JPH03208184A - Character photograph automatic recognition processing system - Google Patents

Abstract

PURPOSE:To decide which of a photograph picture element or a character picture element an input picture element is by storing every picture element of input data on a matrix separately, and comparing the sum of squares of difference between the mean value of the picture element of the matrix and each picture element with a threshold. CONSTITUTION:A reading part 1 converts an original picture in which a character and a photograph are mixed into the many picture elements of different densities, and reads it as digital data. A CPU 2 stores this data temporarily in a memory 3, and controls a mesh processing part 5 and a variance processing part 6 so as to recognize the character and the photograph. Here, the neighboring picture elements of every picture element of the digital data are stored on the matrix, and the sum of squares of the difference between the mean value of each picture element and each picture element is calculated, and the processing part 6 compares this sum of squares with the threshold, and decides which of the photographic element or the character picture element it is. Besides, each area of the original picture is divided into meshes, and the processing part 5 counts the picture element decided to be the photograph picture element for every mesh, and decides the character element or the photograph picture element by comparing it with the threshold. Thus, the precision of the separation of the image area of the character and the photograph is improved, and the character becomes easy to see.

Description

【発明の詳細な説明】 ［産業上の利用分野コ 本発明は文字と写真が混在する原画像を自動的に処理し
て文字と写真とを識別する方式の改良に関する． ［発明の概要］ 文字と写真が混在する原画像を構戒する多数の濃度の異
なるＷＪ素から成る入力データに対し，分散処理、メッ
シュ処理及び又はエッジ処理を施して文字と写真とを正
確に認識する方式に関するものである． ［従来の技術］ 文字画像と写真画像が混在している原稿を光学的に読み
取ってディジタル処理して得られた画像情報を再生した
場合，その再生画像は大変見すらいものであった．その
ため文字と写真の像域分離を自動的に認識し、再生画像
を見やすくする処理方式として、従来からエッジ処理及
びメッシュ処理の方式が提案されている． ［発明が解決しようとする側題］ しかしながら従来の方式は画素の濃度差に着目するのみ
であるため，太い文字の場合、太い文字の非エッジ部は
濃度差がないため文字として認識されず、写真として認
識されてしまう不具合があった・ ［発明の目的］ 従って本発明の目的は文字と写真の像域分離を自動的に
認識し再生された′ｆｇＪ像を見やすくする文字写真自
動認識処理方式を提但するにある．〔課題を解決するた
めの手段〕 本発明は上記目的を達或するため、文字と写真が混在す
る原画像を濃度の異なる複数の画素に変換して成る入力
データを取り込んで、夫々の画素毎にその周囲の画素を
３Ｘ３のマトリクス上に格納し、各マトリクスの画素の
平均値と各画別との差の２乗和を演算し，その２乗和と
しきい値と比較して写真画素か文字画素かを判定する分
散処理手段と、上記ｙＫ画像の各領域を複数個の２×２
のメッシュに区切り、夫々のメッシュについて写真画素
と判定された画素を計数してその計数値をしきい値と比
較し、その結果によりメッシュ毎に写真領域か文字領域
かを決定するメッシュ処理手段と，から成ることを要旨
とする． ［作用］ メッシュ処理手段及び又はエッジ処理手段の他に、分散
処理手段を設けているので、特に太い文字に対する認識
精度が向上する． ［実施例］ 以下図面に示す実施例を参照して本発明を説明する．第
１図は本発明による文字写真自動認識処理方式の一実旅
例の基本的システム概念図を示す．同図において、　１
　は読取部，　２は中央演算部（ＣＰＵ）　、３はメモ
リ（ＲＡＭ）．４はＲＯＭ、５はメッシュ処理部、６は
分散処理部、７はエッジ抽出処細部である． 読取部ｌは公知の光学的文字読取装［　（ＯＣＲ）等か
ら戊り、文字と写真が混在するＪ）Ｘ画像を濃度の具々
る多数の画素に変換してディジタルデータとして読み取
る． ＣＰＵ２は読み取られたデータを一時的にメモリ３に蓄
積しておき、下記のようにメッシュ処理部５，分散処理
部６及びエッジ抽出処理部７を制御して文字と写真とを
認識するように処理させる．なおＲＯＭ４は上記制御に
必要なプログラムを格納している． 而して本発明の第１の方式では上記画素データに対し分
散処理とメッシュ処理を施す．この分散処理は入力デー
タ（１画素＝１ドッ−ト）である注目のｌ画素に対して
その近傍の８画素の状況を見て所定の演算を行い．その
結果をしきい値と比鮫しながらその注目画素が文字画素
か写真画素かを判定する． 上記分散処理の演算は分散処理部６により次のようにし
て行われる． ３×３のマトリクス上に格納された注目画素を含む９個
の画素の濃度平均値を求め，この平均値と各画ｉｉ４濃
度との２乗和を計算する．今、この２乗和をＳＤ．平均
値をｍ，各画素濃度をＸ工（ｉ＝０〜８）とすると， ＳＤ＝Σ　（Ｘｔｍ）” Ｌ＝０ このＳＤは９個の画素に比例する． 文字の入力データではその文字のエッジ濾度分布に対し
ＳＤが大きく、また太い均一濃度の文字領域のＳＤは小
さい．写真の画素濃度はほぼＳＤの分布をもっている． 本発明者の実験（コンピュータシミュレーション）的検
討によればＳＤのしきい値をＳａ，　Ｓｂ，Ｓａ＜Ｓｂ
とすると、文字領域の画素についてはＳＤ（ＳＤａ又は
ＳＤｂ（ＳＤ．写真領域についてはＳＤａ≦ＳＤ≦ＳＤ
ｂとして判別できる．このような分散処理により細い文
字，太い文字及び画素濃度変化の穏やかな写真の判別を
正確に行うことができる． 次にメッシュ処理部５は判別情報をより正確にするため
、入力データの全領域をＭＸＮ個のメッシュに区切り、
更にその１つのメッシュを２×２の部分に区分し、各メ
ッシュについて，その４個の部分の文字画素の数を計数
し、その計数値ＳＩＩがしきい値Ｓｃより大きいと文字
領域、Ｓｃより小さいと写真領域と判定し、像域分離を
行う．第２図は上述した分散処理とメッシュ処理を行う
アルゴリズムを示すフローチャートである．さて本発明
の第２の方式は上述した第１の方式の分散処理に平行し
てエッジ抽出処理を行うことにより特に細い文字の場合
の認識精度を更に上げるようにしている． 上記エッジ抽出処理部７では前記分散処理における９個
の画素が格納されているマトリクスに対し３Ｘ３の加重
マトリクスを乗算して和をとり各Ｉｉｉ素の濃度を変換
してエッジを抽出する．このようにして抽出された画素
の濃度としきい値Ｔとを比較してエッジ画素（文字画素
）か非エッジ画素（写真画素）かを判定する． ＣＰＵ２はエッジ抽出処理部７及び分敗処理部６での判
定結果をとり写Ｘｉ素か文字画素かを判定し、前記メッ
シュ処理に移行する． 第３図は上記第２の方式のアルゴリズムを示すフローチ
ャートである． ［発明の効果］ 以上説明したように本発明によれば、太い文字、細い文
字及び文字と写真の像域分離の粘度が従来の方式よりも
向上して見やすくなり、不自然さのないＷＩ像データが
得られ、特に太い文字に対しても前記アルゴリズムで文
字として判断することが可能となるので実用上の効果顕
著である．DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an improvement in a method for automatically processing an original image containing a mixture of text and photographs to distinguish between text and photographs. [Summary of the invention] Dispersion processing, mesh processing, and/or edge processing is applied to input data consisting of a large number of WJ elements with different densities, which form an original image in which text and photos are mixed, to accurately separate text and photos. It is related to the recognition method. [Prior Art] When the image information obtained by optically reading and digitally processing a document containing a mixture of text images and photographic images is reproduced, the reproduced image is very unpleasant to look at. Therefore, edge processing and mesh processing methods have been proposed as processing methods that automatically recognize the image area separation between text and photos and make the reproduced image easier to see. [Side problem to be solved by the invention] However, since the conventional method only focuses on the density difference between pixels, in the case of a thick character, the non-edge part of the thick character is not recognized as a character because there is no density difference. There was a problem that the image was recognized as a photograph. [Object of the Invention] Therefore, the purpose of the present invention is to provide an automatic text/photo recognition processing method that automatically recognizes the image area separation between text and photos and makes it easier to see the reproduced 'fgJ image. However, I would like to propose the following. [Means for Solving the Problems] In order to achieve the above object, the present invention imports input data obtained by converting an original image containing text and photographs into a plurality of pixels with different densities, and converts each pixel into Then, store the surrounding pixels on a 3x3 matrix, calculate the sum of squares of the difference between the average value of the pixels in each matrix and each image, and compare the sum of squares with the threshold value to determine whether the photo pixel A distributed processing means for determining whether it is a character pixel, and a plurality of 2×2 pixels for each area of the yK image.
mesh processing means for dividing each mesh into meshes, counting pixels determined to be photographic pixels for each mesh, comparing the counted value with a threshold value, and determining whether each mesh is a photographic area or a text area based on the result; The gist is that it consists of . [Operation] Since a distributed processing means is provided in addition to the mesh processing means and/or the edge processing means, recognition accuracy is particularly improved for thick characters. [Examples] The present invention will be described below with reference to examples shown in the drawings. Figure 1 shows a basic system conceptual diagram of an example of the automatic text and photo recognition processing method according to the present invention. In the same figure, 1
is a reading unit, 2 is a central processing unit (CPU), and 3 is a memory (RAM). 4 is a ROM, 5 is a mesh processing section, 6 is a distributed processing section, and 7 is an edge extraction processing section. The reading unit 1 uses a known optical character reading device (OCR), etc., to convert an image containing a mixture of text and photographs into a large number of pixels of varying density and read it as digital data. The CPU 2 temporarily stores the read data in the memory 3, and controls the mesh processing section 5, distributed processing section 6, and edge extraction processing section 7 to recognize characters and photographs as described below. Let it be processed. Note that ROM4 stores programs necessary for the above control. In the first method of the present invention, the above pixel data is subjected to distributed processing and mesh processing. This distributed processing performs a predetermined operation on a pixel of interest, which is input data (1 pixel = 1 dot), by looking at the status of 8 pixels in its vicinity. The result is compared with a threshold value to determine whether the pixel of interest is a text pixel or a photo pixel. The above distributed processing calculations are performed by the distributed processing unit 6 as follows. The average density value of nine pixels including the pixel of interest stored on a 3x3 matrix is determined, and the square sum of this average value and each pixel ii4 density is calculated. Now, convert this sum of squares to SD. If the average value is m and each pixel density is The SD is large with respect to the edge filtration distribution, and the SD of the thick uniform density character area is small.The pixel density of a photograph has an approximately SD distribution.According to the inventor's experimental study (computer simulation), the SD is large. Set the threshold value to Sa, Sb, Sa<Sb
Then, for the pixels in the text area, SD (SDa or SDb (SD). For the photo area, SDa≦SD≦SD
It can be identified as b. This kind of distributed processing makes it possible to accurately distinguish between thin characters, thick characters, and photographs with gentle pixel density changes. Next, in order to make the discrimination information more accurate, the mesh processing unit 5 divides the entire area of the input data into MXN meshes,
Furthermore, one mesh is divided into 2 × 2 parts, and for each mesh, the number of character pixels in the four parts is counted, and if the counted value SII is larger than the threshold value Sc, the character area is If it is small, it is determined to be a photographic area and image area separation is performed. Figure 2 is a flowchart showing the algorithm for performing the above-mentioned distributed processing and mesh processing. Now, the second method of the present invention performs edge extraction processing in parallel to the distributed processing of the first method described above, thereby further increasing the recognition accuracy especially for thin characters. The edge extraction processing unit 7 multiplies the matrix in which the nine pixels in the distributed processing are stored by a 3×3 weighted matrix, adds up the sum, converts the density of each III element, and extracts an edge. The density of the pixel thus extracted is compared with a threshold value T to determine whether it is an edge pixel (text pixel) or a non-edge pixel (photo pixel). The CPU 2 takes the determination results from the edge extraction processing unit 7 and the division processing unit 6, determines whether the pixel is a copy Xi element or a character pixel, and proceeds to the mesh processing. FIG. 3 is a flowchart showing the algorithm of the second method described above. [Effects of the Invention] As explained above, according to the present invention, the viscosity of image area separation between thick characters, thin characters, and characters and photographs is improved compared to conventional methods, making it easier to see, and creating WI images without unnaturalness. This has a significant practical effect because data can be obtained and even thick characters can be judged as characters using the algorithm.

【図面の簡単な説明】[Brief explanation of drawings]

第工図は本発明の一実施例の基本的システム構或を示す
ブロック図、９５２図及び第３図は夫々本発明の第１及
び第２の方式のアルゴリズムを示すフローチャートであ
る． 工・・・・・・・・・画像読取部、２・・・・・・・・
・ＣＰＵ、３・・・・・・・・・メモリ，４・・・・・
・・・・ＲＯＭ，５・・・・・・・・・メッシュ処理部
，６・・・・・・・・・分敗処理部、７・・・・・・・
・・エッジ抽出処理部． 特訂出願人　　京セラ株式会社 第２図Fig. 9 is a block diagram showing the basic system configuration of an embodiment of the present invention, and Fig. 952 and Fig. 3 are flowcharts showing the algorithms of the first and second methods of the present invention, respectively. Engineering... Image reading section, 2...
・CPU, 3...Memory, 4...
ROM, 5...Mesh processing unit, 6...Distribution processing unit, 7...
...Edge extraction processing section. Special applicant Kyocera Corporation Figure 2

Claims (2)

【特許請求の範囲】[Claims] （１）文字と写真が混在する原画像を濃度の異なる複数
の画素に変換して読み取る入力データを取り込んで、夫
々の画素毎にその周囲の画素を３×３のマトリクス上に
格納し、各マトリクスの画素の平均値と各画素との差の
２乗和を演算し、その２乗和としきい値と比較して写真
画素か文字画素かを判定する分散処理手段と、 （ｎは自然数） 上記原画像の各領域をを複数個のｎ×ｎ（ｎは自然数）
のメッシュに区切り、夫々のメッシュについて写真画素
と判定された画素を計数してその計数値をしきい値と比
較し、その結果によりメッシュ毎に写真領域か文字領域
かを決定するメッシュ処理手段と、から成ることを特徴
とする文字写真自動認識処理方式。(1) Input data is read by converting an original image containing a mixture of text and photos into multiple pixels with different densities, and storing the surrounding pixels for each pixel in a 3 x 3 matrix. a distributed processing means that calculates the sum of squares of the difference between the average value of the pixels of the matrix and each pixel, and compares the sum of squares with a threshold value to determine whether the pixel is a photographic pixel or a text pixel; (n is a natural number); Each area of the above original image is divided into multiple n×n (n is a natural number)
mesh processing means for dividing each mesh into meshes, counting pixels determined to be photographic pixels for each mesh, comparing the counted value with a threshold value, and determining whether each mesh is a photographic area or a text area based on the result; An automatic text/photo recognition processing method characterized by comprising the following steps. （２）文字と写真が混在する原画像を濃度の異なる複数
の画素に変換して成る入力データを取り込んで、夫々の
画素毎にその周囲の画素を３×３のマトリクス上に格納
し、各マトリクスの画素の平均値と各画素との差の２乗
和を演算し、その２乗和としきい値と比較して写真画素
か文字画素かを判定する分散処理手段と、 上記画素格納マトリクスに対し３×３の加重マトリクス
を乗算して、和をとり濃度変換された各画素としきい値
とを比較してエッジ画素か非エッジ画素かを判定するエ
ッジ抽出処理手段と、 上記分散処理手段とエッジ抽出処理手段との判定結果に
応じて写真画素か文字画素かを判定する判定手段と、 上記原画像の各領域を複数個のｎ×ｎのメッシュに区切
り、夫々のメッシュについて写真画素と判定された画素
を計数してその計数値をしきい値と比較し、その結果に
よりメッシュ毎に写真領域か文字領域かを決定するメッ
シュ処理手段と、から成ることを特徴とする文字写真自
動認識処理方式。(2) Input data is obtained by converting an original image containing a mixture of text and photos into multiple pixels with different densities, and for each pixel, the surrounding pixels are stored in a 3 x 3 matrix, and each a distributed processing means that calculates the sum of squares of the difference between the average value of the pixels of the matrix and each pixel, and compares the sum of squares with a threshold value to determine whether the pixel is a photographic pixel or a text pixel; an edge extraction processing means that multiplies the pixels by a 3×3 weighted matrix, calculates the sum, and compares each density-converted pixel with a threshold value to determine whether it is an edge pixel or a non-edge pixel; determining means for determining whether a pixel is a photographic pixel or a character pixel according to the determination result with the edge extraction processing means; and dividing each area of the original image into a plurality of n×n meshes, and determining each mesh as a photographic pixel. automatic text/photo recognition processing, comprising: a mesh processing means that counts the pixels that have been detected, compares the counted value with a threshold value, and determines whether each mesh is a photo area or a text area based on the result; method.