photoelectric conversion means, the read character string is compared with a list of candidate names or a list of political party names, and the candidate name or political party name with the highest match rate between the handwritten character string and the character string of the candidate name included in the list of candidate names or political party names is determined as the recognition result of the handwritten character string.
[0003] Furthermore, a known conventional ballot paper reading device is, for example, the ballot paper reading device described in Patent Document 2. This reading device has an imaging device that captures an image of the writing surface of the ballot paper and information processing means that processes the captured image information, and the information processing means determines the content written on the ballot paper. [Prior art documents] [Patent documents]
[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2002-150222 [Patent Document 2] Japanese Patent Application Laid-Open No. 2012-23550 Summary of the Invention [Problem to be solved by the invention]
[0005] However, with the character recognition method disclosed in Patent Document 1 and the ballot reader disclosed in Patent Document 2, if a handwritten string of characters spanning multiple lines is written on a ballot paper bearing the names of candidates or political parties, only one line of the handwritten string is read. In order to compare this one line of handwritten string with a list of candidate names or a list of political parties, it is necessary to register the candidate names or political party names divided at line break positions in the list of candidate names or political party names, corresponding to all expected line break positions, which poses a problem of a large amount of labor required for the registration process.
[0006] In addition, depending on the candidate name or party name and the position of the line breaks in the handwritten string on the ballot paper, there was a problem in that even if the content of the entire handwritten string was not the content for which the vote was to be apportioned, the content of the handwritten string that was read could match the content of the target for which the vote was to be apportioned, and the vote could be recognized as being the target for apportionment.
[0007] This invention has been made to solve such problems, and aims to provide a character recognition method that makes it possible to easily read handwritten character strings written across multiple lines on a ballot paper by connecting and reading the handwritten character strings, and that prevents votes whose original contents in the handwritten character string are not subject to apportionment from being recognized as votes that are subject to apportionment. [Means for solving the problem]
[0008] In order to solve the above problems, the character recognition method of the present invention is a character recognition method that reads handwritten characters written on a ballot paper via a photoelectric conversion means, and includes: a first step of reading the handwritten characters written on the ballot paper via a photoelectric conversion means; a second step of identifying whether the read character string read via the photoelectric conversion means has one line or multiple lines; a third step of recognizing the characters of the read character string as is if the read character string has one line; and a fourth step of combining the characters of each successive line of the read character string to form a single combined character string and recognizing the characters of the combined character string if the read character string has multiple lines.
[0009] Furthermore, the fourth step may include a fifth step of combining the multiple lines of the read character strings into a single combined character string, with the line at one end of the read character string being the first line and the line at the other end being the last line, and a sixth step of combining the lines of the read character string into a single combined character string, with the line at the other end of the read character string being the first line and the line at the one end being the last line.
[0010] The ballot reading device of the present invention reads the handwritten characters written on the ballot paper using the character recognition method described above. [Effects of the Invention]
[0011] The character recognition method of this invention comprises a second step of identifying whether the number of lines of the read string read via a photoelectric conversion means from handwritten characters written on a ballot paper is one or multiple lines; a third step of recognizing the characters of the read string as is if the read string has one line; and a fourth step of combining the characters of each successive line of the read string to form a single combined string and recognizing the characters of the combined string if the read string has multiple lines.Therefore, by connecting and reading handwritten characters written across multiple lines on a ballot paper, handwritten characters can be easily read, and it is possible to prevent votes whose original content in the handwritten string is not subject to apportionment from being recognized as votes subject to apportionment. [Brief explanation of the drawings]
[0012] [Figure 1] 1 is a configuration diagram showing a ballot paper reading device according to a first embodiment of the present invention. [Figure 2] FIG. 2 is a configuration diagram showing the imaging unit of FIG. [Figure 3] FIG. 1 is a diagram showing the writing surface of a ballot paper used in proportional representation elections. [Figure 4] This is a diagram showing the writing surface of a completed ballot paper in a proportional representation election when the number of lines of writing within the writing box is one. [Figure 5] This is a diagram showing the writing surface of a completed ballot paper in a proportional representation election when the number of lines of writing within the writing box is two. [Figure 6] 10 is a flowchart showing the processing flow of the ballot paper reading device in judgment mode. [Figure 7] This is a diagram showing the writing surface of another example of a proportional representation election in which the number of lines of the written content within the writing box of a completed ballot paper is two. [Figure 8] FIG. 1 is a diagram showing the writing surface of a ballot paper used in a single-seat constituency election. [Figure 9] This is a diagram showing the writing surface of a completed ballot paper in a single-seat constituency election when the number of lines of writing within the writing box is one. [Figure 10] This is a diagram showing the writing surface of a completed ballot paper in a single-seat constituency election, where the number of lines of writing within the writing box is two. [Figure 11] This is a figure showing the writing surface of another example of a completed ballot paper in a single-seat constituency election, where the number of lines of writing within the writing box on the completed ballot paper is two. DETAILED DESCRIPTION OF THE INVENTION
[0013] Embodiment 1 (Ballot reader) A first embodiment of the present invention will now be described with reference to the accompanying drawings. 1 is a configuration diagram showing a ballot paper reading device according to embodiment 1. Ballot paper reading device 1 is provided with a reading device main body 1a and an extended stacker unit 2. Reading device main body 1a is provided with a hopper 10, main body side transport means 11, first to fifth stackers 12 to 16, an imaging unit 17, and display means 18.
[0014] The hopper 10 is a storage section in which multiple ballot papers 3 inserted from outside are stored. The main body transport means 11 is composed of multiple transport rollers 11a and forms a first internal transport path 111 that transports the ballot papers 3 from the hopper 10 to the first to fifth stackers 12-16 along a transport direction 110. The main body transport means 11 also separates from the first internal transport path 111 at a midpoint 111a and forms a second internal transport path 112 that transports the ballot papers 3 toward the expansion stacker unit 2. The imaging unit 17 generates image information showing the edge of the ballot paper 3, as described below, and is located downstream of the hopper 10 and upstream of the midpoint 111a along the transport direction 110. The display means 18 is a liquid crystal display that displays information to the user.
[0015] The extended stacker unit 2 is an extended unit that can be separated from the reading device main body 1a. The extended stacker unit 2 is provided with an extended-side conveying means 20 and sixth to thirteenth stackers 21-28. The extended-side conveying means 20 forms an extended-side conveying path 200 that conveys ballot papers 3 conveyed to the extended stacker unit 2 through the second internal main body conveying path 112 to the sixth to thirteenth stackers 21-28.
[0016] Stackers 12-16 and 21-28 are cases in which ballot papers 3 classified as described below are stored. While this first embodiment shows an example in which one expansion stacker unit 2 is connected to the reading device main body 1a, multiple expansion stacker units 2 may be connected to the reading device main body 1a depending on the number of stackers required. Furthermore, while the expansion stacker unit 2 has been described as being separable from the reading device main body 1a, the reading device main body 1a and expansion stacker unit 2 may be provided integrally with each other.
[0017] Next, Fig. 2 is a configuration diagram showing the imaging unit 17 of Fig. 1. The imaging unit 17 is provided with first and second guide plates 170, 171, first to third roller sections 172-174, first and second imaging means 175, 176, and first and second opposing rollers 177, 178.
[0018] The first and second guide plates 170, 171 are plate members extending along the conveying direction 110, and are arranged opposite each other across the first internal main body conveying path 111. These first and second guide plates 170, 171 guide the ballot papers 3 as they pass through the imaging unit 17. Although not shown in detail, the first and second guide plates 170, 171 each have openings at the positions where the aforementioned first to third roller units 172-174 are arranged.
[0019] The first roller unit 172 is composed of a pair of rollers 172a arranged on either side of the first main body conveying path 111 so that their outer circumferential surfaces come into contact with each other through the openings of the first and second guide plates 170, 171. The second and third roller units 173, 174 are also composed of a pair of rollers 173a, 174a, similar to the first roller unit 172. These first to third roller units 172-174 are arranged spaced apart from each other along the conveying direction 110, and feed (convey) ballot papers 3 inserted between the rollers 172a-174a along the conveying direction 110. In other words, the first to third roller units 172-174 constitute part of the main body conveying means 11.
[0020] Here, one end face of the first internal transport path 111 on the side where the first guide plate 170 is located is referred to as the transport path front face 113, and the other end face on the side where the second guide plate 171 is located is referred to as the transport path back face 114. In other words, when the ballot 3 passes through the imaging unit 17, the end faces of the first internal transport path 111 where the first and second end faces 3a, 3b of the ballot 3 face each other are referred to as the transport path front face 113 and the transport path back face 114. Depending on the orientation of the ballot 3, the first end face 3a is either the front (written side) or back face of the ballot 3, and the second end face 3b is the other of the front (written side) or back face. Furthermore, if the front and back directions and the top and bottom directions of the ballot 3 are not aligned with each other, there are four possible orientations for the ballot 3: front-to-top, front-to-bottom, back-to-top, and back-to-bottom.
[0021] The first imaging means 175 is disposed between the first roller unit 172 and the second roller unit 173 so as to face the conveyance path surface 113 through the opening in the first guide plate 170. The first imaging means 175 has a light source unit 175a and an imaging unit 175b, and generates first image information 175c showing the first edge 3a of the ballot paper 3 as the ballot paper 3 passes through the imaging unit 17. Specifically, as the ballot paper 3 passes through the imaging unit 17, imaging light is irradiated from the light source unit 175a onto the first edge 3a of the ballot paper 3, and the imaging light reflected by the first edge 3a is received by the imaging unit 175b, thereby generating the first image information 175c.
[0022] The first opposing roller 177 is disposed between the first roller unit 172 and the second roller unit 173 so as to face the imaging side end face of the first imaging means 175 through the opening in the second guide plate 171. The first opposing roller 177 is disposed close to the imaging side end face of the first imaging means 175 so as to send the ballot paper 3 along the conveying direction 110 when the ballot paper 3 enters between the outer circumferential surface of the first opposing roller 177 and the imaging side end face of the first imaging means 175.
[0023] The second imaging means 176 is disposed between the second roller unit 173 and the third roller unit 174 so as to face the conveyance path rear surface 114 through the opening of the second guide plate 171. That is, the second imaging means 176 is disposed downstream of the first imaging means 175 along the conveyance direction 110 and does not face the first imaging means 175. The second imaging means 176 has a light source unit 176a and an imaging unit 176b, similar to the first imaging means 175 described above, and generates second image information 176c that shows the second edge surface 3b of the ballot paper 3, which is opposite the first edge surface 3a. Both the first image information 175c and the second image information 176c represent images using a collection of dots arranged two-dimensionally.
[0024] The second opposing roller 178 is disposed between the second roller unit 173 and the third roller unit 174 so as to face the second imaging means 176 through the opening of the first guide plate 170. Similar to the first opposing roller 177, the second opposing roller 178 is disposed close to the imaging side end face of the second imaging means 176.
[0025] An information processing means 4 is connected to the first and second imaging means 175, 176. This information processing means 4 is composed of a computer including a storage unit that stores information such as programs and an arithmetic unit that performs processing operations based on the information in the storage unit. In detail, the information processing means 4 is composed of a computer built into the reading device main body 1a, or a combination of a computer provided separately from the reading device main body 1a and a computer built into the reading device main body 1a. As will be described later, this information processing means 4 processes the first and second image information 175c, 176c from the first and second imaging means 175, 176.
[0026] (Character recognition method for proportional representation elections) Next, a method for character recognition on a ballot paper for a proportional representation election using the ballot paper reading device 1 of the first embodiment will be described. FIG. 3 shows the writing side of a ballot paper 3 used in a proportional representation election in which political parties and other political organizations are candidates. This ballot paper 3 is a blank ballot paper 31 that has not yet been filled out by a voter. The blank ballot paper 31 is rectangular, with the direction of arrow Y, which is the longitudinal direction, being the vertical direction, and the direction of arrow X, which is perpendicular to the direction of arrow Y, being the horizontal direction. Predetermined printed information is printed on the writing side of the blank ballot paper 31. This printed information includes printed characters 311 and entry boxes 312. The printed characters 311 represent, for example, the name of the election or notes, and the entry boxes 312 define a writing area in which voters should write the name or abbreviation of the political party or other political organization for which they are voting in a proportional representation election. The entry boxes 312 are rectangular and are vertically long, with the direction of arrow Y, which is the vertical direction of the blank ballot paper 31, being the longitudinal direction.
[0027] The background color of blank ballot papers 31 used in proportional representation elections is colored white to prevent voters from misidentifying the election for which they are voting. Note that the background color of blank ballot papers 31 is one example, and other specific colors may also be used. Furthermore, while the printed matter is colored black, for example, it may be colored in a color other than black as long as it is distinguishable from the background color of blank ballot paper 31 in first and second image information 175c, 176c.
[0028] (Proportional Representation Election Study Mode) The ballot reading device 1 of the present embodiment 1, when operating in a known learning mode, uses a blank ballot 31 to learn the positions of the printed characters 311 and the areas of the entry boxes 312 on the blank ballot 31. The operation of the learning mode will be explained below. After setting the operating mode of the ballot reader 1 to learning mode, the user places a blank ballot 31 into the hopper 10 (see FIG. 1). Next, the first and second imaging means 175, 176 (see FIG. 2) generate first and second image information 175c, 176c, respectively, representing corresponding sides of the blank ballot 31.
[0029] Next, the information processing means 4 identifies the front and back of the blank ballot 31 based on the first and second image information 175c, 176c from the first and second imaging means 175, 176. Here, the information processing means 4 detects the entry box 312 from the front surface, which is either the first or second end surface 3a, 3b, based on the first and second image information 175c, 176c from the first and second imaging means 175, 176. The front and back are then identified by determining whether the entry box 312 was detected from the first end surface 3a or the second end surface 3b. This identifies which of the first and second image information 175c, 176c the writing side of the blank ballot 31 corresponds to.
[0030] Next, the information processing means 4 identifies the top-bottom orientation of the blank ballot paper 31 based on the first image information 175c or the second image information 176c. This is done based on the orientation of the printed characters 311, for example.
[0031] (Proportional Representation Election Judgment Mode) Next, we will explain the operation of the ballot paper reading device 1 in the determination mode. This determination mode is a mode in which handwritten characters written by the voter on the blank ballot paper 31 are recognized and determined.
[0032] Figure 4 is a diagram showing the writing surface of a completed ballot paper 31a, which is a ballot paper on which a voter writes the information by hand on a blank ballot paper 31 (see Figure 3), in a proportional representation election, when the number of lines of content to be written in the writing box 312 is one. Figure 5 is a diagram showing the writing surface of a completed ballot paper 31a, which is a ballot paper on which a voter writes the information by hand on a blank ballot paper 31 (see Figure 3), in a proportional representation election, when the number of lines of content to be written in the writing box 312 is two. Note that the names or abbreviations of political parties and other political organizations, or candidate names including personal names, etc., described below are used for the purpose of explaining the present embodiment 1, and are unrelated to actual political parties, other political organizations, or individual names.
[0033] Figure 6 is a flowchart showing the processing flow of the ballot reading device 1 in the determination mode. After setting the operating mode of the ballot reading device 1 to the determination mode, the user places a completed ballot paper 31a into the hopper 10. There may be multiple completed ballot papers 31a, and the processing of Figure 6 is performed for each completed ballot paper 31a.
[0034] First, as in the learning mode, the first and second imaging means 175, 176 generate first and second image information 175c, 176c, respectively, representing corresponding sides of the completed ballot paper 31a (step S1). Next, as in the learning mode, the information processing means 4 identifies the front and back of the completed ballot paper 31a (step S2). In the following example, we will explain the case where the first image information 175c corresponds to the writing side. Note that the operation is similar when the second image information 176c corresponds to the writing side.
[0035] The information processing means 4 determines whether the written content (handwritten characters) within the entry box 312 detected in the first image information 175c can be read (step S3). If it is determined that the written content can be read, the information processing means 4 reads the written content (step S4) and identifies the top-to-bottom orientation of the completed ballot paper 31a (step S5). The top-to-bottom orientation is identified, for example, based on the orientation of the printed characters 311 and the orientation of the characters of the written content. At this point, it is identified whether the orientation of the completed ballot paper 31a is front-to-top, front-to-bottom, back-to-top, or back-to-bottom. Note that known handwritten character recognition means can be used as the handwritten character recognition means.
[0036] Also, in step S3, if the contents written in the entry box 312 cannot be read, the information processing means 4 determines that the completed ballot paper 31a is an invalid ballot (step S6), and transports the completed ballot paper 31a to the stacker 16 for invalid ballots or unidentifiable ballots, thereby ending the processing.
[0037] Next, the information processing means 4 determines whether the number of lines of the written content in the entry box 312 is one line or two or more lines (multiple lines) (step S7). Here, the number of lines of the written content is counted along the direction of arrow X, which is the horizontal direction of the entry box 312 shown in Figures 4 and 5. For example, if the written content in the entry box 312 of the completed ballot paper 31a is one line as shown in Figure 4, the information processing means 4 determines that the written content is one line. Also, if the written content in the entry box 312 of the completed ballot paper 31a is two lines as shown in Figure 5, the information processing means 4 determines that the written content is two or more lines. Furthermore, if the written content in the entry box of the completed ballot paper 31a (not shown) is three or more lines, the information processing means 4 similarly determines that the written content is two or more lines.
[0038] If the content written in the entry box 312 is two or more lines, the information processing means 4 sequentially connects the read results of each line of the content and performs character recognition as a single read character string (step S8). For example, the content written in the entry box 312 in FIG. 5 is "ABCD Party," with "AB" written vertically in the first line from the right and "CD Party" written vertically in the second line from the right. In this case, the information processing means 4 sequentially connects the content "AB" in the first line from the right and the content "CD Party" in the second line from the right, and obtains the character string "ABCD Party" as a read character string. Note that while the case where the content written spans two lines has been described here, if the content spans three or more lines, the content of each line is similarly sequentially connected to obtain a read character string.
[0039] If the content written in the entry box 312 is one line, the information processing means 4 performs character recognition on the written content as it is as one read character string (step S9).
[0040] Next, if the content written in the entry box 312 is two or more lines, the information processing means 4 compares the read character string "ABCD Party" connected in step S8 with the candidate list. On the other hand, if the content written in the entry box 312 is one line, the information processing means 4 takes the character string "ABCD Party" of the written content as a read character string and compares it with the registered name and alias in the candidate list (step S10). This candidate list is a list that lists the names of candidates in an election. Table 1 is an excerpt from a candidate list used in the proportional representation election of this embodiment 1. The candidate list lists the candidate numbers assigned to political parties and other political organizations that have registered their candidacy with the election chief, i.e., the candidates, the registered names of the candidates corresponding to the candidate numbers, and the aliases for the registered names.
[0041] [Table 1]
[0042] Generally, when a voter handwrites a candidate's name on a ballot in an election, there are various possibilities, such as writing the candidate's registered name exactly, writing the abbreviation registered by the candidate exactly, writing part or all of the registered name in hiragana, or writing an abbreviation with part of the registered name omitted at the voter's discretion. In elections, even if a voter writes part or all of the registered name in hiragana or abbreviates part of the registered name, if the candidate's name is identifiable, it is generally determined that the vote is valid for that candidate. Therefore, the abbreviation registered by the candidate, the character string of the registered name in hiragana part or all, which is assumed to appear on the ballot, and the character string with part of the registered name omitted are listed as alternative names on the candidate list.
[0043] The information processing means 4 compares the read character string with the registered name and alias in the candidate list and determines whether a matching name is detected (step S11). If a matching name is detected, the information processing means 4 counts the number of votes as votes for the candidate with the candidate number corresponding to the matching name (step S12), and ends the process by transporting the completed ballot paper 31a to one of stackers 12-15, 21-27 according to the contents written on it.
[0044] Next, if it is determined in step S12 that the read character string does not match the registered name and alias in the candidate list, the information processing means 4 collates the read character string with the allocation target list (step S13).
[0045] In general, in elections, if the contents of a ballot correspond to multiple candidates, the ballot may be treated as a pro rata ballot, in which the votes are apportioned among the candidates. In step S13, the read character string of the contents of the ballot is compared with the pro rata ballot list to determine whether it is eligible for such apportionment.
[0046] Table 2 is an excerpt from the allocation target list used in the proportional representation election of this embodiment 1, and lists the allocation target character string and the name of the candidate to which the allocation is to be made that corresponds to this read character string. For example, if the read character string in the entry box on completed ballot paper 31a is "CD Party," "CD," "CD Party," or "CD," the contents of completed ballot paper 31a correspond to both candidate number 1 (ABCD Party) and candidate number 2 (EFCD Party), which shares a partial character string with this candidate name, and therefore the votes related to completed ballot paper 31a are subject to allocation.
[0047] [Table 2]
[0048] The information processing means 4 determines whether or not a character string to be allocated that matches the read character string has been detected as a result of the comparison in step S13 (step S14). If a character string to be allocated is detected, the information processing means 4 counts the vote associated with this ballot as an allocation ticket (step S15), and then has the completed ballot 31a transported to the allocation ticket stacker 28, thereby completing the process.
[0049] Furthermore, if the information processing means 4 does not detect a character string to be allocated that matches the read character string, it determines that the read character string does not correspond to the candidate's name, the candidate's abbreviation or alias, or the character string to be allocated, and therefore determines that the contents of the completed ballot paper 31a are an invalid ballot (step S6), and transports the completed ballot paper 31a to the stacker 16 for invalid or unidentifiable ballots, thereby ending the process.
[0050] Next, the advantages of the present embodiment 1 over conventional techniques in proportional representation elections will be described. Referring again to Figure 5, the contents written in entry box 312 of completed ballot paper 31a are "ABCD Party," with "AB" written vertically on the first line from the right and "CD Party" written vertically on the second line from the right. When the contents written in such entry box 312 are read by a conventional ballot paper reading device, either "AB" on the first line or "CD Party" on the second line is read as the read character string.
[0051] For this reason, conventional ballot paper reading devices can only read one line of the written content in the entry box 312. Therefore, if the line break position of the written content changes, the read character string also changes. For example, if the written content is written over two lines and the content is "ABCD Party," the read result may be any of the following read character strings depending on the line break position: "A," "AB," "ABC," "ABCD," "BCD Party," "CD Party," "D Party," or "Party." If these different read results depending on the line break position were to be covered by aliases in the candidate list as shown in Table 1, the alias column in the candidate list would become too long, which would increase the labor required to register aliases in the candidate list. Furthermore, when considering cases where the names or abbreviations of political parties and other political organizations are partially written in kana, the number of read result patterns that must be covered would increase even further.
[0052] 7 is a diagram showing another example of the writing surface when the number of lines of writing in box 312 of completed ballot paper 31a in a proportional representation election is two. In this example, the writing in box 312 of completed ballot paper 31a is "EFCD Party," with "EF" written vertically on the first line from the right and "CD Party" written vertically on the second line from the right.
[0053] In the case of the completed ballot 31a shown in Figure 5, where the read character string in box 312 is "CD Party" on the second line from the right, and in the case of the completed ballot 31a shown in Figure 7, where the read character string in box 312 is "CD Party" on the second line from the right, the candidate name corresponding to the read character string corresponds to both the ABCD Party and the EFCD Party. Therefore, since the original content written in box 312 of the completed ballot 31a shown in Figure 6 is "ABCD Party," and the original content written in box 312 of the completed ballot 31a shown in Figure 7 is "EFCD Party," the candidate names can be identified, but there is a problem in that votes are subject to proration based on the read character string.
[0054] On the other hand, in the ballot reading device 1 of this embodiment 1, if the writing in the entry box 312 of the completed ballot 31a spans multiple lines, the writing on each line is connected to form the read string, thereby preventing the read string from changing due to changes in the line break position of the written content, and also preventing ballots on which the names of the candidates originally written can be identified from being subject to apportionment due to the line break position of the written content.
[0055] (Character recognition method for single-seat constituency elections) Next, we will explain the character recognition method for ballot papers in single-seat constituency elections using the ballot paper reading device 1 of this embodiment 1. Note that detailed explanations of the same content as the character recognition method for ballot papers in proportional representation elections described above will be omitted.
[0056] FIG. 8 shows the writing side of a ballot paper 32 used in a single-seat constituency election in which individuals are candidates. This ballot paper 3 is a blank ballot paper 32 that has not yet been filled out by a voter. The blank ballot paper 32 is rectangular, with the direction of arrow Y, which is the longitudinal direction, being the vertical direction, and the direction of arrow X, which is perpendicular to the direction of arrow Y, being the horizontal direction. Predetermined printed information is printed on the writing side of the blank ballot paper 32. This printed information includes printed characters 321 and entry boxes 322. The printed characters 321 represent, for example, the name of the election or notes, and the entry boxes 322 define a writing area in which a voter should write the name of the candidate for whom they are voting in a single-seat constituency election. The entry boxes 322 are rectangular and are vertically long, with the direction of arrow Y, which is the vertical direction of the blank ballot paper 32, being the longitudinal direction.
[0057] Blank ballot papers 32 used in single-seat constituency elections are colored yellow to prevent voters from misidentifying the election for which they are using blank ballot papers 32. This coloring of the background of blank ballot papers 32 is one example, and other specific colors may also be used. Furthermore, while the printed matter is colored black, for example, it may be colored in a color other than black as long as it is distinguishable from the background color of blank ballot papers 32 in first and second image information 175c, 176c.
[0058] (Study mode for single-seat constituency elections) The ballot reading device 1 of this embodiment 1 operates in a learning mode similar to that used in proportional representation elections, and uses a blank ballot 32 to learn the positions of the printed characters 321 and the entry boxes 322 on the ballot 32, as well as the front and back and top and bottom orientation of the blank ballot 32.
[0059] (Single-seat constituency election judgment mode) Next, we will explain the operation of the ballot reading device 1 in the determination mode. This determination mode is a mode in which handwritten characters written by the voter on the blank ballot paper 32 are recognized and determined.
[0060] Figure 9 is a diagram showing the writing surface of a completed ballot paper 32a, which is a ballot paper on which a voter has handwritten the information to be written on a blank ballot paper 32 (see Figure 8), in a single-seat constituency election, when the number of lines of writing in the writing box 322 is one, and Figure 10 is a diagram showing the writing surface of a completed ballot paper 32a, which is a ballot paper on which a voter has handwritten the information to be written on a blank ballot paper 32 (see Figure 8), in a single-seat constituency election, when the number of lines of writing in the writing box 322 is two.
[0061] As with proportional representation elections, the processing flow of the ballot reader 1 in the determination mode will be explained using Figure 6. After setting the operating mode of the ballot reader 1 to the determination mode, the user places the completed ballot paper 32a into the hopper 10. There may be multiple completed ballot papers 32a, and the processing of Figure 5 is performed for each completed ballot paper 32a.
[0062] The operations from step S1 to step S7 are the same as those in the determination mode for proportional representation elections. If the content written in the entry box 322 is two or more lines, the information processing means 4 sequentially connects the read results of each line of the content and performs character recognition as a single read character string (step S8). For example, the content written in the entry box 322 of FIG. 10 is "Yamada Ichiro," with "Yamada" written vertically in the first line from the right and "Ichiro" written vertically in the second line from the right. In this case, the information processing means 4 sequentially connects the content "Yamada" in the first line from the right and the content "Ichiro" in the second line from the right, and determines the character string "Yamada Ichiro" as the read character string. Note that while the case where the content written spans two lines has been described here, if the content written spans three or more lines, the content of each line is similarly sequentially connected to form a read character string.
[0063] If the content written in the entry box 322 is one line, the information processing means 4 performs character recognition on the written content as it is as one read character string (step S9).
[0064] Next, the information processing means 4 compares the read character string connected in step S8 with the candidate list. Furthermore, if the content written in the entry box 322 is one line or more, the information processing means 4 uses the character string of the written content as it is as a read character string and compares it with the registered name and alias in the candidate list (step S10). The candidate list is a list that lists the names of candidates in an election. Table 3 is an excerpt from a candidate list used in the single-seat constituency election in this embodiment 1. The candidate list lists the candidate numbers assigned to each candidate who has registered their candidacy with the election chief, the candidate names corresponding to the candidate numbers, and the aliases for the candidate names.
[0065] [Table 3]
[0066] Generally, when a voter handwrites a candidate's name on a ballot in a single-seat constituency election, there are various possibilities, such as writing the candidate's name exactly, writing the abbreviation registered by the candidate, writing part or all of the candidate's name in hiragana, or writing an abbreviation with part of the candidate's name omitted at the voter's discretion. In elections, even if a voter writes part or all of the registered name in hiragana or abbreviates part of the registered name, if the candidate's name is identifiable, it is generally considered a valid vote and a vote for that candidate. Therefore, the abbreviation registered by the candidate, the character string of the candidate's name in part or all of hiragana, and the character string of the candidate's name omitted, which are assumed to appear on the ballot, are listed as alternative names on the candidate list.
[0067] The information processing means 4 compares the read character string with the registered name and alias in the candidate list and determines whether a matching name is detected (step S11). If a matching name is detected, the information processing means 4 counts the number of votes as votes for the candidate with the candidate number corresponding to the matching name (step S12), and then transports the completed ballot paper 32a to one of stackers 12-15, 21-27 according to the contents of the ballot, thereby completing the process.
[0068] Next, if it is determined in step S11 that the read character string does not match the registered name and alias in the candidate list, the information processing means 4 collates the read character string with the allocation target list (step S13).
[0069] In general, in elections, if the contents of a ballot correspond to multiple candidates, the ballot may be treated as a pro rata ballot, in which the votes are apportioned among the candidates. In step S13, the read character string of the contents of the ballot is compared with the pro rata ballot list to determine whether it is eligible for such apportionment.
[0070] Table 4 is an excerpt from the allocation target list used in the proportional representation election of this embodiment 1, and lists the allocation target character string and the name of the candidate to which the allocation is to be made that corresponds to this read character string. For example, if the read character string in the entry box on completed ballot paper 32a is "Yamada" or "Yamada," the contents of this completed ballot paper correspond to both Yamada Ichiro, candidate number 1, and Yamada Goro, candidate number 2, who has the same surname as this candidate, and therefore the votes related to this ballot paper are subject to allocation.
[0071] [Table 4]
[0072] The information processing means 4 determines whether or not a character string to be allocated that matches the read character string has been detected as a result of the comparison in step S13 (step S14). If a character string to be allocated is detected, the information processing means 4 counts the vote associated with this ballot as an allocation ticket (step S15), and then has the completed ballot 32a transported to the allocation ticket stacker 28, thereby completing the process.
[0073] Furthermore, if the information processing means 4 does not detect a character string to be allocated that matches the read character string, it determines that the contents of the ballot paper are invalid because the read character string does not correspond to the candidate's name, abbreviation or alias of the candidate, or an item to be allocated (step S6), and then transports the completed ballot paper 32a to the stacker 16 for invalid or unidentifiable ballots, thereby ending the process.
[0074] Next, we will explain the advantages of the present embodiment 1 over conventional techniques in single-seat constituency elections. Referring again to Figure 10, the content written in entry box 322 of completed ballot paper 32a is "Yamada Ichiro," with "Yamada" written vertically on the first line from the right and "Ichiro" written vertically on the second line from the right. When the content written in such entry box 322 is read by a conventional ballot paper reading device, either "Yamada" on the first line or "Ichiro" on the second line is read as the read character string.
[0075] For this reason, conventional ballot paper reading devices can only read one line of the written content when the entry box 322 contains multiple lines. Therefore, if the line break position of the written content changes, the read character string also changes. For example, if the written content is written over two lines and the content is "Yamada Ichiro," the read character string may be any of the following depending on the line break position: "Yama," "Yamada," "Yamada Ichiro," "Yamada Ichiro," "Yamada Ichiro," "Taichiro," "Ichiro," and "Ro." If these different read results depending on the line break position were to be covered by aliases in the candidate list as shown in Table 3, the alias column in the candidate list would become too long, which would increase the labor required to register aliases in the candidate list. Furthermore, considering the case where all or part of the candidate's name is written in kana, the number of read result patterns that must be covered would further increase.
[0076] 11 is a diagram showing another example of the writing surface when the number of lines of writing content within box 322 of completed ballot paper 32a in a single-seat constituency election is two. In this example, the writing content in box 322 of completed ballot paper 32a is "Yamada Goro," with "Yamada" written vertically on the first line from the right and "Goro" written vertically on the second line from the right.
[0077] In the case of the completed ballot 32a shown in Figure 10, where the read character string in entry box 322 is "Yamada" in the first line from the right, and the completed ballot 32a shown in Figure 11, where the read character string in entry box 322 is "Yamada" in the first line from the right, the candidate names corresponding to the read character string are both "Yamada Ichiro" and "Yamada Goro." Therefore, since the original content written in entry box 322 of the completed ballot 32a shown in Figure 10 is "Yamada Ichiro," and the original content written in entry box 322 of the completed ballot 32a shown in Figure 11 is "Yamada Goro," the candidate names can be identified, but there is a problem in that votes are subject to proration based on the read character string.
[0078] On the other hand, in the ballot reading device 1 of this embodiment 1, if the writing in the entry box 322 of the completed ballot 32a spans multiple lines, the writing on each line is connected to form a read string, thereby preventing the read string from changing due to changes in the line break position of the written content, and also preventing ballots on which the name of the candidate originally written can be identified from being subject to apportionment due to the line break position of the written content.
[0079] Thus, the character recognition method of this embodiment 1 is a character recognition method that reads handwritten characters written on the ballot paper 3 via the information processing means 4, and includes the steps of: step S4 of reading handwritten characters written on the completed ballot paper 31a, 32a via the information processing means 4; step S7 of identifying whether the read character string read via the information processing means 4 has one line or multiple lines; step S9 of recognizing the characters of the read character string as is if the read character string has one line; and step S8 of combining the characters of each successive line of the read character string to form a single combined character string and recognizing the characters of the combined character string if the read character string has multiple lines.Therefore, by connecting and reading the handwritten character strings written over multiple lines in the entry boxes 312, 322 of the ballot paper 3, the handwritten character string can be easily read, and it is possible to prevent votes whose original content in the handwritten character string is not subject to apportionment from being recognized as votes subject to apportionment.
[0080] Furthermore, the ballot reading device 1 of this embodiment 1 reads handwritten characters written on the ballot 3 using the character recognition method described above, and therefore provides a ballot reading device that can easily read handwritten character strings written over multiple lines in the entry boxes 312, 322 on the ballot 3.
[0081] Embodiment 2 Next, a character recognition method according to a second embodiment of the present invention will be described. The character recognition method according to the second embodiment is a modification of the character recognition method according to the first embodiment, in that handwritten characters written over multiple lines are read using both a reading method in which the first line is read from the side closest to one end of the ballot paper, and a reading method in which the first line is read from the side closest to the other end of the ballot paper. In the second embodiment, the same reference numerals as those in Figures 1 to 11 indicate the same or similar components as those in the first embodiment, and therefore detailed description thereof will be omitted. Furthermore, the following description of the second embodiment will focus on the case of voting in a proportional representation election, but the same applies to voting in a single-seat constituency election.
[0082] The character recognition method according to the second embodiment is the same as that according to the first embodiment up to step S7 shown in Figure 4. Next, if the content written in the entry box 312 is two or more lines, the information processing means 4 sequentially connects the read results of each line of the content and performs character recognition as a single read character string (step S8). At this time, the information processing means 4 connects the content by both connecting the content sequentially in the order along the direction of the arrow X on the completed ballot papers 31a, 32a, and connecting the content sequentially in the order opposite to the arrow X.
[0083] For example, in entry boxes 312 and 322 in FIG. 6 , the content is "ABCD Party," with "AB" written vertically on the first line from the right and "CD Party" written vertically on the second line from the right. In this case, information processing means 4 sequentially connects the content "AB" on the first line from the right with the content "CD Party" on the second line from the right to create the character string "ABCD Party" as a first read character string, and sequentially connects the content "CD Party" on the second line from the right with the content "AB" on the first line from the right to create the character string "CD Party AB" as a second read character string. Note that while the case where the content spans two lines has been described here, even if the content spans three or more lines, the content of each line is similarly connected sequentially from right to left to create a read character string.
[0084] Next, if the contents of the entry boxes 312, 322 are two or more lines, the information processing means 4 compares both the first read string "ABCD Party" connected in step S8 and the second read string "CD Party AB" with the registered name and alias in the candidate list (step S10).
[0085] Next, the information processing means 4 compares the first and second read character strings with the registered names and aliases in the candidate list and determines whether a matching name is detected (step S11). If a matching name is detected, the information processing means 4 counts the number of votes as votes for the candidate having the candidate number corresponding to the matching name (step S12), and then transports the completed ballot paper 31a to one of stackers 12-15, 21-27 according to the contents of the ballot, thereby completing the process.
[0086] Next, if it is determined in step S11 that the read character string does not match the registered name or alias in the candidate list, the information processing means 4 compares the first read character string and the second read character string with the allocation target list (step S13). As a result of the comparison in step S13, the information processing means 4 determines whether or not an allocation target character string that matches the first read character string or the second read character string has been detected (step S14). If an allocation target character string is detected, the information processing means 4 counts the vote associated with this ballot as an allocation ticket (step S15), and has the allocation ticket stacker 28 transport the completed ballot 31a, thereby completing the process.
[0087] Furthermore, if the information processing means 4 does not detect any character strings to be apportioned that match the first read character string and the second read character string, it determines that the contents of the ballot paper are invalid (step S6), and ends the process by transporting the completed ballot paper 31a to the stacker 16 for invalid or unidentifiable ballots. The rest of the configuration and operation of the ballot paper reading device are the same as in embodiment 1.
[0088] In this way, in step S8, when combining multiple lines of read character strings, the system includes a step of combining the character strings of each line, with the line at one end of the read character string as the first line and the line at the other end as the last line, to form a single combined character string, and a step of combining the character strings of each line, with the line at the other end of the read character string as the first line and the line at one end as the last line, to form a single combined character string.Therefore, even if the candidate's name is written in the opposite line in entry boxes 312, 322 on ballot paper 3, the contents of the handwritten character string can be read.
[0089] In the first and second embodiments, first and second imaging means 175, 176 are provided as imaging means, but an imaging means may be provided alone. In this case, the user will align the front and back of the completed ballot slip 32a before inserting it into the hopper 10.
[0090] Furthermore, the character recognition method of the ballot reading device in embodiments 1 and 2 was used for character recognition of ballot papers in proportional representation elections and single-seat constituency elections, but is not limited to this and can be used for various types of elections such as mayoral elections and multi-seat constituency elections.
[0091] The components included in the first and second embodiments of the present invention and the components included in the modifications thereof can be used in any suitable combination.
[0092] Although the preferred embodiments have been described in detail above, the present invention is not limited to the above-described embodiments, and various modifications and substitutions can be made to the above-described embodiments without departing from the scope of the claims. [Explanation of symbols]
[0093] 1 Ballot paper reading device, 3 Ballot paper, 31 Blank ballot paper, 31a Completed ballot paper, 32 Blank ballot paper, 32a Completed ballot paper, 4 Information processing means (photoelectric conversion means).
Claims
1. A character recognition method for reading handwritten characters written on a ballot paper via a photoelectric conversion means, a first step of reading the handwritten characters written on the ballot paper via the photoelectric conversion means; a second step of identifying whether the number of lines of the read character string read via the photoelectric conversion means is one line or multiple lines; a third step of recognizing characters of the read character string as they are when the number of lines of the read character string is one; a fourth step of combining character strings of successive lines of the read character string into a single combined character string and recognizing characters of the combined character string when the read character string has a plurality of lines; A character recognition method comprising:
2. a fifth step of combining the read character strings of multiple lines in the fourth step, by combining the character strings of each line into one combined character string, with the line at one end of the read character string as the first line and the line at the other end as the last line; a sixth step of combining the character strings of the read character strings into one combined character string, with the other end line of the read character string as the first line and the one end line as the last line.
3. A ballot reading device that reads the handwritten characters written on the ballot paper using the character recognition method described in claim 1 or 2.
Citation Information
Patent Citations
Character recognition method
JP2002150222A
Reading device and reading method of ballot paper
JP2012023550A