JP5677139B2 - Form character recognition device - Google Patents

Form character recognition device Download PDF

Info

Publication number
JP5677139B2
JP5677139B2 JP2011048983A JP2011048983A JP5677139B2 JP 5677139 B2 JP5677139 B2 JP 5677139B2 JP 2011048983 A JP2011048983 A JP 2011048983A JP 2011048983 A JP2011048983 A JP 2011048983A JP 5677139 B2 JP5677139 B2 JP 5677139B2
Authority
JP
Japan
Prior art keywords
character
image
frame
recognition
contact
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Fee Related
Application number
JP2011048983A
Other languages
Japanese (ja)
Other versions
JP2012185713A (en
Inventor
敬 平野
敬 平野
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Mitsubishi Electric Corp
Original Assignee
Mitsubishi Electric Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Mitsubishi Electric Corp filed Critical Mitsubishi Electric Corp
Priority to JP2011048983A priority Critical patent/JP5677139B2/en
Publication of JP2012185713A publication Critical patent/JP2012185713A/en
Application granted granted Critical
Publication of JP5677139B2 publication Critical patent/JP5677139B2/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Description

この発明は、帳票画像上の文字を認識処理する帳票文字認識装置に関し、特に、文字枠内に乱雑に記載された文字や取り消し線あるいはノイズに起因する誤認識を抑制する構成に関するものである。   The present invention relates to a form character recognition apparatus for recognizing characters on a form image, and more particularly to a configuration that suppresses misrecognition caused by characters, strikethrough, or noise that are randomly described in a character frame.

帳票画像上の文字を認識処理する帳票文字認識装置では、誤認識を抑制するために、不明確な認識結果を棄却する棄却判定処理を有している。この棄却判定処理では、一般的に、文字パターンを辞書と照合して得た認識結果の信頼度（スコア）が、所定の閾値を超えた場合に棄却する処理が用いられる。このようなスコアによる棄却判定により、辞書に登録された字形と著しく形状が異なる文字パターンを棄却できる。   A form character recognition apparatus that recognizes characters on a form image has a rejection determination process for rejecting an unclear recognition result in order to suppress erroneous recognition. In this rejection determination process, generally, a process of rejecting when the reliability (score) of a recognition result obtained by comparing a character pattern with a dictionary exceeds a predetermined threshold is used. With the rejection determination based on such a score, it is possible to reject a character pattern whose shape is significantly different from the character shape registered in the dictionary.

しかし、現実的には、スコアによる棄却判定処理単独では誤認識を全て無くすことは困難である。特に、文字が文字枠線と接触した場合や、取り消し線が描かれた場合、ノイズが重畳された場合において、スコアのみでは棄却できずに誤認識する場合がある。そのため、スコア以外の判定基準に基づく棄却判定処理が望まれる。このような棄却判定手段として、幾つかの従来技術があった。   However, in reality, it is difficult to eliminate all misrecognitions only by the rejection determination process based on the score. In particular, when a character comes into contact with a character frame line, when a strikethrough is drawn, or when noise is superimposed, there is a case where the score alone cannot be rejected and erroneously recognized. Therefore, a rejection determination process based on determination criteria other than the score is desired. There are several conventional techniques as such rejection determination means.

例えば、特許文献１に示すように、文字画像上に文字とは異なる線分があるかを調べることで取り消し線の有無を検知し、取り消し線が有ると判定された場合は文字認識結果を棄却する文字認識装置があった。
また、非特許文献１に示すように、帳票画像上のノイズを除去してから文字認識処理を行う方式があった。例えばファクシミリで送信された帳票画像では縦方向のノイズが重畳する場合がある。そのため帳票画像を上下方向に走査して、帳票画像長さに対して一定割合以上の黒線が検出された場合、これをノイズとみなす。検知したノイズを画像中から除去した後、その画像に対して文字認識処理を行うことにより、このような縦方向のノイズを数字の“１”に誤認識する問題を抑制できる。
更に、特許文献２に示すように、文字パターンが文字枠線と接触・交差する場合に、通常の認識辞書との照合を行うと共に、文字枠線と接触・交差した場合用の特殊な辞書を用いて照合処理を行うようにしたものがあった。これにより、文字パターンが文字枠線と接触・交差した場合の認識精度を高めている。また、両方の照合結果が異なる場合は強制的に棄却することで誤認識を抑制している。 For example, as shown in Patent Document 1, the presence or absence of a strikethrough is detected by checking whether there is a line segment different from the character on the character image. If it is determined that there is a strikethrough, the character recognition result is rejected. There was a character recognition device.
Further, as shown in Non-Patent Document 1, there has been a method of performing character recognition processing after removing noise on a form image. For example, in a form image transmitted by facsimile, vertical noise may be superimposed. Therefore, when a form image is scanned in the vertical direction and a black line of a certain ratio or more with respect to the form image length is detected, this is regarded as noise. By removing the detected noise from the image and then performing character recognition processing on the image, it is possible to suppress the problem of erroneously recognizing such vertical noise as the number “1”.
Furthermore, as shown in Patent Document 2, when a character pattern touches / intersects a character frame line, a normal dictionary is collated and a special dictionary for contact / intersect with a character frame line is created. Some of them were used for collation processing. This improves the recognition accuracy when the character pattern contacts / intersects the character frame line. Moreover, when both collation results are different, erroneous recognition is suppressed by forcibly rejecting them.

特許第３３９１２２３号公報Japanese Patent No. 3391223 特許第３７９４２８５号公報Japanese Patent No. 3794285

平野敬，岡田康裕，依田文夫，“表の構造解析によるＦＡＸ送信帳票からの文字認識フィールド抽出法”，電子情報通信学会技術報告，ＰＲＭＵ ２０００−７１ （２０００）Takashi Hirano, Yasuhiro Okada, Fumio Yoda, “Character recognition field extraction method from FAX transmission form by table structure analysis”, IEICE Technical Report, PRMU 2000-71 (2000)

しかしながら、例えば特許文献１に記載されているような装置では、文字ではない直線成分を探索することで取り消し線を検知する処理を行い、取り消し線が見つかった場合に文字認識結果を棄却するものであるが、そのため、直線以外の例えば曲線で描かれた取り消しには対処できないという課題があった。
また、非特許文献１に記載されているようなフィールド抽出法では、ノイズが検出されれば誤認識は抑制可能となるが、検出が困難なノイズには対処できないという課題があった。例えば、非特許文献１において、帳票画像全体を走査して縦方向のノイズを検出する処理があるが、文字記入欄付近にのみ短い縦方向のノイズが存在した場合は、方式上これをノイズとして検出できない。そのために残ったノイズを文字“１”と誤認識するといった問題点を有していた。 However, for example, an apparatus as described in Patent Document 1 performs processing for detecting a strikethrough by searching for a straight line component that is not a character, and rejects a character recognition result when a strikethrough is found. However, for this reason, there is a problem that it is not possible to deal with cancellations drawn with a curve other than a straight line, for example.
In addition, in the field extraction method described in Non-Patent Document 1, erroneous recognition can be suppressed if noise is detected, but there is a problem that noise that is difficult to detect cannot be dealt with. For example, in Non-Patent Document 1, there is a process of detecting the vertical noise by scanning the entire form image. If there is a short vertical noise only in the vicinity of the character entry field, this is regarded as noise in the method. It cannot be detected. Therefore, there is a problem that the remaining noise is erroneously recognized as the character “1”.

更に、特許文献２に記載されているような装置では、通常の認識辞書との照合と共に、文字枠線と接触・交差した場合用の特殊な辞書を用いた照合を行い、双方の認識結果が合致しない場合は認識結果を棄却している。しかしながら、両方の認識結果が誤るケースも存在し、この場合は誤認識を抑制できないという課題があった。
そして、これら文献に記載されている処理は別々の要因による誤認識を抑制するものである。そのため全処理を実装しなければ、ここに記載された要因（取り消し線、ノイズ、文字枠線との接触・交差）に起因した誤認識は抑制できない。しかし、これを実現するには処理コストが増大するという課題があった。 Furthermore, in an apparatus as described in Patent Document 2, collation with a normal recognition dictionary and collation using a special dictionary for contact / intersection with a character frame line are performed, and both recognition results are obtained. If they do not match, the recognition result is rejected. However, there are cases where both recognition results are erroneous, and in this case, there is a problem that the erroneous recognition cannot be suppressed.
The processes described in these documents suppress erroneous recognition due to different factors. For this reason, if all the processes are not implemented, it is not possible to suppress misrecognition caused by the factors described here (strikethrough, noise, contact / intersection with character frame lines). However, in order to realize this, there is a problem that the processing cost increases.

この発明は上記のような課題を解決するためになされたもので、処理コストが小さく、かつ、誤認識を抑制することのできる帳票文字認識装置を得ることを目的とする。   The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a form character recognition device that is low in processing cost and can suppress erroneous recognition.

この発明に係る帳票文字認識装置は、帳票画像上の文字を認識処理する帳票文字認識装置において、帳票画像中から文字認識対象とする文字記入欄位置を抽出するフィールド抽出部と、文字記入欄内の画像から文字枠線を検出する枠線検出部と、枠線検出部の検出結果に基づいて、文字記入欄内の画像から１文字分の文字パターンを取り出して辞書との照合を行う文字パターン認識部と、文字パターン認識部で得た認識結果の信頼度の値を元に認識結果の棄却判定を行うスコアによる棄却判定部と、枠線検出部の検出結果に基づいて、文字記入欄内の画像から１文字分の非文字枠画像を取り出して、非文字枠画像に接触している文字枠線の数を計測する文字・枠線接触数検出部と、非文字枠画像に接触している文字枠線数が所定の閾値を越える場合に、文字枠線に対応した文字の認識結果を棄却する罫線接触棄却判定部と、スコアによる棄却判定部の結果と罫線接触棄却判定部の結果のうち、どちらかの結果が棄却であった場合は、最終的に出力する文字認識結果を棄却する棄却処理統合部を備えたものである。 The form character recognition apparatus according to the present invention is a form character recognition apparatus for recognizing a character on a form image. A field extraction unit for extracting a character entry field position to be a character recognition object from the form image; A character line pattern for detecting a character frame line from the image and a character pattern for extracting a character pattern for one character from the image in the character entry field and collating with a dictionary based on the detection result of the frame line detection unit Based on the detection result of the recognition unit, the score determination unit for determining rejection of the recognition result based on the reliability value of the recognition result obtained by the character pattern recognition unit, and the detection result of the frame line detection unit, A non-character frame image for one character is taken out from the image and a character / frame line contact number detection unit for measuring the number of character frame lines in contact with the non-character frame image; The number of character frame lines exceeds the specified threshold When a ruled line contact rejection determination unit to reject the recognition result of the character corresponding to the character frame line, among the result of the rejection determination unit according to the score and the ruled line contact rejection determination unit results, either result is rejected In this case , a rejection processing integration unit for rejecting the character recognition result to be finally output is provided.

この発明の帳票文字認識装置は、非文字枠画像に接触した文字枠線の本数が閾値以上の場合は文字認識結果を棄却するようにしたので、処理コストが小さく、かつ、誤認識を抑制することができる。   The form character recognition apparatus according to the present invention rejects the character recognition result when the number of character frame lines in contact with the non-character frame image is equal to or greater than the threshold value, so that the processing cost is low and erroneous recognition is suppressed. be able to.

この発明の実施の形態１による帳票文字認識装置を示す構成図である。It is a block diagram which shows the form character recognition apparatus by Embodiment 1 of this invention. この発明の実施の形態１による帳票文字認識装置のフィールド抽出部の処理例を示す説明図である。It is explanatory drawing which shows the process example of the field extraction part of the form character recognition apparatus by Embodiment 1 of this invention. この発明の実施の形態１による帳票文字認識装置の枠線検出部の処理例を示す説明図である。It is explanatory drawing which shows the process example of the frame line detection part of the form character recognition apparatus by Embodiment 1 of this invention. この発明の実施の形態１による帳票文字認識装置の文字パターン認識部の動作を示す説明図である。It is explanatory drawing which shows operation | movement of the character pattern recognition part of the form character recognition apparatus by Embodiment 1 of this invention. この発明の実施の形態１による帳票文字認識装置の文字・枠線接触数検出部の処理例を示す説明図である。It is explanatory drawing which shows the process example of the character and frame line contact number detection part of the form character recognition apparatus by Embodiment 1 of this invention. この発明の実施の形態１による帳票文字認識装置の文字・枠線接触数検出部の接触箇所検出例を示す説明図である。It is explanatory drawing which shows the contact location detection example of the character and frame line contact number detection part of the form character recognition apparatus by Embodiment 1 of this invention. この発明の実施の形態１による帳票文字認識装置のスコアによる棄却判定部の判定処理例を示す説明図である。It is explanatory drawing which shows the example of a determination process of the rejection determination part by the score of the form character recognition apparatus by Embodiment 1 of this invention. この発明の実施の形態１による帳票文字認識装置の罫線接触棄却判定部の判定例を示す説明図である。It is explanatory drawing which shows the example of a determination of the ruled line contact rejection determination part of the form character recognition apparatus by Embodiment 1 of this invention.

実施の形態１．
図１は、この発明の実施の形態１による帳票文字認識装置を示す構成図である。
図１において、帳票画像１０１は、紙の帳票をファクシミリやスキャナで画像化した画像データである。帳票文字認識装置１０２は、帳票画像１０１を解析して、その認識結果１０３を出力する文字認識装置であり、フィールド抽出部１０４、枠線検出部１０５、文字パターン認識部１０６、スコアによる棄却判定部１０７、棄却処理統合部１０８、結果出力部１０９、文字・枠線接触数検出部１１０、罫線接触棄却判定部１１１を備えている。 Embodiment 1 FIG.
FIG. 1 is a block diagram showing a form character recognition apparatus according to Embodiment 1 of the present invention.
In FIG. 1, a form image 101 is image data obtained by imaging a paper form with a facsimile or a scanner. The form character recognition device 102 is a character recognition device that analyzes the form image 101 and outputs the recognition result 103, and includes a field extraction unit 104, a frame line detection unit 105, a character pattern recognition unit 106, and a rejection determination unit based on scores. 107, a rejection processing integration unit 108, a result output unit 109, a character / frame line contact number detection unit 110, and a ruled line contact rejection determination unit 111.

フィールド抽出部１０４は、帳票画像１０１中から文字記入欄の位置と大きさを抽出する。枠線検出部１０５は、文字記入欄の画像中から文字枠線を検出する。文字パターン認識部１０６は、文字枠線の検出結果を元に１文字の文字パターンを取得して辞書と照合することで文字認識結果を得る。スコアによる棄却判定部１０７は、文字認識結果に含まれる認識結果の信頼度（スコア）の値から棄却判定を行う。文字・枠線接触数検出部１１０は、非文字枠画像（文字やノイズ、取り消し線）と文字枠線との接触有無を調べて、接触している文字枠線本数を計測する。罫線接触棄却判定部１１１は、接触している文字枠線の本数を元に文字認識結果を棄却するかを判定する。棄却処理統合部１０８は、スコアによる棄却判定部１０７の結果と罫線接触棄却判定部１１１の結果を元に、文字認識結果を棄却するかを最終決定する。結果出力部１０９は文字認識結果を外部へ出力する。   The field extraction unit 104 extracts the position and size of the character entry field from the form image 101. The frame line detection unit 105 detects a character frame line from the image in the character entry field. The character pattern recognition unit 106 obtains a character recognition result by acquiring a character pattern of one character based on the detection result of the character frame line and collating it with a dictionary. The score rejection determination unit 107 performs rejection determination from the reliability (score) value of the recognition result included in the character recognition result. The character / frame line contact number detection unit 110 checks the presence / absence of contact between a non-character frame image (characters, noise, strikethrough) and a character frame line, and measures the number of character frame lines in contact. The ruled line contact rejection determination unit 111 determines whether to reject the character recognition result based on the number of touching character frame lines. Rejection processing integration unit 108 finally determines whether to reject the character recognition result based on the result of rejection determination unit 107 based on the score and the result of ruled line contact rejection determination unit 111. The result output unit 109 outputs the character recognition result to the outside.

図２は、フィールド抽出部１０４の処理例を示す説明図である。２０１は帳票画像の例である。２０２は該帳票画像から抽出した文字記入欄の画像例である。
図３は、枠線検出部１０５の処理例を示す説明図である。３０１は文字記入欄の画像例２０２に対して、文字枠線を検出した結果例である。
図４は、文字パターン認識部１０６の処理例を示す説明図である。４０１は文字枠線との接触部分を除去して作成した文字パターン、４０２は文字枠線との接触部分を文字の一部とみなして作成した文字パターンである。
図５は、文字・枠線接触数検出部１１０の処理例を示す説明図である。５０１は１個の文字枠を囲む画像バッファ、５０２は文字パターンが上側の文字枠線と接触していると判定された領域、５０３は文字パターンが左側の文字枠線と接触していると判定された領域、５０４は文字パターンが下側の文字枠線と接触していると判定された領域である。 FIG. 2 is an explanatory diagram illustrating a processing example of the field extraction unit 104. 201 is an example of a form image. Reference numeral 202 denotes an image example of a character entry column extracted from the form image.
FIG. 3 is an explanatory diagram illustrating a processing example of the frame line detection unit 105. Reference numeral 301 denotes an example of a result of detecting a character frame line for the image example 202 in the character entry field.
FIG. 4 is an explanatory diagram illustrating a processing example of the character pattern recognition unit 106. 401 is a character pattern created by removing the contact portion with the character frame line, and 402 is a character pattern created by regarding the contact portion with the character frame line as a part of the character.
FIG. 5 is an explanatory diagram illustrating a processing example of the character / frame line contact number detection unit 110. 501 is an image buffer surrounding one character frame, 502 is an area where it is determined that the character pattern is in contact with the upper character frame line, and 503 is determined that the character pattern is in contact with the left character frame line. The region 504 is a region where it is determined that the character pattern is in contact with the lower character frame line.

図６は、文字・枠線接触数検出部１１０で得た、非文字画像と接触する文字枠線の本数を示す説明図である。６０１は２本の文字枠線と接触した文字の例、６０２は３本の文字枠線と接触した文字の例である。
図７は、スコアによる棄却判定部１０７の処理例を示す説明図である。７０１は辞書に登録された文字“３”と形状が類似した文字パターン、７０２は辞書に登録された文字“３”と形状が大きく異なる文字パターンの例である。
図８は、罫線接触棄却判定部１１１の動作例を示す説明図である。８０１は直線の取り消し線が記載された文字記入欄、８０２は直線でない取り消し線が記載された文字記入欄、８０３は短い縦方向のノイズが重畳された文字記入欄、８０４は乱雑な文字が記載された文字記入欄の例である。 FIG. 6 is an explanatory diagram showing the number of character frame lines in contact with the non-character image obtained by the character / frame line contact number detection unit 110. Reference numeral 601 denotes an example of a character in contact with two character frame lines, and reference numeral 602 denotes an example of a character in contact with three character frame lines.
FIG. 7 is an explanatory diagram illustrating a processing example of the rejection determination unit 107 based on a score. 701 is an example of a character pattern having a shape similar to that of the character “3” registered in the dictionary, and 702 is an example of a character pattern having a shape greatly different from that of the character “3” registered in the dictionary.
FIG. 8 is an explanatory diagram illustrating an operation example of the ruled line contact rejection determination unit 111. 801 is a character entry field with a straight strikethrough, 802 is a character entry field with a non-straight strikethrough, 803 is a character entry field with a short vertical noise superimposed, and 804 is a messy character entry It is an example of the written character entry column.

以下、図１〜図８を適宜参照しつつ、実施の形態１の帳票文字認識装置の動作について説明する。
先ず、図１に示すフィールド抽出部１０４は、帳票画像１０１から文字記入欄の位置と大きさを検出する。この文字記入欄の抽出処理は既存の帳票ＯＣＲと同じ方式で実現できる。本処理により、例えば図２に示す帳票画像２０１から、２文字枠の文字記入欄２０２が抽出される。 Hereinafter, the operation of the form character recognition apparatus according to the first embodiment will be described with reference to FIGS.
First, the field extraction unit 104 shown in FIG. 1 detects the position and size of the character entry field from the form image 101. This extraction process of the character entry column can be realized by the same method as the existing form OCR. By this processing, for example, a character entry column 202 having a two-character frame is extracted from the form image 201 shown in FIG.

次に、枠線検出部１０５は、抽出された文字記入欄から文字枠線を検出する。これも既存の帳票ＯＣＲと同じ方式で実現できる。例えば図３に示した文字枠線抽出結果例３０１は、文字記入欄２０２に対して本処理を適用した結果である（図３は９０度回転して表示している）。図３中の文字“５”は縦の文字枠線を示しており、“＃”は横の文字枠線である。また“＠”は非文字枠画像（文字や取り消し線、ノイズ）を示し、“＊”は非文字枠画素と文字枠線との接触位置を示す。   Next, the frame line detection unit 105 detects a character frame line from the extracted character entry field. This can also be realized in the same manner as the existing form OCR. For example, a character frame line extraction result example 301 shown in FIG. 3 is a result of applying this processing to the character entry field 202 (FIG. 3 is displayed after being rotated 90 degrees). The character “5” in FIG. 3 indicates a vertical character frame line, and “#” is a horizontal character frame line. “@” Indicates a non-character frame image (character, strikethrough, noise), and “*” indicates a contact position between the non-character frame pixel and the character frame line.

次に、文字パターン認識部１０６は、枠線検出部１０５の結果に基づいて、１文字分の非文字枠画像を取り出す。これを文字パターンとして、辞書と照合することで、その文字パターンに対する文字認識結果を取得する。例えば、図３に示す文字枠線抽出結果例３０１からは図４の４０１に示す非文字枠画像が得られるため、これを辞書と照合することで“８”という文字認識結果を得る。更に、本処理では、文字パターンが文字枠線と接触・交差している場合、文字パターンと接触した文字枠線部分（図３における“＊”部分）も文字だとみなした文字パターン４０２を作成し、これを辞書と照合する。そして信頼度（スコア）が高い方の文字認識結果を出力する。   Next, the character pattern recognition unit 106 extracts a non-character frame image for one character based on the result of the frame line detection unit 105. This is used as a character pattern and collated with a dictionary to obtain a character recognition result for the character pattern. For example, since the non-character frame image 401 shown in 401 of FIG. 4 is obtained from the character frame line extraction result example 301 shown in FIG. 3, the character recognition result “8” is obtained by collating this with the dictionary. Furthermore, in this processing, when the character pattern is in contact with or intersecting the character frame line, a character pattern 402 is created in which the character frame line part (the “*” part in FIG. 3) in contact with the character pattern is also regarded as a character. This is checked against the dictionary. The character recognition result with the higher reliability (score) is output.

次に、スコアによる棄却判定部１０７は、文字パターン認識部１０６で得た文字認識結果の信頼度（スコア）を調べ、スコアの値が閾値を越える場合は、その文字認識結果を出力する。逆にスコアの値が閾値を超えない場合は、文字認識結果を棄却とする。本処理では文字パターン認識部１０６で得た信頼度（スコア）を指標として棄却判定を行うため、辞書に記述された文字形状と異なる形状の文字パターンが棄却できる。
例えば、辞書に活字の文字パターン“３”が登録されている場合、図７に示すように辞書に登録された文字パターンと形状が類似した文字パターン７０１が入力された場合は正しい認識結果を返し、形状が異なる文字パターン７０２が入力された場合は、認識結果が疑わしいと判断して棄却する。
ここまでは通常の帳票ＯＣＲと同様な処理である。 Next, the rejection determination unit 107 based on the score checks the reliability (score) of the character recognition result obtained by the character pattern recognition unit 106, and outputs the character recognition result if the score value exceeds the threshold value. Conversely, if the score value does not exceed the threshold value, the character recognition result is rejected. In this process, since the rejection determination is performed using the reliability (score) obtained by the character pattern recognition unit 106 as an index, a character pattern having a shape different from the character shape described in the dictionary can be rejected.
For example, if the type character pattern “3” is registered in the dictionary, and a character pattern 701 similar in shape to the character pattern registered in the dictionary is input as shown in FIG. 7, a correct recognition result is returned. When a character pattern 702 having a different shape is input, it is determined that the recognition result is suspicious and is rejected.
Up to this point, the processing is the same as that of normal form OCR.

次に、文字・枠線接触数検出部１１０は、１文字ずつ、文字枠領域付近の画像を調べて、非文字枠画像が、文字枠線と接触した箇所を特定して、その文字枠線の本数を計測する。具体的には図３に示した文字枠線の解析結果データから、図５に示すように１文字分の画像５０１を切り出す。そして内部を１画素ずつ走査して、上下に“＊＠”が並んだ画像領域５０２が見つかった場合、上側の横文字枠線と非文字枠画像との接触有と判断する。同様に、左右に“＊＠”が並んだ画像領域５０３が見つかった場合、左側の縦文字枠線との接触有と判断する。また、上下に“＠＊”が並んだ画像領域５０４が見つかった場合、下側の横文字枠線との接触有と判断する。このような処理を行うことで、文字やノイズ、取り消し線と接触した文字枠線の本数を測定する。このように非文字枠画像と接触した文字枠線の本数０〜４を利用して、文字の乱雑度を数値的に表現する。   Next, the character / frame line contact number detection unit 110 examines an image in the vicinity of the character frame area for each character, specifies a position where the non-character frame image is in contact with the character frame line, and determines the character frame line. Measure the number of Specifically, from the character frame line analysis result data shown in FIG. 3, an image 501 for one character is cut out as shown in FIG. When the inside is scanned pixel by pixel and an image area 502 in which “* @” is lined up and down is found, it is determined that the upper horizontal character frame line and the non-character frame image are in contact. Similarly, when an image region 503 in which “* @” is aligned on the left and right is found, it is determined that there is contact with the left vertical character frame line. Further, when an image area 504 in which “@ *” is lined up and down is found, it is determined that there is contact with the lower horizontal character frame line. By performing such processing, the number of character frame lines in contact with characters, noise, and strikethrough is measured. In this way, the character randomness is numerically expressed using the number 0 to 4 of the character frame lines in contact with the non-character frame image.

図６は本処理の結果例を示す。ここで図中の○印は文字枠と非文字枠画像との接触位置を示す。文字６０１は２本の文字枠線と接触していると判定され、文字６０２は３本の文字枠線と接触していると判定される。   FIG. 6 shows an example of the result of this processing. Here, a circle in the figure indicates a contact position between the character frame and the non-character frame image. It is determined that the character 601 is in contact with two character frame lines, and the character 602 is determined to be in contact with three character frame lines.

次に、罫線接触棄却判定部１１１は、非文字枠画像に接触した文字枠線の本数が閾値以上の場合、乱雑な文字であると判断して文字認識結果を棄却する。例えば閾値＝２の場合、図８に示した文字記入欄８０１では、全ての文字枠において、左右の文字枠線と取り消し線とが接触しており、非文字枠画像に接触した文字枠線の本数＝２となる。そのために、文字認識結果が全て棄却される。また、取り消し線が曲線で描かれた文字記入欄８０２でも、やはり非文字枠画像に接触した文字枠線の本数が２以上となり、左３個の文字枠で文字認識結果が棄却となる。また、途切れた縦方向のノイズが重畳した文字記入欄８０３（ノイズが途切れて短いためにノイズ除去が困難）でも、左端の文字枠において非文字枠画像が上下２本の文字枠線と接触しているため棄却される。さらに、乱雑な文字が記入された文字記入欄８０４では、乱雑さに応じて非文字枠画像に接触した文字枠線の本数が増加し、２本以上の文字枠線に接触した文字が棄却となる。   Next, the ruled line contact rejection determination unit 111 determines that the character is a messy character and rejects the character recognition result when the number of character frame lines in contact with the non-character frame image is equal to or greater than a threshold value. For example, when threshold value = 2, in the character entry field 801 shown in FIG. 8, the left and right character frame lines and the strikethrough line are in contact with each other in all character frames, and the character frame line that has contacted the non-character frame image. The number is 2. Therefore, all character recognition results are rejected. Also, in the character entry field 802 in which the strikethrough is drawn with a curve, the number of character frame lines in contact with the non-character frame image is 2 or more, and the character recognition result is rejected in the left three character frames. In addition, even in the character entry field 803 in which discontinuous vertical noise is superimposed (noise removal is short and noise removal is difficult), the non-character frame image is in contact with the upper and lower two character frame lines in the leftmost character frame. Is rejected. Furthermore, in the character entry field 804 in which random characters are entered, the number of character frame lines in contact with the non-character frame image increases according to the randomness, and characters in contact with two or more character frame lines are rejected. Become.

次に、棄却処理統合部１０８は、スコアによる棄却判定部１０７の結果と、罫線接触棄却判定部１１１のどちらかで棄却と判定された場合、最終的に出力する文字認識結果を棄却する。双方の棄却判定部で棄却と判定されなかった場合は、文字パターン認識部１０６で得た文字認識結果を出力する。最後に結果出力部１０９は、棄却処理統合部１０８で得た最終的な文字認識結果を外部へ出力する。   Next, the rejection processing integration unit 108 rejects the character recognition result that is finally output when the result of the rejection determination unit 107 based on the score or the ruled line contact rejection determination unit 111 determines that the rejection is made. When both rejection determination units do not determine rejection, the character recognition result obtained by the character pattern recognition unit 106 is output. Finally, the result output unit 109 outputs the final character recognition result obtained by the rejection processing integration unit 108 to the outside.

このように実施の形態１の帳票文字認識装置は二つの棄却手段を持つ。第一に、スコアによる棄却判定部１０７では文字パターン認識部１０６で得た信頼度を指標として棄却判定を行うため、辞書と異なる形状を持つ文字パターンを棄却し、誤認識を抑制できる。第二に、罫線接触棄却判定部１１１では非文字枠の画素に接触した文字枠線の本数を指標として棄却判定を行うため、直線・非直線からなる取り消し線がある文字枠や、ノイズ除去が難しいノイズが重畳された文字枠、乱雑な文字が記入された文字枠を棄却し、さらに誤認識を抑制できる。
また、本帳票文字認識装置では、非文字枠画像に接触した文字枠線の本数を指標として棄却判定を行う簡便な処理で、三つの要因（取り消し線、ノイズ、乱雑な文字）に起因した誤認識を抑制できるため、処理コストが少ないという効果を有する。 Thus, the form character recognition apparatus of Embodiment 1 has two rejection means. First, since the rejection determination unit 107 by score performs rejection determination using the reliability obtained by the character pattern recognition unit 106 as an index, a character pattern having a shape different from that of the dictionary can be rejected, and erroneous recognition can be suppressed. Second, the ruled line contact rejection determination unit 111 performs rejection determination using the number of character frame lines that have contacted non-character frame pixels as an index. Character frames with difficult noise superimposed on them and character frames with messy characters filled in can be rejected, and erroneous recognition can be further suppressed.
In addition, this form character recognition device is a simple process that makes a rejection decision using the number of character frame lines in contact with a non-character frame image as an index, and errors caused by three factors (strikethrough, noise, and messy characters). Since recognition can be suppressed, there is an effect that processing cost is low.

以上説明したように、実施の形態１の帳票文字認識装置によれば、帳票画像上の文字を認識処理する帳票文字認識装置において、帳票画像中から文字認識対象とする文字記入欄位置を抽出するフィールド抽出部と、文字記入欄内の画像から文字枠線を検出する枠線検出部と、枠線検出部の検出結果に基づいて、文字記入欄内の画像から１文字分の文字パターンを取り出して辞書との照合を行う文字パターン認識部と、文字パターン認識部で得た認識結果の信頼度の値を元に認識結果の棄却判定を行うスコアによる棄却判定部と、枠線検出部の検出結果に基づいて、文字記入欄内の画像から１文字分の非文字枠画像を取り出して、非文字枠画像に接触している文字枠線の数を計測する文字・枠線接触数検出部と、非文字枠画像に接触している文字枠線数が所定の閾値を越える場合に、文字枠線に対応した文字の認識結果を棄却する罫線接触棄却判定部と、スコアによる棄却判定部の結果と罫線接触棄却判定部の結果のうち、どちらかの結果が棄却であった場合は、最終的に出力する文字認識結果を棄却する棄却処理統合部を備えたので、処理コストが小さく、かつ、誤認識を抑制することができる。 As described above, according to the form character recognition apparatus of the first embodiment, the form character recognition apparatus for recognizing the characters on the form image extracts the character entry column position as the character recognition target from the form image. Based on the detection result of the field extraction unit, the character frame line from the image in the character entry field, and the detection result of the frame line detection unit, the character pattern for one character is extracted from the image in the character entry field A character pattern recognition unit that performs collation with the dictionary, a rejection determination unit based on a score that performs a rejection determination of the recognition result based on the reliability value of the recognition result obtained by the character pattern recognition unit, and detection of the frame line detection unit Based on the result, a character / frame contact number detection unit that takes out a non-character frame image for one character from the image in the character entry field and measures the number of character frame lines in contact with the non-character frame image; Sentences that touch non-character frame images If the border number exceeds a predetermined threshold value, the ruled line contact rejection determination unit to reject the recognition result of the character corresponding to the character frame line, among the result of the rejection determination unit according to the score and the ruled line contact rejection determination unit results, When either result is rejection, since the rejection processing integration unit that rejects the character recognition result to be finally output is provided, the processing cost is low and erroneous recognition can be suppressed.

また、実施の形態１の帳票文字認識装置によれば、非文字枠画像は、文字、ノイズ及び取り消し線のうち少なくともいずれかを含むようにしたので、乱雑な文字やノイズあるいは取り消し線による誤認識を抑制することができる。   Further, according to the form character recognition apparatus of the first embodiment, the non-character frame image includes at least one of characters, noise, and strikethrough, so that misrecognition due to messy characters, noise, or strikethrough Can be suppressed.

尚、本願発明はその発明の範囲内において、実施の形態の任意の構成要素の変形、もしくは実施の形態の任意の構成要素の省略が可能である。   In the present invention, any constituent element of the embodiment can be modified or any constituent element of the embodiment can be omitted within the scope of the invention.

１０１ 帳票画像、１０２ 帳票文字認識装置、１０３ 認識結果、１０４ フィールド抽出部、１０５ 枠線検出部、１０６ 文字パターン認識部、１０７ スコアによる棄却判定部、１０８ 棄却処理統合部、１０９ 結果出力部、１１０ 文字・枠線接触数検出部、１１１ 罫線接触棄却判定部。   DESCRIPTION OF SYMBOLS 101 Form image, 102 Form character recognition apparatus, 103 Recognition result, 104 Field extraction part, 105 Frame line detection part, 106 Character pattern recognition part, 107 Rejection determination part by score, 108 Rejection process integration part, 109 Result output part, 110 Character / frame line contact number detection unit, 111 ruled line contact rejection determination unit.

Claims (2)

帳票画像上の文字を認識処理する帳票文字認識装置において、
前記帳票画像中から文字認識対象とする文字記入欄位置を抽出するフィールド抽出部と、
前記文字記入欄内の画像から文字枠線を検出する枠線検出部と、
前記枠線検出部の検出結果に基づいて、前記文字記入欄内の画像から１文字分の文字パターンを取り出して辞書との照合を行う文字パターン認識部と、
前記文字パターン認識部で得た認識結果の信頼度の値を元に当該認識結果の棄却判定を行うスコアによる棄却判定部と、
前記枠線検出部の検出結果に基づいて、前記文字記入欄内の画像から１文字分の非文字枠画像を取り出して、当該非文字枠画像に接触している文字枠線の数を計測する文字・枠線接触数検出部と、
前記非文字枠画像に接触している文字枠線数が所定の閾値を越える場合に、当該文字枠線に対応した文字の認識結果を棄却する罫線接触棄却判定部と、
前記スコアによる棄却判定部の結果と前記罫線接触棄却判定部の結果のうち、どちらかの結果が棄却であった場合は、最終的に出力する文字認識結果を棄却する棄却処理統合部を備えたことを特徴とする帳票文字認識装置。 In a form character recognition device that recognizes characters on a form image,
A field extraction unit for extracting a character entry column position to be a character recognition target from the form image;
A frame line detection unit for detecting a character frame line from the image in the character entry field;
A character pattern recognition unit that takes out a character pattern for one character from the image in the character entry field and collates it with a dictionary based on the detection result of the frame detection unit;
Rejection determination unit with a score that performs rejection determination of the recognition result based on the reliability value of the recognition result obtained by the character pattern recognition unit;
Based on the detection result of the frame line detection unit, a non-character frame image for one character is extracted from the image in the character entry field, and the number of character frame lines in contact with the non-character frame image is measured. A character / frame contact detection unit;
A ruled line contact rejection determination unit that rejects a recognition result of characters corresponding to the character frame line when the number of character frame lines in contact with the non-character frame image exceeds a predetermined threshold;
When one of the results of the rejection determination unit based on the score and the result of the ruled line contact rejection determination unit is rejection, a rejection processing integration unit that rejects the character recognition result that is finally output is provided. A form character recognition device characterized by that. 前記非文字枠画像は、文字、ノイズ及び取り消し線のうち少なくともいずれかを含むことを特徴とする請求項１記載の帳票文字認識装置。 The form character recognition apparatus according to claim 1, wherein the non-character frame image includes at least one of a character, noise, and a strikethrough.
JP2011048983A 2011-03-07 2011-03-07 Form character recognition device Expired - Fee Related JP5677139B2 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP2011048983A JP5677139B2 (en) 2011-03-07 2011-03-07 Form character recognition device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
JP2011048983A JP5677139B2 (en) 2011-03-07 2011-03-07 Form character recognition device

Publications (2)

Publication Number Publication Date
JP2012185713A JP2012185713A (en) 2012-09-27
JP5677139B2 true JP5677139B2 (en) 2015-02-25

Family

ID=47015746

Family Applications (1)

Application Number Title Priority Date Filing Date
JP2011048983A Expired - Fee Related JP5677139B2 (en) 2011-03-07 2011-03-07 Form character recognition device

Country Status (1)

Country Link
JP (1) JP5677139B2 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2018156308A (en) * 2017-03-16 2018-10-04 富士ゼロックス株式会社 Information processing device, information processing system, and program

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP7338158B2 (en) 2019-01-24 2023-09-05 富士フイルムビジネスイノベーション株式会社 Information processing device and program
JP7326753B2 (en) 2019-01-24 2023-08-16 富士フイルムビジネスイノベーション株式会社 Information processing device and program
WO2023248449A1 (en) * 2022-06-23 2023-12-28 株式会社東芝 Character recognition device, character recognition method, and program

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2877380B2 (en) * 1989-10-05 1999-03-31 株式会社東芝 Optical character reader
JP3345224B2 (en) * 1995-03-06 2002-11-18 富士通株式会社 Pattern extraction device, pattern re-recognition table creation device, and pattern recognition device
JPH0981666A (en) * 1995-09-08 1997-03-28 Toshiba Corp Character recognition device and recognized result display method
JP2000155803A (en) * 1998-11-20 2000-06-06 Nec Corp Character reading method and optical character reader
WO2000060531A2 (en) * 1999-04-07 2000-10-12 Raf Technology, Inc. Extracting user data from a scanned image of a pre-printed form
JP4580520B2 (en) * 2000-08-28 2010-11-17 株式会社日本デジタル研究所 Character recognition method and character recognition apparatus

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2018156308A (en) * 2017-03-16 2018-10-04 富士ゼロックス株式会社 Information processing device, information processing system, and program
US10796143B2 (en) 2017-03-16 2020-10-06 Fuji Xerox Co., Ltd. Information processing apparatus, information processing system, and non-transitory computer readable medium

Also Published As

Publication number Publication date
JP2012185713A (en) 2012-09-27

Similar Documents

Publication Publication Date Title
JP6286866B2 (en) Image processing apparatus and image processing method
US10740899B2 (en) Image processing apparatus for identifying region within image, information processing method, and storage medium
WO2013112746A1 (en) Detecting and correcting skew in regions of text in natural images
US8027539B2 (en) Method and apparatus for determining an orientation of a document including Korean characters
US8200043B2 (en) Page orientation detection based on selective character recognition
JP5677139B2 (en) Form character recognition device
Ayesh et al. A robust line segmentation algorithm for Arabic printed text with diacritics
JP6116531B2 (en) Image processing device
CA2999099A1 (en) Method and system for extracting information from hand-marked industrial inspection sheets
US9251412B2 (en) Segmentation of devanagari-script handwriting for recognition
JP7326753B2 (en) Information processing device and program
JP5039659B2 (en) Character recognition method and character recognition device
Mahastama et al. Improving Projection Profile for Segmenting Characters from Javanese Manuscripts
US20130051681A1 (en) System and method for script and orientation detection of images
Tikader et al. Edge based directional features for English-Bengali script recognition
Hanmandlu et al. Segmentation of handwritten Hindi text: A structural approach
JP2020119291A (en) Information processing device and program
Seraogi et al. Automatic Orientation Correction of AEC Drawing Documents
JP5669044B2 (en) Document verification system and document verification method
US20210073567A1 (en) Systems and methods for separating ligature characters in digitized document images
JP4270767B2 (en) Ruled line information processing method, ruled line information processing program, image processing apparatus, and storage medium
JP2002366900A (en) Optical character reader
JP6833324B2 (en) Information processing equipment, programs and information processing methods
JP3428504B2 (en) Character recognition device
JP2013152523A (en) Document character difference detection device

Legal Events

Date Code Title Description
A621 Written request for application examination

Free format text: JAPANESE INTERMEDIATE CODE: A621

Effective date: 20131030

A977 Report on retrieval

Free format text: JAPANESE INTERMEDIATE CODE: A971007

Effective date: 20140620

A131 Notification of reasons for refusal

Free format text: JAPANESE INTERMEDIATE CODE: A131

Effective date: 20140701

A521 Request for written amendment filed

Free format text: JAPANESE INTERMEDIATE CODE: A523

Effective date: 20140828

TRDD Decision of grant or rejection written
A01 Written decision to grant a patent or to grant a registration (utility model)

Free format text: JAPANESE INTERMEDIATE CODE: A01

Effective date: 20141202

A61 First payment of annual fees (during grant procedure)

Free format text: JAPANESE INTERMEDIATE CODE: A61

Effective date: 20141226

R150 Certificate of patent or registration of utility model

Ref document number: 5677139

Country of ref document: JP

Free format text: JAPANESE INTERMEDIATE CODE: R150

LAPS Cancellation because of no payment of annual fees