JP2002297638A - Method for extracting title from document image - Google Patents

Abstract

PROBLEM TO BE SOLVED: To automatically extract a title from a document image according to features characteristic of the title without depending upon any specific document format. SOLUTION: A character string area is cut rectangularly out of the document image, and points of title likelihood are added according to properties of the character string area to extract the title. The points are given on the basis of the features unique to the tile, e.g. the distance between the lines before and behind the object line, whether or not the line is meshed or framed, the difference in character size between the object line and other character lines, whether the object line is independent, etc., and the points are added to extract the line of a character string reaching specific points as the title.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for accurately expressing document contents from a database of document image data input from an image input device such as a facsimile or an image scanner in order to improve the convenience of retrieval. The present invention relates to title extraction for extracting a title area in a document.

[0002]

2. Description of the Related Art Conventionally, when searching for a document image, an operator manually operates the content of the document separately from inputting the document image from an image input device in order to improve the convenience of the subsequent search. Is extracted / created and added to the title information and the keyword information that accurately expresses, and a specific position (character string) in the document is cut out as a title / keyword for a fixed-form document.

[0003] Further, as a method of extracting a title from an irregular document using only layout features, for example, Japanese Patent Application Laid-Open No. 9-134406, entitled "Apparatus and method for extracting title from document image" Japanese Patent Application Laid-Open No. 5-274471 discloses a “method of extracting a title region of an image document”.

However, in the above-described conventional technology, the addition of title information or keyword information by an operator increases the work load as the document size increases, thereby causing an increase in work load. I will. In addition, since automatic extraction of a specific position is performed only on a fixed document, it cannot be used for an unfixed document, and there is a problem that convenience is lacking.

That is, in the documents disclosed in JP-A-9-134406 and JP-A-5-274471, titles are extracted by paying attention only to layout features, so that the contents of a document are accurately represented. However, the hit rate of the title to be obtained is not always satisfactory, and there is a problem in that it hinders later document search and the like.

Therefore, by using a feature unique to a title as a point without depending on a specific document format, a character string region having a large number of points is automatically extracted as a title, and the accuracy of title extraction and the convenience in document search are improved. Japanese Patent Application Laid-Open No. 2000-148
Japanese Patent Publication No. 788 discloses that a character string region with a high score is automatically extracted as a title by using a feature unique to the title as a point of the likeness of the title.

According to this method, it is possible to extract a title. However, since extraction feature points in a document are limited, when a document having various layouts is input, title extraction accuracy is poor. In some cases. For example,
In the case where the title line is divided into two lines instead of one line, there are cases where language processing cannot determine the likelihood of a title using the method of the previous technique. Furthermore, if the character string of the title is almost the same size as the character string of the body, the point of the title may not be obtained well enough to extract the title.

[0008]

SUMMARY OF THE INVENTION In the present invention, Japanese Patent Application Laid-Open
-148788 New features are added to those disclosed in the Japanese Patent Publication, so that the title-like features can be pointed as much as possible even from document images where title extraction cannot be performed well in the past, and the overall accuracy of title extraction is enhanced. It is intended to be.

[0009]

According to a first aspect of the present invention, there is provided a method for extracting a title by extracting a character string region from a document image in a rectangular shape and adding a point of the title likeness based on the attribute of the character string region. Measuring the line spacing before and / or after the target line, comparing the measured line spacing with a predetermined value, and giving a point of title-likeness to a character line larger than the predetermined value. This is a title extracting method characterized by comprising steps.

According to a second aspect of the present invention, in the title extracting method according to the first aspect, a character of a character line in the character string area is recognized, and whether or not the confidence of the character recognition result is lower than a predetermined value. And a step of determining whether the character line has a negative title-like point when the character line is low.

According to a third aspect of the present invention, in the title extracting method according to the first aspect, the centering position in the document is obtained, and it is determined whether or not the target line is the centering position, and it is determined that the target line is at the centering position. Then, a centering position based on the extracted character area portion is determined, and when it is determined that the target line is located at the centering position based on the character area portion, title-likeness is assigned to the character line. This is a title extracting method characterized by comprising each step of giving points.

According to a fourth aspect of the present invention, in the title extracting method according to the first aspect, it is determined whether or not another line exists next to the target line, and it is determined that another line exists. When it is determined, the title extracting method is characterized by comprising the steps of giving a negative title-like point (assuming that the target line is not independent).

According to a fifth aspect of the present invention, in the title extracting method according to the fourth aspect, when it is determined that another line exists next to the target line, the character of the target line and the character of the another line are determined. Judgment is made as to whether there is a difference in size, and if it is judged that there is a difference in size, a line with a small character size is given a negative title-like point (assuming it is not independent) and a line with a large character size Is a title extraction method, which comprises the steps of giving a point of title-likeness (assuming that it may be independent).

According to a sixth aspect of the present invention, in the title extracting method according to the first aspect, a difference in character size between a line before and after the target line is measured, and the difference in character size exceeds a predetermined value. It is a title extraction method comprising the steps of: determining whether or not a title is exceeded and, if it is determined to be exceeded, giving a title-like point to a target line.

According to a seventh aspect of the present invention, in the title extracting method according to the sixth aspect, it is determined whether or not the character size has changed so as to decrease the size by looking at the line from top to bottom. For the lines that are changing to become, give points of negative title-likeness,
This is a title extraction method comprising the steps of giving a title-like point to a line above the title.

According to an eighth aspect of the present invention, in the title extracting method according to the sixth aspect, it is determined whether or not the character size has changed so as to increase the character size by looking at the line from top to bottom. Give a point of the title-likeness to the line determined to be changing so that
A title extracting method characterized by comprising each step of giving a negative title-like point to a line above it.

According to a ninth aspect of the present invention, there is provided a method for extracting a title by extracting a character string region from a document image in a rectangular shape and adding title-like points based on the attribute of the character string region. Larger than the other lines,
Moreover, it is characterized in that each step determines whether or not the lines appear at equal intervals, and gives a point of the title-likeness to the line determined to have a larger space and also appears at equal intervals. This is a title extraction method.

According to a tenth aspect of the present invention, in the method for extracting a title by extracting a character string area from a document image in a rectangular shape, adding a point of title likeness based on the attribute of the character string area, and extracting a title. It is determined whether or not it is surrounded by a frame, and when it is determined that it is surrounded by a frame, a title extraction is characterized by comprising a step of giving a title-like point to the target line. Is the way.

According to an eleventh aspect of the present invention, in the title extracting method according to the tenth aspect, it is determined whether or not a plurality of character areas exist in the frame (for example, multi-column), and it is determined that a plurality of character areas exist. In such a case, the judgment of being surrounded by a frame is invalidated (that is, the point of the previous title is cleared, and the title is extracted without the information surrounded by the frame). It is an extraction method.

According to a twelfth aspect of the present invention, in the method for extracting a title by extracting a character string region from a document image in a rectangle and adding a point of the title likeness based on the attribute of the character string region, A title extracting method characterized by comprising a step of determining whether or not a screen is shaded and giving a point of title-likeness to a line determined to be shaded.

According to a thirteenth aspect of the present invention, in a method for extracting a title by extracting a character string area from a document image in a rectangular shape, adding points of the title likeness based on the attribute of the character string area, and extracting a title. It is determined whether the number of characters is smaller than that of the other lines and the centering is performed,
When it is determined that the number of characters is smaller than that of the other lines and that it is centered, a title which comprises a step of giving a negative point of title-likeness to a line below the line based on the line. It is an extraction method.

According to a fourteenth aspect of the present invention, there is provided a method for extracting a title by extracting a character string area from a document image in a rectangular shape, adding a title-like point based on the attribute of the character string area, and extracting the title. By comparing with the specific word dictionary, it is determined whether or not the target line is composed of only specific characters (for example, “notation”). If it is determined that the target line is composed of only specific characters, And a step of separately giving title-like points to lines above and below the line based on the line.

According to a fifteenth aspect of the present invention, there is provided a program for causing a computer to execute the method according to any one of the first to fourteenth aspects.

According to a sixteenth aspect of the present invention, there is provided a computer-readable recording medium on which the program according to the fifteenth aspect is recorded.

According to a seventeenth aspect of the present invention, there is provided a title extracting apparatus including the program according to the fifteenth aspect.

[0026]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The extraction of a title from a document image according to the present invention will be described below with reference to the accompanying drawings.

FIG. 1 is a block diagram showing a system configuration for performing a title (region) extraction process according to an embodiment of the present invention. In the figure, reference numeral 101 denotes an area identification unit as an area identification unit that cuts out a character string area from a document image input from an image input device (not shown) such as a facsimile or an image scanner, and 102 denotes identification of the area identification unit 101. A character recognition unit as a character recognition unit that performs character recognition based on the result, 103 is a font identification unit as a font identification unit that performs font identification based on the identification result of the area identification unit 101,
104 is a natural language analysis unit as a natural language analysis means for analyzing natural language title likelihood based on the character code obtained by the recognition result of the character recognition unit 102, and 105 is centering, underlining, character rectangle size, and shading , A point-attaching unit as a point-attaching means for assigning a point of a title likeness using a frame or the like. Reference numeral 106 denotes a specific word dictionary in which specific characters referred to in extracting a title are stored.

In the configuration shown in FIG. 1, when a document image is input from an image input device (not shown), preprocessing such as skew correction is performed, an area identification process is executed by an area identification unit 101, and a character string is output. Obtain information on the coordinates and size of the rectangle. Next, using the result of the area identification processing by the area identification unit 101, character recognition by the character recognition unit 102 and font identification by the font identification unit 103 are performed.

In the character recognizing unit 102, a character code, a certainty factor, a coordinate value and a size of a character rectangle for each character can be obtained as a point of the likeness of a title. Also, font identification unit
In 103, the font type for each character is obtained as a point of the title-likeness.

The character code obtained by the character recognizing unit 102 is also supplied to a natural language analyzing unit 104 and a natural language analyzing routine. give. Furthermore, in natural language processing, the statistical information dictionary of endings frequently appearing in the title is compared with the character code string in the character area, and the character string area including the ending that matches the frequent occurrence of the ending in the character string area has the point of title-likeness. give.

Further, in addition to the above-described points of the likelihood of points, the total points of the likelihood of the title are calculated using the conventionally used centering processing, underlining processing, the size of the character string rectangle, etc., and the title is identified. .

Next, a series of title extracting methods of the present invention will be described in order with reference to flowcharts shown in FIGS. It should be noted that this title extraction method can be performed in a combination of a plurality of methods, alone, or selectively according to the configuration of FIG.

(Embodiment 1) In this embodiment, a line-to-line distance of a character line is extracted as a feature, and a line having a change in the line-to-line distance is given a point of a title likeness. For example, when the top line is regarded as a line distance from the upper end of the image, it is determined that the line has a large line distance, and a title-like point is attached.

FIG. 3 is a flowchart showing an execution procedure of the title extracting method according to the first embodiment. First, a document image is input from a document input device (not shown) (S301), and a character string area is identified by the area identification unit 101 (S302). Subsequently, the line spacing between the character strings in the character string area is measured (S303), and when the difference between the lines before and / or after the character string is larger than a predetermined line spacing difference, that is, when the other line spacing is used. If the distance between the lines greatly changes (S304, Yes), a title-like point is given to the line. By summing up these points, a title area is determined and extracted as a title (S305).

(Second Embodiment) In a second embodiment, in addition to the processing procedure in the first embodiment, a point of the likelihood of a title is given based on the certainty factor of the character as a result of character recognition. Here, a character string with low confidence is nothing but low reliability as a line, so if such a line is extracted as a title, it becomes a meaningless character string and can be a valid keyword. Absent. Therefore, a certain threshold value is obtained for the certainty factor of the character recognition result, and when the certainty factor is equal to or higher than the certain threshold value, a point of the title-likeness is given. By adding information that does not seem to be a title, that is, by giving a point of minus title likeness, a correct title can be extracted.

FIG. 4 is a flowchart showing an execution procedure of the title extracting method according to the second embodiment. The processing procedure for title extraction will be described with reference to the figure. First, a document image is input from a document input device (not shown) (S40).
1), a character string area is identified by the area identification unit 101 (S40)
2). Subsequently, the distance between lines of the character string in the character string region is measured (S403), and when the line distance difference before and after the character string is larger than a predetermined line distance difference, that is, when the line distance greatly changes (S404, Yes) If the certainty factor of the character code identification is equal to or more than a certain threshold (S405, Yes), a title-like point is given to the line, and the points are summed to determine a title area and extracted as a title. (S406). If the certainty of the character code identification does not reach a certain threshold value, information that the character is unlikely to be a title is added. The extraction accuracy of

(Third Embodiment) In a third embodiment, a centering position in a document is obtained, and when it is determined that the target line is centered, a point similar to a title is given. Conventionally, when the distances from the left and right edges of a row to the left and right edges of an image are equal, it is most commonly determined that the row is centered. However, in the third embodiment, not only that, The entire image is searched for the maximum and minimum coordinate values of a certain range of the character area, the character existence range is found, and it is checked whether the character is centered within the character existence range. This allows, for example, B5
If you use an image copied without enlarging the image to A4, you can find the centering position in a stable area called the character existence range, even though the centering position is shifted from normal. , Features can be taken stably.

FIG. 5 is a flowchart showing the execution procedure of the title extracting method according to the third embodiment. The processing procedure for title extraction will be described with reference to the figure. First, a document image is input from a document input device (not shown) (S50).
1), the character string area is identified by the area identification unit 101 (S50)
2). Next, it is determined whether or not the target line is centered in the document (S503). If the line is centered (S503, Yes), the maximum and minimum coordinate values within a certain range of the character area are searched, and the character is searched. The existence range is identified (S504), and it is determined whether or not the identified character string area is centered in the character existence range (S505).
05, Yes), a title-like point is given to the line, and the points are summed to determine a title area and extracted as a title (S506).

Fourth Embodiment In the fourth embodiment, the independence of a row is considered. That is, when there is no other row next to the row, that is, left and / or right, the row is located alone. That is, I think there is a title. On the other hand, the case where there is another line besides the case corresponds to the case where a plurality of lines are described in parallel, and it is difficult to consider that the lines themselves are independent. Therefore, in such a case, the point of the title likeness is reduced.

FIG. 6 is a flowchart showing an execution procedure of the title extracting method according to the fourth embodiment. The processing procedure for extracting a title will be described with reference to the figure. First, a document image is input from a document input device (not shown) (S60).
1), a character string area is identified by the area identification unit 101 (S60)
2). Next, it is determined whether there is a line of the character string on the left or right (left or right) of the identified character string (S603), and if there is no other line (S6).
03, No), giving that line a title-like point,
By summing up these points, a title area is determined and extracted as a title (S604). If there is another line of the character string (S603, Yes), a negative point of the title likeness is given (S605).

(Fifth Embodiment) In the fifth embodiment, in addition to the fourth embodiment, considering the balance of the line size and the like, when the font sizes are the same, it is determined that the fonts are written together, If the font size is large, it is difficult to imagine that it is written together. Therefore, it is considered that the line with the large font size is actually independent, and gives a point of title-likeness to it. Furthermore, the smaller the font size, the less likely it is to become the title, so the points are reduced.

FIG. 7 is a flowchart showing the execution procedure of the title extracting method according to the fifth embodiment. The processing procedure for title extraction will be described with reference to the figure. First, a document image is input from a document input device (not shown) (S70).
1), the character string area is identified by the area identification unit 101 (S70)
2). Next, it is determined whether there is a line of the character string on the left or right (left or right) of the identified character string (S703), and if there is no other line (S7).
03, No), giving that line a title-like point,
By summing up these points, a title area is determined and extracted as a title (S704). Up to this point, the operation is the same as in the fourth embodiment. Next, if there is another line beside (S703, Ye
s) and measure the font size of each line (S70
5) If the font size of the line of the identified character string is larger than the font size of the characters of the left and right character strings (S706,
Yes), a title-like point is given to the line, and the points are summed to determine a title area and extracted as a title (S704). If the font size of the line of the identified character string is not larger than the font size of the character of the horizontal character string (S706, No), a minus point of title-likeness is given to the line (S707).

In the sixth, seventh and eighth embodiments, information on the character size is used. In addition to simply adding a large character size as a title-like point, the difference in character size between the lines above and below it is captured as a feature. Since the title is often written in a conspicuous manner, it is often the case that the style is deliberately changed from the preceding and following lines, as in the first embodiment. Therefore, either the line spacing or the character size often changes, and even if the font size is not as large as the overall size, the font size is clearly different from the previous and next lines. There are many cases where titles exist at positions where is noticeable.

(Embodiment 6) FIG. 8 is a flowchart showing an execution procedure of a title extracting method according to Embodiment 6. The processing procedure for title extraction will be described with reference to the figure. First, a document image is input from a document input device (not shown) (S801), and a character string area is identified by the area identification unit 101 (S802). Next, it is determined whether or not there is a line of the character string in the horizontal direction of the identified character string (S803). If there is no other line (S803, No), a point of title-likeness is given to that line, and this point is The title area is determined by summing up and extracted as a title (S804). The steps up to here are the same as in the fourth and fifth embodiments. Next, if there is another line on the left and right (S803, Yes), the font size of each line is measured (S805), and the line of the extracted character string and the font size are larger than the font size of the character on the horizontal line. If the difference is larger and the difference is larger than a predetermined value, that is, if the difference is larger (S806, Yes), a title-like point is given to the line, and the points are summed to determine a title area and the title area is determined. It is extracted (S804). If the difference between the font sizes is not large in step 806 (S806, No), a minus point of title-likeness is given (S807).

(Embodiment 7) Embodiment 7 is different from Embodiment 6
In the title extraction method of, when the line is viewed from the top to the bottom, for a line that changes so that the character size becomes smaller, a minus point is given to the point of title-likeness, and the line above it is given This is a title extraction method characterized by giving a point of title-likeness. FIG. 9 is a flowchart illustrating an execution procedure of the title extracting method according to the seventh embodiment. The processing procedure for extracting the main title will be described with reference to the figure. First, a document image is input from a document input device (not shown) (S901), and a character string area is identified by the area identification unit 101 (S902). Next, the character size of the identified character string area is measured (S903), and if the character of the character string in the extracted line is smaller than the character size of the character string in the upper line (S904, Yes), Giving (adding) negative points of title-likeness (S905),
A title-like point is given to the upper line, and a title area is determined by adding up the points, and is extracted as a title (S906).

(Embodiment 8) The embodiment 8 is similar to the embodiment 6
In the title extraction method described in, when looking at the line from top to bottom, give a point of title-likeness to the line that changes so that the character size increases, and give the line above it Give the points of title-like a minus point. FIG. 10 is a flowchart illustrating an execution procedure of the title extracting method according to the eighth embodiment.
The processing procedure for extracting the title will be described with reference to the figure. First, a document image is input from a document input device (not shown) (S1001), and a character string area is identified by the area identification unit 101 (S1002). Next, the character size of the identified character string area is measured (S1003), and if the character of the character string in the extracted line is larger than the character size of the character string in the upper line (S1003).
4, Yes), give the line a point of title-likeness,
By summing up these points, a title area is determined and extracted as a title (S1005), and a minus point of title-likeness is given to the upper line (S1006).

(Embodiment 9) A ninth aspect is a process of giving a point of a title-like character to a place where a space occupied by characters has a large proportion and characters appear at equal intervals. FIG. 11 is a flowchart illustrating an execution procedure of the title extracting method according to the ninth embodiment. The processing procedure for extracting the title will be described with reference to the figure. First, a document image is input from a document input device (not shown) (S110).
1), a character string area is identified by the area identification unit 101 (S110)
2). Next, the space between the characters of the identified character string is measured (S1103), and if the space between the characters of the target line is larger than the space between the characters of the other lines at equal intervals (S1103).
104, Yes), a title-like point is given to the line, and the points are summed to determine a title area and extracted as a title (S1105).

(Embodiment 10) The embodiment 10 is a title extracting method utilizing the fact that there are many titles in a portion surrounded by an enclosing frame as a layout feature other than the underline. FIG. 12 is a flowchart illustrating an execution procedure of the title extracting method according to the tenth embodiment. The processing procedure for extracting the title will be described with reference to the figure. First, a document image is input from a document input device (not shown) (S1201), and a character string area is identified by the area identification unit 101 (S1202). Next, if the identified character string is surrounded by a frame (S1203, Yes), a title-like point is given to the line, and the points are summed to determine a title area and extracted as a title (S1204). ).

(Eleventh Embodiment) In the eleventh embodiment, in the title extraction processing of the tenth embodiment, if the entire document is surrounded by a frame, the information on the frame is useless. This is a title extraction method that starts over. FIG. 13 is a flowchart illustrating an execution procedure of the title extracting method according to the eleventh embodiment. The processing procedure for extracting the main title will be described with reference to the figure. First, a document image is input from a document input device (not shown) (S1301), and a character string area is identified by the area identification unit 101 (S1302). Next, if the identified character string is surrounded by a frame (S1303, Yes), it is determined whether or not a plurality of character areas exist in the frame (S1304). (S1304, No), a title-like point is given to the target line, and the points are summed to determine a title area and extracted as a title (S1304). If there are a plurality of character areas in the frame (S1304, Yes), the other processing, in this case, the processing in which the frame is surrounded by the frame in the tenth embodiment is invalidated (reset). That is, processing is performed assuming that there is no frame.

(Twelfth Embodiment) A twelfth embodiment is a title extracting method utilizing the fact that there are many titles in a shaded line as a layout feature other than the underline. Especially in Japanese documents, shading is often used for emphasis, which is very effective. FIG. 14 is a flowchart illustrating an execution procedure of the title extracting method according to the twelfth embodiment. The processing procedure for extracting the title will be described with reference to the figure. First, a document image is input from a document input device (not shown) (S1401), and a character string area is identified by the area identification unit 101 (S1402). Next, it is determined whether or not the identified character string is woven (S1403). If the character string is shaded (S1403Yes), a title-like point is given to the target line, and the points are totaled. Thus, a title area is determined and extracted as a title (S1404).

(Thirteenth Embodiment) In the thirteenth embodiment, if a line having a small number of characters and being centered is present in a sentence, it is a line such as a letter such as a letter written in a notice. This is a title extraction method characterized by giving a minus point of the title-likeness to a line below it, since a title is usually written above it. FIG. 15 is a flowchart illustrating an execution procedure of the title extracting method according to the thirteenth embodiment. The processing procedure for extracting the title will be described with reference to the figure. First, a document input device (not shown)
, A document image is input (S1501), and a character string area is identified by the area identification unit 101 (S1502). Next, the number of characters in the line of the identified character string is measured (S1503), and it is determined whether the number of characters in the target line is smaller and centered than in the other lines (S1504). If it is a line that has been set (S1503 Yes), a title-like point is given to the target line, and the points are summed to determine a title area and extracted as a title (S1505).

(Embodiment 14) In Embodiment 14, unlike Embodiment 13, a word having a characteristic as a break between sentences such as "" is stored as a dictionary of specific words.
It is characterized in that the point of the title-likeness is changed with a line composed of only specific characters as a boundary. For example, words such as "dono" at the end of the sentence are examples of cases in which it is difficult to become a title in itself. It is possible to search for documents, collect such keywords, and save them as a database to improve accuracy.

FIG. 16 is a flowchart showing an execution procedure of the title extracting method according to the fourteenth embodiment. The processing procedure for extracting the main title will be described with reference to the figure. First, a document image is input from a document input device (not shown) (S1601), and a character string area is identified by the area identification unit 101 (S1602). Next, it is determined whether or not the extracted character string is composed only of special characters registered in the specific word dictionary (S1603), and if it is composed only of specific characters (S1603).
(3, Yes), giving a different title-like point to the upper and lower lines with that line as a boundary (S1604).

The embodiment of the title extracting method according to the present invention has been described above. A series of procedures for executing each of the above methods is described as a computer program, and this program is written on a flexible disk, CD-ROM, DVD-ROM,
A method for extracting a title from an input document image by automatically recording a title on an input document image by configuring a title extracting device for a document image by recording the image on an arbitrary recording medium such as an MO and reading the same by an arbitrary computer. Can be implemented.
In addition to recording this program on a recording medium, it is of course possible for a computer to directly read the program via an arbitrary network such as the Internet or an intranet.

[0055]

According to the first to fourteenth aspects, a character string area having a large number of points can be automatically extracted as a title with high accuracy, and the title can be accurately extracted. The convenience at the time of searching for a document can be improved. Effects corresponding to Claims 15 and 16: By allowing any computer to read, it is possible to easily perform a process of extracting a title from an input document image. Effect corresponding to claim 17: A process of extracting a title from an input document image can be easily performed.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating an example of a system configuration for performing a title extraction process.

FIG. 2 is a diagram showing a title likelihood that is secondarily obtained among points of the title likeness used in the title extraction process.

FIG. 3 is a flowchart according to a first embodiment of the title extracting method of the present invention.

FIG. 4 is a flowchart according to a second embodiment of the title extracting method of the present invention.

FIG. 5 is a flowchart according to a third embodiment of the title extracting method of the present invention.

FIG. 6 is a flowchart according to a fourth embodiment of the title extracting method of the present invention.

FIG. 7 is a flowchart according to a fifth embodiment of the title extracting method of the present invention.

FIG. 8 is a flowchart according to a sixth embodiment of the title extracting method of the present invention.

FIG. 9 is a flowchart according to a seventh embodiment of the title extracting method of the present invention.

FIG. 10 is a flowchart according to an eighth embodiment of the title extracting method of the present invention.

FIG. 11 is a flowchart according to a ninth embodiment of the title extracting method of the present invention.

FIG. 12 is a flowchart according to a tenth embodiment of the title extracting method of the present invention.

FIG. 13 is a flowchart according to an eleventh embodiment of the title extracting method of the present invention.

FIG. 14 is a flowchart according to a twelfth embodiment of the title extraction method of the present invention.

FIG. 15 is a flowchart according to a thirteenth embodiment of the title extraction method of the present invention.

FIG. 16 is a flowchart according to a fourteenth embodiment of the title extraction method of the present invention.

[Explanation of symbols]

101: area identification unit, 102: character identification unit, 103: font identification unit, 104: natural language identification unit, 105: point attaching unit, 106: specific word dictionary

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 5B050 AA10 BA10 BA16 EA03 EA04 5B064 AA01 AB19 BA01 CA08 EA27 EA28 EA32 5B075 ND07 NK04 NK07 NK31 PP04 PR06 QM08

Claims (17)

[Claims] 1. A method for extracting a title by extracting a character string region from a document image in a rectangle and adding a point of a title likeness based on an attribute of the character string region, and before and / or after a target line. A title extraction method comprising: measuring a line spacing of a character line, comparing the measured line spacing with a predetermined value, and giving a point of title-likeness to a character line larger than the predetermined value. . 2. The title extraction method according to claim 1, wherein a character of a character line in the character string area is recognized, and it is determined whether or not the degree of certainty of the character recognition result is lower than a predetermined value. A title extracting method, comprising: giving each character line a negative title-like point when the character line is low. 3. The title extracting method according to claim 1, wherein a centering position in the document is obtained, and it is determined whether or not the target line is at the centering position. Determining a centering position based on the character region portion, and determining that the target line is at the centering position based on the character region portion, giving a point of title-likeness to the character line. Title extracting method characterized by comprising steps. 4. The title extracting method according to claim 1, wherein it is determined whether or not another line exists next to the target line, and when it is determined that another line exists, a minus is determined. A title extraction method comprising the steps of giving title-like points. 5. The title extracting method according to claim 4, wherein when it is determined that another line exists next to the target line,
It is determined whether there is a difference in the character size between the target line and the another line, and when it is determined that there is a difference in the size, a line having a small character size is given a point of negative title likeness, and the character size is changed. A title extraction method comprising the steps of: giving a point of title-likeness to a line having a large size. 6. The title extracting method according to claim 1, wherein a difference in character size between a line before and after the target line is measured, and it is determined whether or not the difference in character size exceeds a predetermined value. And a step of giving the point of the title likeness to the target line when it is determined that the title is exceeded, comprising the steps of: 7. The title extracting method according to claim 6, wherein the line is viewed from top to bottom, and it is determined whether or not the character size is changed so as to be reduced. A title extracting method, comprising: giving a negative title-like point to a current line, and giving a title-like point to a line above the negative line. 8. The title extracting method according to claim 6, wherein the line is viewed from top to bottom, and it is determined whether or not the character size is changed so as to increase, and the character size is changed so as to increase. A title extraction method comprising: giving a title-like point to a line determined to be present; and giving a minus title-like point to a line above the title extraction method. 9. A method of extracting a title by extracting a character string region from a document image by a rectangle and adding a point of a title likeness based on the attribute of the character string region, wherein a space between characters is larger than other lines. And determining whether or not the lines appear at equal intervals, and giving points of title-likeness to the lines determined to have larger spaces and appearing at equal intervals. Title extraction method. 10. A method of extracting a title by extracting a character string area from a document image in a rectangle and adding a point of a title likeness based on the attribute of the character string area, wherein a target line is surrounded by a frame. A title extraction method comprising the steps of: giving a point of title-likeness to the target line if it is determined whether the line is surrounded by a frame. 11. The title extracting method according to claim 10, wherein it is determined whether or not a plurality of character areas exist in the frame, and when it is determined that a plurality of character areas exist, it is determined that the character area is surrounded by the frame. A title extraction method including steps for invalidating the judgment. 12. A method for extracting a title by extracting a character string region from a document image in a rectangle and adding a point of a title likeness based on the attribute of the character string region, wherein a target line is shaded. A title extraction method comprising: determining whether or not a line is shaded and giving a title-like point to a line determined to be shaded. 13. A method for extracting a title by extracting a character string region in a rectangle from a document image and adding a point of a title likeness based on the attribute of the character string region, wherein the target line is the number of characters from other lines. Is determined and whether or not it is centered, and when it is determined that the number of characters is smaller than that of the other lines and that it is centered, the title likeness is placed in a line below the line based on the line. A title extraction method comprising the steps of giving a negative point. 14. A method for extracting a title by extracting a character string area from a document image in a rectangular shape, adding a point of a title likeness based on the attribute of the character string area, and comparing with a specific word dictionary existing in the method. By doing so, it is determined whether or not the target line is composed of only specific characters, and when it is determined that the target line is composed of only specific characters, the line above the line is determined based on that line. A title extraction method comprising: a step of separately giving a title-like point to a line and a line below the line. 15. A program for executing the method according to claim 1 on a computer. 16. A computer-readable recording medium on which the program according to claim 15 is recorded. 17. A title extracting device equipped with the program according to claim 15.