JPH01320171A - Method of processing framed character - Google Patents

Abstract

PURPOSE: To efficiently form a data base by using primary scanning block labelling and a structurizing character cut-out method to automatically separate printing element character data into blocks and subsequently attaching a frame to characters in a character string. CONSTITUTION: A document is scanned by an optical reader to read printing element character data such as character, image and figure data or the like being the image data of the document to input the same and the input data (binarized materials) is separated into a character string (character block), an image block and a figure block by a block separation means 20 performing block separation by a primary scanning block labelling method. Subsequently, the characters in the character string separated by the block separation means 20 are cut-out by a character cut-out means 30 by judging a square attribute (kanji or English character) by a structurizing character cut-out method to perform character cut-out processing.

Description

Detailed Description of the Invention (Industrial Application Field) The present invention relates to a frame character processing method. Text information (binary data)
The present invention relates to a method for processing cut-out frame characters by first separating them into character strings, image blocks, and graphic blocks, and then adding frames to the characters in the character string.

(Summary of the Invention) The frame character processing method of the present invention uses primary scanning block labeling (○ne Path Block Labeling).
) and structured character extraction method (5tructvre 工5o
(latin) is used to generate print character information, such as information in which characters and figures coexist, information in which uppercase and lowercase letters are mixed, character information with uneven character spacing, information consisting of letters, alphanumeric characters, and general symbols, etc. By automatically separating the string into blocks, and then cutting out the characters in the string by placing a frame around them,
Efficient (database).

(Prior art) Nowadays, with the progress of computerization, text in documents (Te
xt ), image (Image), graph
There is a need to efficiently compile information such as c) into a database.

For this reason, it is important for document analysis systems to reduce the amount of data accumulated by automatically classifying information such as text, images, and figures in documents and processing each type of information using appropriate technology. . In addition, we will use an optical character recognition system (0ptical char recognition system) to accurately cut out areas of characters and symbols from the framed text.
Acter Recognition:0CR) is equally important. In general, automatically classifying text, images, and graphic information falls within the scope of block separation processing, but Extracting characters and symbols from a character string falls within the scope of character extraction processing. These are called kajidokoro ■ri.

The purpose of this frame processing is to classify the object to be processed, which is l in the class 4 analytical nostem. The following block separation processing techniques have been known so far: (1) Projection method (Histogram); Avi
nash Kak and Azriel Raaenf
eld″Digital Picture Process
-See ssing'.

(2) Expansion and contraction method (ExpAnsion and S
hrinking); Nakamura, Buke, Kaimoto, Minami “Character area extraction algorithm for mixed-mode communication” IEICE theory (D) J67-D, Oborott, PP 1277-1
284 (1984) and Akita, Naito, Masuda 6 Individual character extraction method from vertically and horizontally written characters "Shingi Giho, PRL-
See 83-7, PP49-56 (t983).

(3) Con5train Run Length
Law (hereinafter abbreviated as CRL); F, M,
Yea, R, S, Ca5ey; Block
Segmentation and Text Ex
Tractionin Mixed Text / I
mageDocuments” + Com-pu
ter Graphic & Image process
ssing t 20 rPP375-390 (198
2) See.

(4) Small region separation method; N, H, Yea, R, S,
Leuand'G,H,Lin''CHARACTE
RRECOGNI-TION ByL BOARD O
CRRART: TRAlN5EPARATION
&CHARAC:TEREXTRAGT knee○N″+
Processing of International
al Computer Bymposium' *D
ember l 7-19 r PP l 29~
137 (1986) Reference 0 (Problems to be Solved by the Invention) However, the above block separation processing techniques (1) to (4) all have advantages and disadvantages, and are not satisfactory.

That is, (1) Projection method (Histogram)
Although this method has a high processing speed, there is a problem that it cannot be applied to recording paper on which text and other materials are slanted, or on recording paper where text and graphics coexist.

In addition, (2) expansion, contraction method and +3) CRL method,
Although it is possible to process complex documents, the processing speed is slow,
In addition, (4) the small area separation method has a high processing speed and can process even slanted text materials, but the format of the input recording paper This invention is intended to solve the above-mentioned problems of the prior art. The objective is to provide a method for processing frame characters that can be efficiently converted into data databases by eliminating problems such as inability to process tilted documents and inability to process complex documents. 0 (Means to solve problems) In the first invention, image data of a document read into a computer is separated into character strings (character blocks), image blocks and graphic blocks by a block separation processing method,
Then, using the character extraction method, kanji,
In a frame character processing method for cutting out alphanumeric characters and punctuation marks with a frame, the block separation processing method is applied to a bitmap image length code (Ru
n LengthCOae)<CRLC to convert
(CnstrainedRun L+length Co
de) conversion process and labeling (Labelin)
g) A block separation process in which adjoining GRLCs are made consecutive and separated into -blocks (independent blocks), and the original bitmap is created from the CRLCs recorded in the independent blocks.
By calculating each corresponding position (X, Y) of the image, the image extraction process of each block (&'?: extracting the image) and the run length (1 and run length of the block) are performed.
th) A block image type discrimination process of determining the block type from the average value.

Further, in the second invention, the character extraction method includes a rectangle search process of framing rectangles separated on the left and right sides of the character string;
Find the height of a kanji character using the height scoring method. A character string height calculation process, a priority extraction process that cuts out kanji and punctuation marks based on their attributes, and a comparatively relaxed condition to calculate the height of all the cut out rectangles. The present invention is characterized in that it is constructed using a structured character extraction method that includes a rightward and leftward extraction process that searches for a Kanji rectangle located on the left side. The image data of a document read into a computer is separated into character strings (character blocks), image blocks, and graphic blocks, and then, using the structured character extraction method described above, kanji, alphanumeric characters are extracted from the separated character strings. It is characterized by adding a frame to punctuation marks and cutting them out.

In this invention, "Run J" refers to black (or white) pixels in a continuous image row.

Also, "Run Length Code"
hCode)J refers to a data code having II Run 11 starting point position and length.

It's the top, [Con5tratn]
-ed Run )J refers to "Run5" where the horizontal distance is smaller than a certain constant.

Also, “Constrained Run Code”
Rained Run Code) J is “Co”.
"n5train-edRun" is a data code that has the leftmost starting point position and total length.

(Function of the Invention) In order to solve the problem of inapplicability when the recorded content of the recording paper is tilted or when text and figures coexist, the present invention uses a primary scanning block labeling method that is an improvement on the conventional C3L method. A block separation method called . As a result, all blocks can be found and transformed simply by transversely distorting the -cycle training image plane. That is, by scanning the original image plane once instead of requiring three image planes (original image plane, horizontal extension plane, and vertical extension plane) and having to scan this image plane multiple times. I'm trying to get by with just that.

Therefore, the processing speed can be increased.

In addition, for character extraction processing, the outer rectangle of each character is usually found using the projection method, but if kanji and English characters coexist in the text, it is necessary to determine the rectangles of kanji and English characters. There is. The following are possible causes that make this character extraction process difficult.

(1) This is because the pitches do not match. In other words, since the pitches in the shapes of the printed characters do not match, it is not possible to automatically cut out the characters as pitch markers.

(2) Due to noise interference. (3) Due to separation of characters. For example, the double rectangular structure ``hi'' may be cut out as two characters ``yumi'' and ``1.'' (4) Due to the contact of characters. For example, “I” and °“
When the two characters ``i'' and ``(kun)'' are printed in type, the space between the characters is too small, so there is a possibility that they will be cut out as a single character.

In this invention, based on the principle that all judgment parameters used for character extraction processing are extracted from input recording paper, the attributes of rectangles (kanji or English characters) are
By adopting a method of determining the above, it is possible to eliminate the difficulties associated with the above-mentioned character segmentation process and to perform character segmentation accurately.

(Embodiment) An embodiment of the present invention will be described below in detail with reference to the drawings.

The third invention is a combination of the block separation processing method that is a feature of the first invention and the character extraction method that is a feature of the second invention. Therefore, in the following example, @
The third invention will be explained in detail.0 The first and second inventions will be explained in the process of explaining the third invention.

FIG. 2g1 is a block diagram showing an apparatus for executing the method for processing text, image, graphic information, etc. of the third invention. In the figure, reference numeral IO denotes an optical reading device, 20 denotes block separation means for separating blocks by the primary scanning block labeling method, and 30
is a character extraction means that performs character extraction processing using a structured character extraction method.

The optical reading device 10 scans the document to read and input printed text information such as text, images, and graphic information, which is image data of the document, and this input data (binarized material) is processed by the block separation means 20. It is separated into a character string (character block), an image block, and a graphic block, and then the character cutting means 20 cuts out the characters in the character string separated by the block separating means 20.

FIG. 2 is an ordinary document, which includes character strings (character blocks), halftone images, and @ line figures. When the image data of this document is read by the optical reading device 10 and processed by the block separation means 20, as shown in FIG. 3, character strings, image blocks,
This separation into graphic blocks has the following characteristics.

(1) The white shaded areas are the main locations of block separation.

(2) The margin distance within the same block is small.

(3) The margin distance between different blocks is large.

Here, the basic principle of the primary scanning block labeling method is as follows.

(1) Add bitmap image to run length code (Run Length)
e) Convert to 0 (2) Label "Run" that is close to each other as the same block due to the neighboring area relationship.

(3) If Runl+ in the neighborhood of the block no longer exists, this block is considered an independent block. (4) All Runs in the independent block are returned to a bitmap image, and the attributes of the block (character string , image block, graphic block).

In this primary operation block labeling method, operations (1) to (4) above are executed by scanning the screen once, and all blocks can be separated.

The 2g4 diagram shows the block separator I! 12 is a flowchart showing the processing content of the primary scanning block labeling method executed by the i20.

First, CRLC (Constrained Run L)
In the conversion process, the bitmap image read by the character reading device 10 is converted to CRLc f:. In other words, a bitmap with a large number of data.

(bitmap) images, each Con5trained
) Only the values of the start and end positions of fun were recorded (, converted to RLC to reduce the amount of data.

The reason why CRLC is adopted in this case is to generate a local expansion effect in the horizontal direction.

Here, since the selection of constant';tI CH has a strong influence on the quality of the bar row result, it is necessary to set it to an appropriate value. In other words, if the constant ch is too thick, the local expansion effect will be meaningless, and two blocks with an inclination angle will overlap each other.Also, if the constant ch is too small, Not only will no expansion effect be obtained, but the amount of data processing during block separation will increase.Next, in the block separation process, labeling (Label
CRLCs in an adjacent relationship by adopting the method of
Obtain the position of the - block (separate into independent blocks) by making it consecutive. Here, the labeling (La
Q. First, explain the method of ``Be ling''.
’ This 6 neighboring area” is the 5th
As shown in the figure, Con5trained Run (
This refers to the part expanded from the black part in the figure, that is, the part surrounded by dotted lines. Blocks that are in contact with this neighboring area are called neighboring blocks of this Run. In addition,
In the figure, XGAP is the amount of horizontal expansion, and YGAP is the amount of vertical expansion.

Next, the processing procedure will be explained.

I, CRLC (Starting X, Length
th)? (Sta-rtingX, endi
ngX).

TI, (a) R from among the already existing blocks
Find and record neighboring blocks of un.

(b) Record all blocks other than the neighboring blocks that already exist (Ca5e l) If the above Run has no neighboring blocks, the core Run
't'') (DInsert into a new block.

(Ca5e 2 ) If the above Run has neighboring blocks, incorporate the Run into the highest neighboring block M. And l of this neighboring block M? RLc to 1 in the remaining neighboring blocks.
IIc so that the new block M formed by the correction has a continuous relationship with all neighboring blocks.

■, Check all blocks other than the already existing neighboring blocks, and if there is no possibility of contacting any new neighboring blocks, send the 0RLC data of the block to the next work unit as independent data. O ■, New CRLC; 0. As is clear from the above processing steps, labeling (La
The method described above compensates for the lack of CRLCO by locally expanding it in the vertical and horizontal directions based on the definition of 6 neighboring blocks, and also makes the dense region of GRLC independent. It is separated as a block.

After processing by the above labeling method, an image extraction process extracts the original bitmap from the CRLC recorded in the independent block.
p) Calculate each corresponding position (X, Y) in the image, -
Extract images of blocks.

Thereafter, the image type (text, image, figure) of each block is analyzed by an image type discrimination process. In other words, it is separated into character strings, image blocks, or graphic blocks.

The emphasis of the analysis here is to determine whether the blocks obtained so far are character strings, but from the results of this analysis
n) Can also judge diagrams, etc.

This analysis requires the block height (H) data and the 0 average stroke width in the running direction (or ``average Run Le
ngth” value (RL)) data is used.

Here, RC = oCrossi, which is the black phase recognition on the scanning direction/Crossing Count on the scanning direction.
ngCount refers to the number of times a pixel is converted from white to black in the scanning direction.

Next, the analysis contents (judgment principles) are shown in Figure 6 (a) and (b).
I will listen to and explain.

In addition, HO = Maximum possible height of the character Hl Maximum possible height of the example character = HO + Maximum deviation of the slope = HO + W Half ton y (Half ton) Can distinguish shapes, geometric figures, or vertical directions. Experimental value of average Run Length TH2 = Can distinguish horizontal straight lines, character strings, or short vertical lines. Experimental value of average Run Length TH3 = Average Run Len that can separate character strings or short vertical lines from the vertical direction
Experimental value of gth TH4=Experimental value of average Run Length that can distinguish a geometric figure or vertical if line from the vertical direction.

First, based on the height of the blocks, they are divided into the following two groups.

(1) Items whose height exceeds the character example range (F(O
>Ml).

This is a geometric figure, FlffA or halftone (Ha
lfton).

Here, judging from the average horizontal stroke width,
The larger 1i6 (RC>THI in the horizontal direction) is
It is a half ton (Half ton) figure. Also,
Judging from the average stroke width in the vertical direction, a line with a large width (RG>TH3 in the vertical direction) is a vertical line. What remains are geometric figures.

(2) Those whose height is within the range of the character string (HO≧H1)
.

This can be determined to be a type of character string or horizontal line. In addition to this, there are some that are confused because the height is similar to that of a character string. This includes only a few short vertical lines, small geometric figures or halftone figures.

Here, judging from the average stroke width in the horizontal direction,
Larger widths (Re>T in the horizontal direction I2) are horizontal lines or halftones included in this group.
n) is a diagram. Also, the average vertical stroke 1
Judging from 1G, the widest one (vertical Re
)TH4) is the vertical line placed in this group. What remains is a string. As an exception, if a character string or a small geometric school shape whose height is similar to that of a character string cannot be clearly distinguished, the length and width of each part will be determined after being sent to the next character cutting process. By the ratio of
A judgment is made as to whether it is a character or not.

Once the character string is separated in this way, the character string is sent to character extraction means 30, where it is subjected to character extraction processing using a structured character extraction method. Next, the contents of this character extraction process will be explained.

The purpose of character extraction processing is to attach rectangular frames to characters (kanji, alphanumeric characters) and symbols (punctuation marks, special symbols) for each separated character string.

For example, if you frame it like Le・□川Q0 In this case, if you directly adopt the projection method, there will be the following problem. In other words, rectangles for alphanumeric characters and general symbols belong to a single rectangle, but rectangles for kanji belong to numbers (1 ~ s i[!if )
is not constant. Therefore, it is not possible to correctly frame a kanji unless it is checked whether a single rectangle satisfies the kanji conditions. Also, for multiple rectangles, merge the rectangles and check whether the merged rectangle meets the conditions of the kanji restriction. If there is a match, this indicates that what is inside the rectangle is a kanji, so the rectangular fusion object must be stopped. If this is not done, the fusion operation will continue until the fused rectangle exceeds the limit.

By the way, statistics show that kanji have the following characteristics.

(1) Among the 5,401 commonly used kanji, 87 characters belong to a single rectangular structure, and 11 characters belong to a double rectangular structure.
%, and those that do not belong to either of these account for 2.

(2) Kanji generally have a square shape, 1, and the aspect ratio is normal; j! 11: t, but kanji with a single structure have a large change in aspect ratio, for example "°日" is 1.6:
1. "JIII" is 0.77:l. Note that kanji characters with a birectangular or larger structure are relatively stable, and 98% of them have an aspect ratio between 0.9:l and 1.1:l.

(3) Kanji appear consecutively (Kanji are always next to each other)
.

The A11η artificial character extraction method of the present invention incorporates these characteristics of kanji characters and allows characters in a character string to be extracted with a frame, without causing the above-mentioned problems. The basic contents are as follows. That is, (1) First, examine the characteristics of all single rectangular structures and extract kanji rectangles that meet strict criteria.

(2) Based on the characteristic that kanji characters are always adjacent to each other, if there are characters to be cut out on the left and right, but nothing in between has been cut out yet, a single rectangle or a fused rectangle is determined under relaxed conditions.

(3) After all kanji have been cut out, the remaining uncut characters shall be alphanumeric characters or symbols.

FIG. 7 is a flowchart showing the processing contents executed by the character cutting means 30.

First, in the rectangle search process, frames are placed around all left and right rectangles in the character string. This is done as follows.

(1) Find the positions of the left and right ends of the rectangle from the projection of the image on the horizontal axis (X) (see Xi r X2 in Figure 8),
Find the horizontal position of the rectangle.

(2) Vertical axis (Y) of the image within the left and right end areas of each rectangle
Find the positions of the upper and lower ends of the rectangle by projecting them onto (Yl, Y in the same figure)
2), determine the vertical position of the rectangle.

Next, calculate the height of the kanji in the string. This is done because the parameters for determining whether a rectangle matches a Kanji character shape are all related to the height of the Kanji character. The principle of measuring the height of a character string using the speaking score method is as follows.

(1) When a certain height h occurs, h and the area before and after it have an influence on the corresponding height score. (2) When the rectangle of a kanji is larger than the rectangle of an alphanumeric character, when does the kanji become Qf-like? Increase the weight of points.

(3) - Set the height of the maximum score in the character string as the height of the kanji.

Figures 9(a) and 9(b) show the score status of the height of a certain single rectangle. As shown in the figure, the height of a character string is statistically calculated from the height scores of all rectangles in the character string.

Next, perform a waste cutting process. This priority extraction process processes a single rectangle, extracts three types of stable and error-prone Kanji and punctuation symbols, and gives display attributes to each.

The first category is a proper kanji, and has an attribute of 1.

The cutting conditions are (1) 1en-3<rectangle length<ten-L(2
)wid-3<width of rectangle (wid-L(3) r
It is assumed that at-3<aspect ratio of rectangle<rat-L (4) distance between left and right rectangles>bnk.

The second type is a comma (,), a comma (1), and a period (o), and the attribute is 2. The extraction conditions are: (1) Distance between left and right rectangles > Comma (2) dat-8 < rectangle Length, width (dat-L) The third class is "-", ", '='", and the attribute is 3.

(1) dat-3<the length of the rectangle (dat-3+ d
at-L) > 1 (2) Width of the rectangle <wid-3〇Here, ten-3 is the minimum allowable length of the rectangle for a proper kanji, and 1en-L is the allowable length for a proper kanji. wid-3 is the minimum allowable rectangular width of a neat kanji, wid-L is the maximum allowable rectangular width of a neat kanji, rat-3 is a neat kanji. rat-L is the maximum permissible rectangular aspect ratio of a proper kanji; Comma is the minimum permissible rectangular length of phrases, commas, etc., and left and right rectangles; dat-8 is the minimum permissible rectangular length of a rectangle such as a phrase or comma, dat-L is the maximum permissible rectangular length of a rectangle such as a phrase or comma, bnk is the distance between two kanji, i.e. Refers to the width of the margin.

FIG. 10 shows the correspondence between rectangles and attributes processed in the above priority extraction process.

Next, execute the rightward extraction process.0 In this rightward extraction process, the left side rectangle is cut out to the left by taking advantage of the property that kanji are adjacent to each other, and the rectangle next to it (right side) is still cut out. For those that have not been cut out, structural cutout judgments regarding single rectangles and fused rectangles are made under relatively relaxed conditions, and a kanji rectangle located on the right side of all cut out rectangles is searched and hung.

The conditions for cutting out are: (1) length-3< (fusion) length of rectangle length-L (2) width -3< (fusion) width of rectangle (w
idth-L (3) Distance between centers of (fusion) rectangle and adjacent cut-out rectangle > pitch-8 (4) For a single rectangle, cam-3 < rectangle aspect ratio <
com-L (5) For the fused rectangle, m1d-3<the aspect ratio of the rectangle<
m1d-L If there are multiple rectangles, 64 plus the above conditions,
(Fusion) Select the one with the smallest difference between the width of the rectangle and the height of the kanji, which is the one most similar to a square, and cut it to make a hanging.

After this, a leftward cutting process is performed. In this process, the same operation as the rightward cutting process is performed, but the target is that the right rectangle has already been cut out, but the rectangle on the left side (1ftl+') has not been cut out yet. .

In this way, characters are cut out from a string with frames attached to them, but some of the strings have their leftmost rectangles and rightmost rectangles left uncut. For such an audience, special processing is applied to the rectangle at both ends using the end point extraction process to ensure that all kanji are accurately extracted.Q Figure 11 shows the above character extraction process. This shows the results.

After this character extraction process, an optical character recognition system (Op-tic
alCharacter Recognition
; be forwarded to OCR).

In order to confirm the practicality of the frame character method, an experiment was conducted using the C language on a PC knee AT, and the results shown in the following table were obtained.

The following variable values were used. In addition, 5ize
represents the height of the kanji in the string.

len −L = 1.25Xsize fen −
5=0.87Xsisewid-L=1.15
X 5ize width-3= 0.83
erat −L = 1.37 rat −8
= 0.751ength - L = 1.35 X s
size length -S= 0.58
width −L = 1.45
th −S= 0.57 X sizecom −L
= 1.8 com - S = 0.48m
1d −LJ = 1.55 mj d −S
= 0.58pitch-S = 0.95X
sizecomma = 0.35 x 5ize
dat-L=0.45Xsize dat-3=O
, 1Xsize In addition, the processing targets are electronic development monthly magazines and other magazines that contain figures, tables, and sentences (including kanji, alphanumeric characters, and special symbols, and the distance between characters is not fixed). , the number of pages was 10.

Table test material 12345678910 Total number of characters 33
0300220300320220720390460
800 incorrect characters 3 2 2 4 3 2
1 12 9 15 Cutting rate 99. l 99.39
9.198.799, 199.199, 896.998
．． 098.1 (Effects of the Invention) This has been explained above. According to the invention of J:5;j';l,
Since the primary scanning block labeling method is adopted,
- Character strings, image blocks, and graphic blocks can be separated and processed in one scan, and databases can be created efficiently. Further, according to the second invention, since a structured character extraction method is adopted, characters can be accurately extracted from a character string with a frame attached to them, and a database can be created efficiently. Furthermore, according to the third invention, after character strings, image blocks, and graphic blocks are separated by a primary scanning block labeling method, characters are cut out from the character string with a frame attached to them by a structured character extraction method. Because
Block separation does not take much time, character extraction can be performed accurately, and a database can be created efficiently.

[Brief explanation of the drawing]

Figure 'yJ1 is a block diagram of a device implementing the third invention,
Fig. 2 is a diagram showing class V before block separation, Fig. 3 is an explanatory diagram showing a document after block separation processing, Fig. 4 is a flowchart showing the processing contents by the primary scanning block labeling method, and Fig. 5 is a diagram showing the document after block separation processing. °Explanatory diagram explaining the yen request of "neighboring territory/ekisu", Figures 6 (a) and (b) are the content of analysis (judgment principle)'? : A block characteristic distribution diagram to be explained; Figure 7 is a flowchart showing the processing contents by the structured character extraction method;
Figure 8 is an explanatory diagram of the operation (rectangle extraction process) of projecting the character ``wa'' (boku) in the X and Y directions.
(b) is an explanatory diagram and graph showing the score status of the height of a certain single rectangle, Figure 10 is an explanatory diagram showing the correspondence status between rectangles processed in the priority extraction process and attributes, and Figure 11 is FIG. 3 is an explanatory diagram showing the result of character extraction processing. io...Optical reading device, 20...Block separation means, 30...Character cutting means. 1st Okoo 2 Kuni #, 3 Fig. 5 Fig. 4 Seki Pro v7 Kenjii 1st Kosui + way I8″1RC 1-1o l-I+ HO Kawakae, 7 Kochii Maii Rei UゑDekume Figure 8 Qin 10 concave pass construction ≧ mo 11 concave τu・) Nyoopa statue primary 8 separate '44 m1 Rorororororo = Rorororo purification point ``7''
Mouth Roro Ishika Shun 177 Hitoshaguchi Mouth pigeon,! ! ! Shun-out frame damage h + Rorororororororo fixed 4 surnames 110013101 Meme quantification/becomes pb 4 Roro hits Rorororororororo Rorororororororororororo n mouth + 1 1 1 21001

Claims (3)

[Claims] (1) The image data of the document read into the computer is converted into character strings (character blocks) using the block separation processing method.
, a frame character processing method in which kanji, alphanumeric characters, and punctuation marks are separated into image blocks and graphic blocks, and then cut out from the separated character strings by adding a frame to kanji, alphanumeric characters, and punctuation marks using a character extraction method, Map (bitmap) image by run length
CRLC (Cn strained Run Le
ngth code conversion process, a block separation process that separates adjacent CRLCs into one block (independent block) by making them contiguous using the labeling method, and a block separation process that separates CRLCs recorded in independent blocks. Each corresponding position (X
. A frame character processing method characterized in that it is configured by a primary scanning block labeling method consisting of. (2) The image data of the document read into the computer is converted into character strings (character blocks) using the block separation processing method.
, a frame character processing method in which kanji, alphanumeric characters, and punctuation marks are separated into image blocks and graphic blocks, and then cut out from the separated character strings by adding a frame to kanji, alphanumeric characters, and punctuation marks using a character extraction method, Comparison of a rectangle search process that frames rectangles separated on the left and right sides of a column, a character string height calculation process that calculates the height of kanji using the height scoring method, and a priority extraction process that cuts out kanji and punctuation marks based on attributes. It is characterized by being constructed by a structured character extraction method that includes rightward and leftward extraction processes that search for Kanji rectangles located to the right or left of all the extracted rectangles under strict conditions. Frame processing method. (3) Primary scanning block labeling that divides the image data of a document read into a computer into character strings (character blocks), image blocks, and graphic blocks, and frames the character strings into kanji, alphanumeric characters, and punctuation marks. and a structured character extraction method that extracts the bitmap image by adding a run length to the bitmap image.
CRLC (Cn strained Run Le
ngth code conversion process, a block separation process that separates adjacent CRLCs into one block (independent block) by making them contiguous using the labeling method, and a block separation process that separates CRLCs recorded in independent blocks. Each corresponding position (X,
The image extraction process extracts the image of each block by calculating the block height and run length (Run Len-g
th) a block image type discrimination process that discriminates the block type from the average value; A character string height calculation process that calculates the height of kanji using the height scoring method, a priority extraction process that extracts kanji and punctuation marks based on their attributes, and a relatively loose condition that calculates the height of kanji to the right or left of all the extracted rectangles. A frame character processing method characterized by comprising a rightward and leftward cutting process of searching for a kanji rectangle located in .