JP2004038756A - Document conversion method and document conversion device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a document conversion method capable of realizing highly accurate conversion from static document type data to dynamic document type data. <P>SOLUTION: Character information of each character is extracted from the inputted static document data (S11), and a plurality of area dividing candidates are prepared on the basis of the extracted character information (S12). Line extraction in each area is carried out per each area dividing candidate, and character strings are prepared by searching extracted lines (S13). A transition probability of all character strings in one area dividing candidate is calculated (S14), and on the basis of the transition probability per each area dividing candidate, an area with the highest transition probability of all character strings is determined as an optimum area (S15). Line extraction in each area is carried out again with respect to the optimum area, and an extracted line is searched to prepare the correct sentence (S16). By this, the static document data is converted into the dynamic document data including the correct sentence. <P>COPYRIGHT: (C)2004,JPO

Description

【００７４】
式（１）より、任意長の文字列Ｃに関する遷移確率値Ｐ（Ｃ）は、２文字組（ｄｉｇｒａｍ）と３文字組（ｔｒｉｇｒａｍ）との出現頻度（または出現確率）テーブルを用意することで得ることができる。ただし、対象文字コードが複数バイトコードの場合（日本語などの場合）、文字のカテゴリ数が数千オーダーとなり、特にｔｒｉｇｒａｍを確保するためのメモリ量の確保が難しい。そのため、式（１）を直接採用することは適当ではない。したがって、本実施の形態においては、ｄｉｇｒａｍと１文字との出現頻度（ｕｎｉｇｒａｍ）を使用した近似式である次の式（２）にて文字列の遷移確率値Ｐを計算する。
【００７５】
【数２】

【００７６】
図９に示される領域分割候補文字列バッファごとに、その中に存在する文字列数５１１で示される複数の文字列バッファ５１２〜５１５に対して、個々に式（２）で遷移確率値Ｐを計算し、それらの積により該当領域候補内の文字列全体の文字列遷移確率を計算する。具体的には、領域分割候補内の文字列数をＬ、各文字列をＣ１，Ｃ２，・・・ＣＬとすると、次の式（３）により領域分割候補文字列全体の文字列遷移確率を計算する。
【００７７】
【数３】

【００７８】
ステップＳ１４では、文字列確率計算部１０１４は、全ての領域分割候補に対して、式（３）により文字列遷移確率を計算し、ステップＳ１５で最適領域決定部１０１５がその文字列遷移確率が最大となる領域分割候補を最適な領域分割結果として採用し、ステップＳ１６で正解文書作成部１０１６が正解の文字列を作成する。なお、ステップＳ１６での正解の文字列を作成する方法は、図８に示される文字列作成処理とほぼ同様の方法である。すなわち、図８のステップＳ１３１，Ｓ１３２、およびＳ１４１をなくし、領域分割候補カウンタＪを最適と判断された領域分割候補番号とすれば、図８に示される処理と同じ処理で正解の文字列を作成することができる。
【００７９】
本実施の形態におけるドキュメント処理装置であるＰＣ１で以上の処理が実行されることで、静的ドキュメント形式から動的ドキュメント形式への、高精度な変換を実現することができる。
【００８０】
［第２の実施の形態］
次に、図１０は、第１の実施の形態において図５に示される入力画像３０１と同様のドキュメントに対する領域分割候補と各領域分割に対する行抽出との具体例を示す図である。
【００８１】
図１０を参照して、領域分割候補６１〜６３は、第１の実施の形態において領域分割に用いる距離パラメータδを変化させた場合に得られる領域分割候補を示す。さらに、分割領域６１１〜６１３は領域分割候補６１における分割領域を示し、分割領域６２１，６２２は領域分割候補６２における分割領域を示し、分割領域６３１は領域分割候補６３における分割領域を示す。また、各分割領域内の太線の横長矩形は行を、その中の細線の矩形は文字を示している。なお、図５に示されるように、領域分割候補６１〜６３は、横書きの３カラムのドキュメントに対する領域分割候補であるため、領域分割は、分割領域６１１〜６１３の３領域からなる領域分割候補６１が正解となる。
【００８２】
第１の実施の形態においては、各領域分割候補内の各領域に対して行抽出を行ない、文字を連結して文字列を作成し、文字列遷移確率により最適な分割候補を決定した後に、再度、その最適な分割候補（図１０では、領域分割候補６１）に対して文字列を抽出し、正解の文章を作成した。このような、第１の実施の形態における処理が有効に動作する理由は、不正解の領域分割候補より抽出した文字列では別領域の文字が連結されるため、文字列としての遷移確率が低下するからである。例えば、図１０においては、分割領域６２１や分割領域６３１に示されるように別領域の文字が連結され、その結果、別領域間をまたがる部分の文字列が言語的に意味不明な文字列となる。そのため、文字列としての遷移確率が低下する。
【００８３】
そこで、第２の実施の形態においてはこのことをさらに利用し、第１の実施の形態におけるドキュメント処理装置と同様のＰＣ１において、より高精度の変換を行なう場合について説明する。
【００８４】
具体的に、図１０に示される全ての領域分割候補６１〜６３での行抽出結果において、他の文字連結と異なる部分の文字連鎖だけを評価した方がより有効に正解の文章が作成できることは、上述の式（２）よりも明らかである。すなわち、図１１に、図１０に示される領域分割候補６１〜６３を比較した場合の、文字連結と異なる部分の文字を暗転して示す。図１１に示されるように、領域間の境界となる文字が連結の異なる文字である。そのため、領域候補ごとにその文字を含めた前後数文字で構成される文字列を作成し、その文字列の遷移確率値Ｐを式（２）で求め、それらを式（３）で評価すればよいことになる。
【００８５】
第２の実施の形態においてＰＣ１で行なわれる変換処理は、第１の実施の形態において図３に示される変換処理とほぼ同様であるため、ここでの、同様の部分についての説明は繰返さない。第２の実施の形態においては、図３のステップＳ１３での文字列作成部１０１３で実行される文字列作成処理が、第１の実施の形態における処理とは異なる。そこで、第２の実施の形態における文字列作成処理について、図１２にフローチャートを示す。なお、図１２にフローチャートが示される文字列作成処理は、図８にフローチャートが示される第１の実施の形態における文字列作成処理とほぼ同様である。すなわち、図１２を参照して、ステップＳ２３１〜Ｓ２３４における処理は、図８のステップＳ１３１〜Ｓ１３４における処理と同様である。そこで、以降においては、ステップＳ２３５の処理以降について説明する。
【００８６】
図１２を参照して、文字カウンタＩで示される文字の連結が、全ての領域候補で同じかどうかを判断することにより、連結が異なる文字を判定する（Ｓ２３５）。具体的には、該当文字情報の領域・行リストの全ての領域・行情報の前文字情報アドレスと次文字情報アドレスとが全て同じ場合でない場合（１つでも異なるアドレスが付与されている場合）に、該当文字を連結が異なる文字と判断する。
【００８７】
１つでも連結が異なる場合は（Ｓ２３５でＹＥＳ）、その文字から所定の文字数であるＮ文字まで前の全文字コードを抽出する（Ｓ２３６）。これは、文字情報の前文字情報アドレスを走査することで可能である。同様に、文字情報の次文字情報アドレスを走査することで、その文字からＮ文字まで次の全文字コードも抽出する（Ｓ２３７）。
【００８８】
次に、ステップＳ２３６およびＳ２３７で抽出された文字コードを、それらの連結順に並べて文字列を作成し、文字列バッファ１０２１へ格納する（Ｓ２３８）。その後、図８のステップＳ１４０以降の処理と同様に、文字カウンタＩをインクリメントし（Ｓ２３９）、ステップＳ２３４へ戻る。また、ステップＳ２３５において文字カウンタＩで示される文字の連結が、全ての領域候補で同じである場合にも（Ｓ２３５でＮＯ）、同様に文字カウンタＩをインクリメントし（Ｓ２３９）、ステップＳ２３４へ戻る。すなわち、次の行に対してステップＳ２３４〜Ｓ２３８の処理を繰返す。
【００８９】
このように、本実施の形態において、上述の、連結が異なる文字の近傍の文字で構成される文字列だけを評価する文字列作成処理を行なうことで、第１の実施の形態における変換処理よりもさらに高精度の領域判定が実現可能となる。
【００９０】
さらに、上述のドキュメント変換装置が行なう変換方法を、プログラムとして提供することもできる。このようなプログラムは、コンピュータに付属するフレキシブルディスク、ＣＤ−ＲＯＭ、ＲＯＭ、ＲＡＭおよびメモリカードなどのコンピュータ読取り可能な記録媒体にて記録させて、プログラム製品として提供することもできる。あるいは、コンピュータに内蔵するハードディスクなどの記録媒体にて記録させて、プログラムを提供することもできる。また、ネットワークを介したダウンロードによって、プログラムを提供することもできる。
【００９１】
提供されるプログラム製品は、ハードディスクなどのプログラム格納部にインストールされて実行される。なお、プログラム製品は、プログラム自体と、プログラムが記録された記録媒体とを含む。
【００９２】
今回開示された実施の形態はすべての点で例示であって制限的なものではないと考えられるべきである。本発明の範囲は上記した説明ではなくて特許請求の範囲によって示され、特許請求の範囲と均等の意味および範囲内でのすべての変更が含まれることが意図される。
【図面の簡単な説明】
【図１】本実施の形態におけるドキュメント処理装置であるＰＣ１の構成の具体例を示すブロック図である。
【図２】変換処理を行なうためのＰＣ１の構成を示すブロック図である。
【図３】本実施の形態のＰＣ１で行なわれる変換処理について示すフローチャートである。
【図４】文字情報バッファ１０２１の具体例を示す図である。
【図５】領域３１１〜３１３を含む入力画像３１に対する、結果表示画像３２を示す図である。
【図６】ステップＳ１２で実行される領域分割候補の作成処理について示すフローチャートである。
【図７】連結関係の具体例を示す図である。
【図８】ステップＳ１３で実行される文字列作成処理を示すフローチャートである。
【図９】文字列バッファ１０２２の具体例を示す図である。
【図１０】第１の実施の形態において図５に示される入力画像３０１と同様のドキュメントに対する領域分割候補と各領域分割に対する行抽出との具体例を示す図である。
【図１１】図１０に示される領域分割候補６１〜６３を比較した場合の、文字連結と異なる部分の文字を暗転して示した図である。
【図１２】第２の実施の形態における文字列作成処理を示すフローチャートである。
【符号の説明】
１　ＰＣ、１０　静的ドキュメントデータ、１１　文字コード、１２　開始ｘ座標、１３　開始ｙ座標、１４　ｘサイズ、１５　ｙサイズ、１６　フォント、１７　属性情報、１８　領域・行情報リスト、２０　動的ドキュメントデータ、３１　入力画像、３２　結果表示画像、４１　先頭文字、４２　最終文字、５１〜５５　領域分割候補文字列バッファ、６１〜６３　領域分割候補、１０１　制御部、１０２　記憶部、１０３　入力部、１０４　出力部、１８１〜１８５　領域・行情報、３１１〜３１３，３２１，３２２　領域、５１１　文字列数、５１２〜５１５　文字列バッファ、６１１〜６１３，６２１，６２２，６３１　分割領域、１０１１　文字情報抽出部、１０１２　領域分割候補作成部、１０１３　文字列作成部、１０１４　文字列確率計算部、１０１５　最適領域決定部、１０１６　正解文書作成部、１０２１　文字情報バッファ、１０２２　文字列バッファ、１８１１　領域ラベル、１８１２　前文字情報アドレス、１８１３　次文字情報アドレス。[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a document conversion method and a document conversion device, and more particularly, to a document conversion method and a document conversion device capable of realizing high-precision conversion from static document format data to dynamic document format data.
[0002]
[Prior art]
With the development of information-related technology and document-related technology in recent years, various types of electronic documents are flooding. These digitized document files can be roughly classified into those having positional information of each object such as characters and images and those having no positional information.
[0003]
Representative examples of the former include a document file of DTP (Desktop Publishing) software and a PDF (Portable Document Format) file. Representative examples of the latter include a plain text file (a file containing only text) and an HTML (Hypertext Markup Language) file. In the following description, the former is called a static document format, and the latter is called a dynamic document format.
[0004]
The dynamic document format having no position information is arranged and displayed by a browser (browsing software) dedicated to the file based on a certain rule with characters and images in the document. A certain rule is, for example, a rule for arranging characters in order in the case of plain text and performing a line feed based on a line feed code, and in the case of HTML, an arrangement rule based on each tag.
[0005]
On the other hand, as a platform for browsing various electronic documents, a personal computer has been generally used in the past. However, in the future, a shift to portable terminals such as PDA (Personal Digital Assistants) and PDC (Personal Digital Cellular) will be made. is expected. In fact, even now, Web browsing with a mobile terminal is realized.
[0006]
However, mobile devices have the problem that their screen resolution is smaller than that of personal computers and they are not suitable for displaying static document formats. Currently and in the future, mobile devices are mainly used in dynamic document formats. It is expected to be.
[0007]
For this reason, in the future, with the shift of the digitized document browsing platform to mobile terminals, a conversion technique for converting information in a static document format into a dynamic document format with high efficiency will be important.
[0008]
Conventionally, as a conversion technique of such an electronic document, a technique such as an electronic filing system disclosed in Japanese Patent Application Laid-Open No. 8-147446 has been proposed. The above-described electronic filing system is a system that converts an electronic document into data of the electronic filing system based on information transmitted from an application to a printer driver. That is, the system performs data conversion by performing an operation of “printing” a corresponding document on a printable application using a conversion printer driver.
[0009]
For example, in the case of Windows (R) which is one of the representative OSs, data transferred to a printer driver when an application performs printing is converted into a format unified by a GDI (Graphics Device Interface) which is a kernel module. . Therefore, in the above-described electronic filing system, a printer driver that interprets a common GDI command and performs data conversion is provided, thereby absorbing many different electronic document formats and realizing many-to-one data conversion. are doing.
[0010]
[Problems to be solved by the invention]
However, in a conventional technology such as the electronic filing system disclosed in the above-mentioned Japanese Patent Application Laid-Open No. 8-147446, the character code of the data passed to the printer driver is converted into a series of sentences, particularly in a specific application such as DTP software. As a result, there is a problem that it is difficult to convert data in a static document format to a dynamic document format. The reason for this is that the DTP software expresses a sentence that exceeds the expressive ability of the GDI code and draws one character at a time.
[0011]
To solve this problem, in the related art, a character string, a word, a line, and the like are determined (cut out) using a fixed ratio threshold value such as 1/2 of the character height. Then, there is a problem that performance is reduced.
[0012]
Also, unlike the conventional technology, even in the case of a conversion technology that does not use a printer driver and has a means for directly interpreting the format of the target document, in a static document format such as DTP or PDF, the extracted characters are arranged as text. There is also a problem that it is difficult to convert to a dynamic document format. The reason why the extracted characters are not arranged as sentences is that DTP, PDF, and the like store characters in the order of sentences in the stored data because there is individual position information to be displayed or printed. There is no need for it, and it is stored in a sequence unique to the application that creates the document.
[0013]
Therefore, when converting data in a static document format to a dynamic document format by a printer driver or direct format interpretation, it is necessary to create a correct sentence based on the obtained individual character position information. However, if the text creation is performed by threshold processing such as character size, there is a problem that accuracy is reduced.
[0014]
The present invention has been made in view of these problems, and an object of the present invention is to provide a document conversion method and a document conversion apparatus capable of realizing high-precision conversion from static document format data to dynamic document format data. Aim.
[0015]
[Means for Solving the Problems]
In order to achieve the above object, according to an aspect of the present invention, a document conversion method is a document conversion method for converting document data including character position information into document data not including character position information. A character information extraction step of extracting character information including the position information of each character from the document data including the position information of the character; and a division candidate for dividing the document data into a plurality of regions based on the position information of the extracted characters. A region dividing candidate creating step to be created, and a character string creating step of extracting a line based on the position information of each extracted character for each created region dividing candidate, searching the extracted line, and creating a character string And a character string probability calculating step of calculating a transition probability value of the generated character string. A region determining step of determining a, based on the result of the determined optimum area, and a text creation step of creating a text sentence does not include the location information of each character.
[0016]
According to another aspect of the present invention, a document conversion method is a document conversion method for converting document data including character position information into document data not including character position information, wherein the document data includes character position information. A character information extracting step of extracting character information including position information of each character; and an area division candidate creating step of creating a division candidate for dividing document data into a plurality of areas based on the extracted character position information. For each of the created region division candidates, a line is extracted based on the position information of each extracted character, and the connection of the characters as the extraction result is different for all the created region division candidates. A character string creating step of creating a partial character string consisting of a character and a predetermined number of characters before and after the character, and transition of the created partial character string A character string probability calculation step of calculating a probability value, an area determination step of determining an optimal area from the created area division candidates based on the calculation result, and a character string probability calculation step of each character based on the determined optimal area result. A text creation step of creating a text sentence that does not include position information.
[0017]
In addition, the character information extracting step includes interpreting data in a common format converted based on the document data including the character position information, and directly interpreting the document data including the character position information. It is desirable to extract character information by at least one of the methods.
[0018]
According to another aspect of the present invention, a document conversion apparatus is a document conversion apparatus for converting document data including character position information into document data not including character position information, the document conversion apparatus including a document data including character position information. Character information extraction means for extracting character information including position information of each character, and region division candidate generation for generating division candidates for dividing document data into a plurality of regions based on the extracted character position information A character string creating means for extracting a line based on the position information of each extracted character, searching for the extracted line, and creating a character string for each created region division candidate; A character string probability calculating means for calculating a transition probability value of a column; an area determining means for determining an optimum area from the created area dividing candidates based on the calculation result; Based on the results of such area, and a text creation means for creating a text sentence does not include the location information of each character.
[0019]
According to still another aspect of the present invention, a document conversion device is a document conversion device for converting document data including character position information into document data not including character position information, the document conversion device including a document including character position information. Character information extraction means for extracting character information including position information of each character from data, and area division candidate generation for generating a division candidate for dividing document data into a plurality of areas based on the extracted character position information Means and a line is extracted based on the position information of each extracted character for each of the created region division candidates, and the connection of characters as an extraction result is different even in all of the created region division candidates. A character string creating means for creating a partial character string consisting of a character and a predetermined number of characters before and after the character, and a transition probability of the created partial character string A character string probability calculating means, an area determining means for determining an optimum area from the created area dividing candidates based on the calculation result, and a position information of each character based on the determined optimum area result. And a sentence creating means for creating a text sentence not including
[0020]
Further, the character information extracting means described above interprets data in a common format converted based on document data including character position information, and directly interprets document data including character position information. It is desirable to extract character information by at least one of the methods.
[0021]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the following description, the same parts and components are denoted by the same reference numerals. Their names and functions are the same. Therefore, detailed description thereof will not be repeated.
[0022]
[First Embodiment]
FIG. 1 is a block diagram illustrating a specific example of a configuration of a personal computer (hereinafter, referred to as a PC) 1 which is a document processing apparatus according to the present embodiment.
[0023]
Referring to FIG. 1, PC 1 in the present embodiment is entirely controlled by a control unit 101 including a CPU (Central Processing Unit) and performs various processes. The program executed by the control unit 101 is stored in the storage unit 102. The storage unit 102 also serves as a buffer and a temporary work area when the control unit 101 executes a program. The input unit 103 includes a mouse, a keyboard, and the like, and receives various instructions from the user. In the case where the apparatus is constituted by a reading device, it also receives information from other devices, data stored on a flexible disk or the like. The output unit 104 may output to display information such as a document on a display (not shown). Further, information such as a document may be output to a printer (not shown) as a printer I / F (interface).
[0024]
The configuration of the PC 1 shown in FIG. 1 is a configuration of a general personal computer, and the configuration of the PC 1 is not limited to the configuration shown in FIG. In this embodiment, the document processing apparatus will be described as a personal computer whose configuration is shown in FIG. 1 as an example. However, the present invention is also applicable to a mobile terminal such as a mobile phone or a PDA (Personal Digital Assistants). You may.
[0025]
In the present embodiment, the above-described PC 1 performs a conversion process of converting data in a static document format into data in a dynamic document format. That is, from document format data (hereinafter, referred to as static document data) including position information of each object such as characters and images, to document format data (hereinafter, referred to as dynamic document data) not including position information. ) Is performed.
[0026]
FIG. 2 is a block diagram showing a configuration of the PC 1 for performing the above-described conversion processing. Referring to FIG. 2, input unit 103 receives an input of static document data 10. The control unit 101 detects that the input unit 103 has received the input of the static document data 10, and thereby detects the character information extraction unit 1011, the area division candidate creation unit 1012, and the character string creation unit 1013 included in the control unit 101. , The character string probability calculating unit 1014, the optimal area determining unit 1015, and the correct answer document creating unit 1016 are controlled in conjunction with the character information buffer 1021 and the character string buffer 1022 included in the storage unit 102 to control the static document data 10. It is converted into dynamic document data 20. Then, the converted dynamic document data 20 is output from the output unit 104.
[0027]
FIG. 3 is a flowchart showing a conversion process performed by PC 1 of the present embodiment. The process illustrated in FIG. 3 is realized by the control unit 101 of the PC 1 reading and executing a program stored in the storage unit 102. Hereinafter, the conversion processing shown in FIG. 3 will be described with reference to the block diagram of FIG.
[0028]
First, the PC 1 receives an input of the static document data 10 through the input unit 102 (S10). As described above, the static document data 10 is data in which font information and attribute information such as underline, italic, and bold are stored in addition to the position information.
[0029]
Next, the character information extracting unit 1011 extracts character information of each character from the input static document data 10 (S11). The character information extraction unit 1011 may extract the character information by interpreting data in a common format converted by a kernel module such as GDI, as is implemented in a general printer driver. Then, a method of extracting character information by directly interpreting the static document data 10 may be used, or another method may be used. Here, the method of extracting the character information can be arbitrarily determined. Then, the character information extracted in step S11 is stored in the character information buffer 1021.
[0030]
Next, the area division candidate creating unit 1012 performs area division based on the character information extracted in step S11 to create a plurality of area division candidates (S12). The process of creating the region division candidate in step S12 will be described later in detail with reference to a subroutine.
[0031]
Further, the character string creation unit 1013 extracts lines in each area for each of the area division candidates created in step S12, and extracts the extracted lines from top to bottom in the case of a horizontal writing area, and in the case of a vertical writing area. A line is searched from right to left to create a character string (S13). The line extraction and character string creation processing in step S13 will be described later in detail with reference to a subroutine.
[0032]
Subsequently, the character string probability calculation unit 1014 calculates the transition probabilities of all character strings in one area division candidate created in step S13 (S14). Then, based on the transition probabilities for each of the region division candidates calculated in step S14, the optimal region determining unit 1015 determines the region having the highest transition probability of all character strings as the optimal region (S15). The transition probabilities of all character strings calculated in step S14 will be described later in detail.
[0033]
Next, in the correct answer document creation unit 1016, a line in each region is extracted again with respect to the optimum region determined in step S15, and the extracted lines are arranged from top to bottom in the case of the horizontal writing region, and in the case of the vertical writing region. Searches for a line from right to left, and creates a correct sentence (S16).
[0034]
Finally, the file of the dynamic document data 20 including the correct sentence created in step S16 is output from the output unit 104 (S17).
[0035]
Thus, the conversion process in PC1 is completed. By executing the above-described processing, the static document data 10 input to the PC 1 is converted into dynamic document data 20 and output.
[0036]
Further, a specific example of the character information buffer 1021 for storing the character information extracted from the static document data 10 in step S11 described above will be described with reference to FIG.
[0037]
Referring to FIG. 4, character information buffer 1021 stores information on each character from the first character to the last character in the order of the characters extracted in step S11. In the case of the static document data 10 including the position information of each object, even if the characters are arranged in the extraction order shown in FIG. 4, the text is not a correct answer.
[0038]
The information of each character stored in the character information buffer 1021 includes a character code 11, a starting x coordinate (horizontal axis coordinate) 12, a starting y coordinate (vertical axis coordinate) 13, an x size (horizontal axis size) 14, and y. It comprises a size (vertical axis size) 15, font information 16, attribute information 17, and area / line information list 18. The configuration of the information of each character is not limited to the above configuration. For example, the font information 16 may be the font information itself or an address of a table indicating the font information. Which one to adopt can be determined arbitrarily. In this embodiment, the format is such that font information is stored for each character.
[0039]
The attribute information 17 is information on the attribute of each character, and corresponds to attribute information such as underline, italic, and bold.
[0040]
The area / line information list 18 is secured for each character by the number of area division candidates obtained as a result of the area division in step S12. That is, as a result of the region division in step S12, when m region division candidates are created, the region / line information list 18 includes first region / line information 181 to m-th region / line information 185. Further, one area / line information includes an area label 1811, a preceding character information address 1812, and a succeeding character information address 1813. In the process of creating a region division candidate in step S12, data is stored in the region label of the corresponding region. Specifically, the area label of the corresponding character in the first area division candidate is stored in the area label 1811 of the first area / line information 181. Thereafter, the area label of the corresponding character in each area division candidate is sequentially It is stored in the area label of the corresponding area / line information.
[0041]
Next, a process of creating a region division candidate executed by the region division candidate creation unit 1012 in step S12 described above will be described using a specific example. In the process of creating a region division candidate according to the present embodiment, an improved method of the image region division method disclosed in Japanese Patent No. 3019287 is adopted.
[0042]
The image region dividing method disclosed in the above-mentioned Japanese Patent No. 3019287 is a method mainly used as preprocessing of a character recognition device, and extracts an area by labeling a neighboring rectangle with respect to rectangular coordinates extracted from an image. How to More specifically, for example, when the image is a character “river”, an area having three vertically long rectangular coordinates from the image before recognition is extracted. Next, the distance parameter δ to be integrated is changed within a certain range, and the optimum distance parameter δ is determined by threshold value determination. Then, this is a method of determining (dividing) an area. That is, this is a method of determining an area only by physical rectangular coordinates without using a character code or the like, and is basically a threshold processing. Therefore, an error as shown in FIG. 5 may occur.
[0043]
FIG. 5 shows a result display image 32 for an input image 31 including regions 311 to 313. That is, as shown in FIG. 5, in the above-described image area dividing method, even if the input image 31 in which the three areas are correct (for example, the character “river”), the area 311 and the area 312 Since the distance between them is small, the result display image 32 may be determined to be an image composed of two regions, the region 321 and the region 322, depending on the threshold value employed. Therefore, in the process of creating a region division candidate in the present embodiment, the process of determining the optimal distance parameter δ is not performed, and the labeling result obtained with each distance parameter δ is extracted as each region division candidate. The process of creating a region division candidate according to the present embodiment will be described.
[0044]
FIG. 6 shows a flowchart of the process of creating a region division candidate executed in step S12 described above. Referring to FIG. 6, first, inconvenient characters are excluded from character elements according to a predetermined rule (S121). Inconvenient for the determination is, for example, a large character that cannot be considered as a character of the implementation or a component thereof.
[0045]
Next, an initial value is set for the distance parameter δ (S122). Subsequently, all two character elements (C _i , C _j And However, i ≠ j) is extracted once (S123). Here, a connected portion of black pixels may be used as a character element, or a run of black pixels may be used as a character element.
[0046]
Next, it is determined whether or not the two character elements extracted in step S123 are close to each other (S124). Here, Rδ (C _i , C _j ) Is determined to determine whether two character elements are close to each other.
[0047]
Here, Rδ (C _i , C _j ) Means "two character elements C _i And C _j Is less than (or less than) the distance parameter δ ”. Regarding the two character elements c and d, "a sequence of certain character elements {x _i ｝ (I = 0, 1,... N−1) exists, and Rδ (c, x ₀ ), Rδ (x _i , X _{i + 1} ) (0 ≦ i ≦ n−2), Rδ (x _n-1 , D) are satisfied (hereinafter referred to as Sδ (c, d)) is a kind of relation mathematically called an equivalence relation. The equivalence relation has a remarkable property that the entire set in which it is defined is decomposed into several groups consisting only of those relations that hold.
[0048]
Therefore, as is performed in step S123, if it is checked whether or not the above-described equivalence relation is established for all combinations of two character elements in the image, all character elements in the image are Can be decomposed into groups such as Sδ (a, b). Note that the distance parameter δ is a vector quantity (δ _x , Δ _y ), Δ _x , Δ _y Are varied separately and the same is done. In this case, "the distance between the two character elements a and b is equal to or less than (or less than) the distance parameter δ" means that the distance in the x direction between the two character elements is δ. _x Less (or less), the distance in the y direction is δ _y The following (or less) "(the same applies hereinafter).
[0049]
Then, Rδ (C _i , C _j ) Holds (YES in S124), a character element label corresponding to each of the two character elements on a one-to-one basis, and all character element labels having the same value as one of the two character element labels, A common new value is substituted (S125). Then, it is determined whether or not all combinations of i and j have been processed (S126). Rδ (C _i , C _j If) is not satisfied (NO in S124), it is determined whether or not all combinations of i and j have been processed without executing the processing of step S125 (S126).
[0050]
If all combinations of i and j have not been processed (NO in S126), the process returns to step S123, and if all combinations of i and j have been processed (YES in S126), division is performed. The result is stored (S127).
[0051]
Next, it is determined whether or not all the distance parameters δ and a set 値 of values that can take the distance parameter δ are processed (S128). If the process has been performed (YES in S128), the process of creating a region division candidate ends, and the process returns to the main routine shown in FIG. If the processing has not been performed for all the distance parameters δ and the set 値 of values that can take the distance parameters δ (NO in S128), the distance parameters δ are changed (S129), and the processing from step S123 is performed again.
[0052]
Thus, the process of creating the area division candidates is completed, and the process returns to the main routine shown in FIG.
[0053]
In step S127, the division result is stored in the area / line information list 18 shown in FIG. In this case, if m candidates are created as a result of the region division (that is, if the region division is performed with m distance parameters δ), the region label 1811 of the first region / row information includes , The area label of the corresponding character in the first area division candidate is stored. Thereafter, the area labels of the second area / line information 182 to the m-th area / line information 185 include the second area division candidate to the m-th area division candidate. Are sequentially stored.
[0054]
Next, the line extraction processing and the character string creation processing performed by the character string creation unit 1013 in step S13 described above will be described using specific examples. In step S13, specifically, a row direction determination process is performed for all the region division candidates for each region in the candidate, and a row extraction process is performed along the determination direction.
[0055]
The above-described line direction determination processing is, for example, a joint invention between the inventor of the present application and other inventors, and is performed by the character string direction detecting device disclosed in Japanese Patent No. 3124854. Can be applied. That is, a criterion for determining whether the character string in the image is in the x direction or the y direction is determined in advance, and a method of determining whether the character string in the x direction or the y direction is based on the position coordinates of the input area rectangle is applied. be able to. For example, for each region rectangle, Rxy = (length in the y direction) / (length in the x direction) is calculated in advance, and RA = (the number of region rectangles where Rxy is less than 1.0) / (in the image) When (the total number of areas) is less than 0.5, the character string can be considered to be in the y direction on the page, and otherwise, the character string can be in the x direction.
[0056]
Further, for the above-described row extraction processing, for example, it is possible to apply the row extraction method disclosed in Japanese Patent Publication No. 8-16918, which is a joint invention between the present inventor and other inventors. it can. That is, it is possible to apply a line extraction method in which rectangles are connected to the left and right (in the case of horizontal writing) and lines are extracted according to the overlapping relationship between the rectangles.
[0057]
As the line information connected in the first area in the above-described line extraction processing, the character information buffer 1021 storing the previous character in the previous character information address 1812 shown in FIG. 4 for each character. The start address of the corresponding character information in is stored. Also, the next character information address 1813 stores the head address of the corresponding character information in the character information buffer 1021 in which the next character is stored. This is the same for the second region and the other regions. Here, the previous character refers to the character left one in the line in the case of horizontal writing, the character up one in the line in the case of vertical writing, and the next character refers to the character one in the line in the case of horizontal writing. The character immediately to the right, or in the case of vertical writing, the character immediately below in the line. If the character is at the beginning of the line, the previous character information address 1812 stores the start address of the last character of the previous line. If the character is at the end of the line, the next character information address 1813 stores the start address of the first character of the next line. The address is stored. However, the previous character information address 1812 of the first character of the first line in the area is NULL (0), and the next character information address 1813 of the last character of the last line in the area is also NULL (0).
[0058]
Next, FIG. 7 shows a specific example of the connection relationship based on the result of the above-described row extraction processing. FIG. 7 is a diagram illustrating a specific example of the result of the line extraction process in a horizontally written document. In FIG. 7, a rectangle indicates a character rectangle, and a right arrow (→) indicates a reference destination of the next character information address. The next character is the character indicated by the next (right next) rectangle in the line, and the first character at the end of the line. A left arrow (←) indicates a reference destination of the preceding character information address. The previous character is a character indicated by the previous (left adjacent) rectangle in the line, and the last character of the previous line at the beginning of the line. That is, in the case of the first character 41 in the area, NULL (0) is stored in the previous character information address. In the case of the last character 42 in the area, NULL (0) is stored in the next character information address.
[0059]
Next, a character is scanned for the line thus extracted for each region in the candidate to create a sentence. FIG. 8 is a flowchart showing the character string creation processing executed by the character string creation unit 1013 in step S13 described above. Referring to FIG. 8, first, a region division candidate counter J is initialized to 0 (S131). This area division candidate counter J uses a register in the CPU.
[0060]
Next, it is determined whether or not the area division candidate counter J is less than the number of area division candidates (S132). If the number is less than the number of area division candidates (YES in S132), the character counter I is subsequently initialized to 0 (S133). This character counter I also uses registers in the CPU.
[0061]
Next, it is determined whether or not the character counter I is less than the number of character information (S134). If the number is less than the number of character information (YES in S134), subsequently, the area / line information of the character information indicated by the character counter I indicated by the area division candidate counter J (that is, stored in the character information buffer 1021). It is determined whether the previous character information address of the J-th area / line information of the I-th character information is NULL (0). Thus, it is determined whether or not the corresponding character information is the first character (S135).
[0062]
If the character information is the first character (YES in S135), the address of the character information found in S135 (that is, the I-th character information buffer) is entered in the character information pointer (S136). This character information pointer uses a register in the CPU. Then, the character code referred to by the character information pointer is stored in the character string buffer 1022 (S137).
[0063]
Next, it is determined whether or not the next character information address referred to by the character information pointer is NULL (0). Accordingly, it is determined whether or not there is a next character (that is, a character connected by line extraction) (S138).
[0064]
If there is a next character (that is, if the next character information address is not NULL) (YES in S138), the next character information address is set in the character information pointer (S139), and the process returns to step S137.
[0065]
When there is no next character (that is, when the next character information address is NULL) (NO in S138), the character counter I is incremented (S140), and the process returns to step S134. If the character information is not the first character in step S135 (NO in S135), the character counter I is similarly incremented (S140), and the process returns to step S134. That is, the processing of steps S134 to S139 is repeated for the next row.
[0066]
When scanning of all character information in the area is completed, that is, when the character counter I is equal to or more than the number of character information (NO in S134), the area division candidate counter J is incremented (S141), and step S132 is performed. Return to That is, the processing of steps S132 to S140 is repeated for the next area.
[0067]
When the processing has been completed for all the area division candidates, that is, when the area division candidate counter J has become equal to or greater than the number of area division candidates (NO in S132), the present processing is ended and shown in FIG. Return to the main routine.
[0068]
FIG. 9 shows a specific example of the character string buffer 1022. Referring to FIG. 9, character string buffer 1022 includes first area division candidate character string buffers 51 to m-th area division candidate character string buffers 55 created for the number of area division candidates (in FIG. 9, the number of area division candidates is m is shown).
[0069]
One area division candidate character string buffer further includes the number of character strings and character string buffers for the number of character strings. That is, the first area division candidate character string buffer 51 includes the number 511 of character strings included in the first area division candidate, and the first character string buffer 512 to the L-th character string buffer 515 (in FIG. The case where the number of character strings included in one area division candidate is L is illustrated). The number L of character strings corresponds to the number determined as the first character (that is, the character whose previous character information address of the character information is NULL (0)) in step S135 in FIG. Reserved dynamically.
[0070]
Next, the calculation of the transition probabilities of all the character strings performed in step S104 of FIG. 3 will be specifically described.
[0071]
The transition probability value P of the character string is to calculate a probability value of a sentence “Today is fine weather”, considering a specific example. That is, it is as shown in the following equation.
[0072]
P (sunny weather today) = 0.73
According to the document “Speech Recognition by Probability Model” (written by Seiichi Nakagawa, Institute of Electronics, Information and Communication Engineers, Corona, first edition, 1988), the transition probability value P is expressed as C = (c1, c2,. ., Cn) and F () are given by the following equation (1) when a combination of characters in parentheses is used.
[0073]
(Equation 1)

[0074]
From equation (1), the transition probability value P (C) for the character string C having an arbitrary length can be obtained by preparing an appearance frequency (or appearance probability) table of a two-character set (digram) and a three-character set (trigram). Obtainable. However, when the target character code is a multi-byte code (for example, in Japanese), the number of character categories is on the order of thousands, and it is particularly difficult to secure a memory amount for securing a trigger. Therefore, it is not appropriate to directly employ Equation (1). Therefore, in the present embodiment, the transition probability value P of the character string is calculated by the following expression (2), which is an approximation expression using the occurrence frequency (unigram) of digram and one character.
[0075]
(Equation 2)

[0076]
For each of the plurality of character string buffers 512 to 515 indicated by the number of character strings 511 existing therein, the transition probability value P is individually calculated by the equation (2) for each of the region division candidate character string buffers illustrated in FIG. Then, the character string transition probability of the entire character string in the corresponding area candidate is calculated by the product thereof. Specifically, assuming that the number of character strings in the region division candidate is L and each character string is C1, C2,... CL, the character string transition probability of the entire region division candidate character string is calculated by the following equation (3). calculate.
[0077]
[Equation 3]

[0078]
In step S14, the character string probability calculation unit 1014 calculates the character string transition probabilities for all the region division candidates according to equation (3). In step S15, the optimal region determination unit 1015 determines that the character string transition probabilities are the maximum. Is adopted as the optimal area division result, and in step S16, the correct answer document creation unit 1016 creates a correct character string. The method of creating the correct character string in step S16 is almost the same as the character string creation processing shown in FIG. That is, if steps S131, S132, and S141 in FIG. 8 are eliminated and the area division candidate counter J is set to the area division candidate number determined to be optimum, a correct character string is created by the same processing as the processing shown in FIG. can do.
[0079]
By executing the above processing on the PC 1 which is the document processing device in the present embodiment, highly accurate conversion from a static document format to a dynamic document format can be realized.
[0080]
[Second embodiment]
Next, FIG. 10 is a diagram showing a specific example of region division candidates for a document similar to the input image 301 shown in FIG. 5 and line extraction for each region division in the first embodiment.
[0081]
Referring to FIG. 10, region division candidates 61 to 63 indicate region division candidates obtained when distance parameter δ used for region division is changed in the first embodiment. Further, the divided regions 611 to 613 indicate the divided regions in the region dividing candidate 61, the divided regions 621 and 622 indicate the divided regions in the region dividing candidate 62, and the divided region 631 indicates the divided regions in the region dividing candidate 63. The bold horizontal rectangle in each divided area indicates a line, and the thin rectangle therein indicates a character. As shown in FIG. 5, the region division candidates 61 to 63 are region division candidates for a horizontally written three-column document. Therefore, the region division is performed by the region division candidate 61 composed of three regions of the divided regions 611 to 613. Is the correct answer.
[0082]
In the first embodiment, a line is extracted for each region in each region division candidate, a character string is created by connecting characters, and an optimal division candidate is determined based on a character string transition probability. Again, a character string was extracted for the optimal division candidate (the region division candidate 61 in FIG. 10), and a correct sentence was created. The reason why the processing according to the first embodiment operates effectively is that the character string extracted from the incorrectly divided area division candidate is connected to characters in another area, so that the transition probability as a character string decreases. Because you do. For example, in FIG. 10, characters in different regions are connected as shown in the divided regions 621 and 631, and as a result, a character string in a portion extending between different regions becomes a character string whose language is meaningless. . Therefore, the transition probability as a character string decreases.
[0083]
Therefore, in the second embodiment, a description will be given of a case in which this fact is further utilized and a higher-precision conversion is performed in the PC 1 similar to the document processing apparatus in the first embodiment.
[0084]
Specifically, in the line extraction results of all the area division candidates 61 to 63 shown in FIG. 10, it is more effective to evaluate only the character chain of a part different from the other character concatenation to create a correct sentence more effectively. This is clearer than the above equation (2). That is, FIG. 11 shows a darkened portion of a character different from the character connection when comparing the area division candidates 61 to 63 shown in FIG. As shown in FIG. 11, characters that are boundaries between regions are characters having different concatenations. Therefore, a character string composed of several characters before and after that character is created for each region candidate, the transition probability value P of the character string is obtained by Expression (2), and these are evaluated by Expression (3). It will be good.
[0085]
The conversion process performed by PC 1 in the second embodiment is substantially the same as the conversion process shown in FIG. 3 in the first embodiment, and therefore, the description of the same portions will not be repeated. In the second embodiment, the character string creation processing executed by the character string creation unit 1013 in step S13 in FIG. 3 is different from the processing in the first embodiment. Therefore, a flowchart of the character string creation processing according to the second embodiment is shown in FIG. The character string creation processing shown in the flowchart in FIG. 12 is almost the same as the character string creation processing in the first embodiment shown in the flowchart in FIG. That is, referring to FIG. 12, the processing in steps S231 to S234 is the same as the processing in steps S131 to S134 in FIG. Thus, hereinafter, the processing after step S235 will be described.
[0086]
With reference to FIG. 12, it is determined whether or not the connection of the character indicated by the character counter I is the same in all the region candidates, thereby determining a character having a different connection (S235). Specifically, when the previous character information address and the next character information address of all the areas and line information of the area / line list of the corresponding character information are not the same (when at least one different address is assigned) Then, the corresponding character is determined to be a character having a different concatenation.
[0087]
If even one connection is different (YES in S235), all character codes before that character and up to N characters which is a predetermined number of characters are extracted (S236). This can be done by scanning the character information address preceding the character information. Similarly, by scanning the next character information address of the character information, all the next character codes from that character to the N character are also extracted (S237).
[0088]
Next, a character string is created by arranging the character codes extracted in steps S236 and S237 in the order of their concatenation, and stored in the character string buffer 1021 (S238). Thereafter, similarly to the processing after step S140 in FIG. 8, the character counter I is incremented (S239), and the process returns to step S234. Also, if the connection of the character indicated by the character counter I in step S235 is the same for all area candidates (NO in S235), the character counter I is similarly incremented (S239), and the process returns to step S234. That is, the processing of steps S234 to S238 is repeated for the next row.
[0089]
As described above, in the present embodiment, by performing the above-described character string creation processing that evaluates only the character string composed of characters in the vicinity of the character having different concatenation, the conversion processing in the first embodiment is improved. This also makes it possible to realize a more accurate area determination.
[0090]
Further, the conversion method performed by the above-described document conversion apparatus can be provided as a program. Such a program can be recorded on a computer-readable recording medium such as a flexible disk, a CD-ROM, a ROM, a RAM, and a memory card attached to the computer, and can be provided as a program product. Alternatively, the program can be provided by being recorded on a recording medium such as a hard disk incorporated in the computer. Further, the program can be provided by downloading via a network.
[0091]
The provided program product is installed and executed in a program storage unit such as a hard disk. Note that the program product includes the program itself and a recording medium on which the program is recorded.
[0092]
The embodiments disclosed this time are to be considered in all respects as illustrative and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the description above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a specific example of a configuration of a PC 1 that is a document processing device according to the present embodiment.
FIG. 2 is a block diagram showing a configuration of a PC 1 for performing a conversion process.
FIG. 3 is a flowchart illustrating a conversion process performed by the PC 1 according to the present embodiment.
FIG. 4 is a diagram illustrating a specific example of a character information buffer 1021;
FIG. 5 is a diagram showing a result display image 32 with respect to an input image 31 including regions 311 to 313.
FIG. 6 is a flowchart illustrating a process of creating a region division candidate executed in step S12.
FIG. 7 is a diagram illustrating a specific example of a connection relationship.
FIG. 8 is a flowchart showing a character string creation process executed in step S13.
FIG. 9 is a diagram showing a specific example of a character string buffer 1022.
FIG. 10 is a diagram showing a specific example of region division candidates for a document similar to the input image 301 shown in FIG. 5 and line extraction for each region division in the first embodiment.
FIG. 11 is a diagram in which characters in portions different from character concatenation when the region division candidates 61 to 63 shown in FIG. 10 are compared are darkened.
FIG. 12 is a flowchart illustrating a character string creation process according to the second embodiment.
[Explanation of symbols]
1 PC, 10 static document data, 11 character code, 12 start x coordinate, 13 start y coordinate, 14 x size, 15 y size, 16 font, 17 attribute information, 18 area / line information list, 20 dynamic document data , 31 input image, 32 result display image, 41 first character, 42 last character, 51-55 area division candidate character string buffer, 61-63 area division candidate, 101 control unit, 102 storage unit, 103 input unit, 104 output unit 181 to 185 area / line information, 311 to 313, 321, 322 area, 511 character string number, 512 to 515 character string buffer, 611 to 613, 621, 622, 631 divided area, 1011 character information extraction unit, 1012 area Division candidate creation unit, 1013 character string creation unit, 1014 character string probability calculation unit, 1015 optimal area determination Fixed part, 1016 correct answer document creation part, 1021 character information buffer, 1022 character string buffer, 1811 area label, 1812 previous character information address, 1813 next character information address.

Claims (6)

A document conversion method for converting document data including character position information into document data not including character position information,
Character information extraction step of extracting character information including the position information of each character from document data including the position information of the character;
An area division candidate creating step of creating a division candidate for dividing the document data into a plurality of areas based on the extracted position information of the character;
For each of the created region division candidates, extract a line based on the position information of the extracted each character, a character string creating step of creating a character string by searching the extracted line,
A string probability calculation step of calculating the transition probability value of the created string,
An area determination step of determining an optimal area from the created area division candidates based on the calculation result,
A text creating step of creating a text sentence that does not include the position information of each character based on the result of the determined optimal area. A document conversion method for converting document data including character position information into document data not including character position information,
Character information extraction step of extracting character information including the position information of each character from document data including the position information of the character;
An area division candidate creating step of creating a division candidate for dividing the document data into a plurality of areas based on the extracted position information of the character;
For each of the created region division candidates, a line is extracted based on the position information of each of the extracted characters, and the connection of the character as the extraction result is one in all of the created region division candidates. A character string creating step of creating a partial character string including the character and a predetermined number of characters before and after the character, centering on different characters,
A string probability calculation step of calculating a transition probability value of the created partial string,
An area determination step of determining an optimal area from the created area division candidates based on the calculation result,
A text creating step of creating a text sentence that does not include the position information of each character based on the result of the determined optimal area. The character information extracting step is to interpret data of a common format converted based on document data including the position information of the character, and to directly interpret the document data including the position information of the character, 3. The document conversion method according to claim 1, wherein the character information is extracted by at least one method. A document conversion device for converting document data including character position information into document data not including character position information,
Character information extraction means for extracting character information including the position information of each character from document data including the position information of the character,
Area division candidate creating means for creating a division candidate for dividing the document data into a plurality of areas based on the extracted position information of the character;
For each of the created region division candidates, extract a line based on the position information of the extracted each character, a character string creating unit that creates a character string by searching for the extracted line,
String probability calculation means for calculating the transition probability value of the created character string,
Area determination means for determining an optimal area from the created area division candidates based on the calculation result,
A document conversion device, comprising: a text creating unit that creates a text sentence that does not include the position information of each character based on the result of the determined optimal area. A document conversion device for converting document data including character position information into document data not including character position information,
Character information extraction means for extracting character information including the position information of each character from document data including the position information of the character,
Area division candidate creating means for creating a division candidate for dividing the document data into a plurality of areas based on the extracted position information of the character;
For each of the created region division candidates, a line is extracted based on the position information of each of the extracted characters, and the connection of the character as the extraction result is one in all of the created region division candidates. A character string creating means for creating a partial character string consisting of the character and a predetermined number of characters before and after the character, centering on different characters,
String probability calculating means for calculating the transition probability value of the created partial character string,
Area determination means for determining an optimal area from the created area division candidates based on the calculation result,
A document conversion device, comprising: a text creating unit that creates a text sentence that does not include the position information of each character based on the result of the determined optimal area. The character information extracting means interprets data in a common format converted based on document data including the position information of the character, and directly interprets the document data including the position information of the character, 6. The document conversion device according to claim 4, wherein the character information is extracted by at least one method.

Images