JPH11338977A - Method and device for character processing and storage medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize a recognition capable of obtaining a high recognition rate without operator's recognition and instruction for language type by individually performing the character recognition processing suitable to each character area in accordance with the discriminated language type. SOLUTION: After preprocessings such as noise elimination and inclination correction, individual rectangular areas such as text blocks, figure blocks, and picture blocks are extracted from picture data of one page by area division processing, and area information indicating positions of individual areas and attribute information indicating classifications of areas are stored in a memory 102. Each of text blocks out of these blocks is subjected to language type discrimination processing, and the obtained result is stored in the memory 102 as the language attribute of the block correspondingly to area information and attribute information of the block. When the language attribute is English, a picture in the block is subjected to character recognition by a character recognition device for English; but when the language attribute is not English, it is subjected to character recognition by character recognition device for Japanese.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character processing method and apparatus for character recognition of a document image in which different languages are mixed, and a storage medium.

[0002]

2. Description of the Related Art An ordinary character recognition device basically targets a single language. For example, a Japanese character recognition device is used for a Japanese document, and an English character recognition device is used for an English document. Character recognition was performed using a device.

[0003] The English document referred to here is a document containing only characters used in English, that is, letters, numbers and symbols, and the English character recognition device can recognize only those characters.

On the other hand, Japanese documents are rarely composed of only Japanese characters such as Kana-Kanji characters, and generally include alphabets and numerals. That is, even if it is a character recognition device for Japanese,
It is also required that it can recognize English characters such as alphabets.

However, the recognition performance of a Japanese character recognition device for a wider range of characters is inferior to that of an English character recognition device for recognizing an English document in advance by recognizing it as English. I can't deny it. If the character recognition device must randomly recognize Japanese and English documents, or if it recognizes a document that is mainly Japanese but contains some English document blocks, these are all used for Japanese. When the recognition is performed only by the character recognition device, the problem is the low recognition performance of the English part.

In order to eliminate such disadvantages, one character recognition device has two units, an English character recognition unit and a Japanese character recognition unit, and the operator can use Japanese for each document or for each part of the document. A Japanese character recognition part is used for the Japanese sentence part where the area and language type are specified by the operator's manual operation. By recognizing the sentence part using the English character recognition unit,
Although the operation is complicated, there is a character recognition device that can obtain a higher recognition rate.

[0007]

However, in the above-mentioned example, high recognition performance cannot be obtained unless the operator properly specifies the area before the recognition and selects the recognition unit to be used for each area. . In particular, when recognizing a document in which text regions in different languages are mixed, the trouble is great.
This greatly hindered full automation, which greatly impaired the practicality of the character recognition device.

[0008]

In order to solve the above problems, the present invention extracts a plurality of character regions from an image, determines a language type for each of the plurality of extracted character regions,
Provided are a character processing method and apparatus, and a storage medium that individually perform character recognition processing suitable for each character area according to the determined language type.

In order to solve the above problem, the present invention preferably extracts characters from an image in the character area, and determines the language type of the character area based on the width of each character.

In order to solve the above-mentioned problem, the present invention preferably obtains an average value of the width of the character for each line, and determines the language type based on comparison with a predetermined threshold value.

In order to solve the above problem, the present invention preferably compares the average value of the width of the character for each line with a threshold value, and finds that the ratio of the number of lines whose average value exceeds the threshold value to the total number of lines is determined. The language type is determined based on the language type.

In order to solve the above problems, the present invention preferably calculates the mode from the peak of each distribution for the pitch and spacing of characters in a line, and uses the difference as the average value for each line.

In order to solve the above-mentioned problem, the present invention preferably extracts characters from an image in the character area, and determines the language type of the character area based on the number of horizontal lines included in each character.

In order to solve the above-mentioned problem, the present invention preferably divides the image of the extracted character into two dots in horizontal width, scans the character vertically, and counts a specific arrangement pattern. Determine the number.

In order to solve the above problem, the present invention preferably extracts a line from an image in the character area, and determines a language type of the character area based on a reference line of each line.

In order to solve the above-mentioned problem, in the present invention, preferably, the reference line of the line is a line representing the upper and lower ends of characters included in the line.

In order to solve the above problem, the present invention preferably extracts characters from an image in the character area, divides the image of each character into a fixed width, and sets upper and lower black pixels detected in each of the images. A reference line for each row is determined from the distribution at the end.

In order to solve the above problems, the present invention preferably includes a plurality of the language type discriminating methods.

In order to solve the above-mentioned problem, the present invention preferably determines a line direction of the extracted character area, and if the determined line direction is a specific direction, performs the language type determination. Instead, the language type is determined in advance.

In order to solve the above problem, the present invention preferably provides information indicating a position of the extracted character area,
The information indicating the determined language type is stored in association with the information.

In order to solve the above problems, the present invention preferably outputs the image information and the result of the character recognition by the same output device.

In order to solve the above problem, the present invention preferably sets one of the determined language types to Japanese.

In order to solve the above-mentioned problem, in the present invention, preferably, one of the determined language types is English.

[0024]

FIG. 1 is a block diagram of an apparatus according to the present invention.

In FIG. 1, reference numeral 101 denotes a CPU (Central Processing Unit) which controls processing according to the present invention in accordance with a control program stored in a memory 102. Each step of the flowchart described later is also executed by the CPU 101. The memory 102 is a storage device including a RAM, a ROM, a hard disk, and stores a control program of the CPU 101, various parameters, input image data, a character recognition dictionary, and the like. Reference numeral 103 denotes an external storage medium such as an optical disk, a magnetic disk, a magneto-optical disk, or a magnetic tape, which is detachable from the apparatus. Programs and data stored in the memory 102 are read from the external storage medium. The processing result is stored in the external storage medium 103.
Is output to Reference numeral 104 denotes a communication I / F for exchanging data with the other end via a network or a public line, and programs and data stored in the memory 102 and processing results are input and output via the communication I / F. It may be done. Reference numeral 105 denotes an input unit such as a keyboard and a pointing device, and transmits an operator's instruction. Instructions for inputting an image, starting character recognition, or storing a text as a result of the recognition processing are input from the input unit 105. Reference numeral 106 denotes a scanner device that optically reads a document and inputs the document as an electric signal to the device. 107 is C
A display device such as an RT or a liquid crystal displays a text of a processing result and serves as one of interfaces for an operator's operation. Reference numeral 108 denotes an LBP or inkjet printer, which outputs the text of the processing result on paper in a font.

FIG. 2 is a functional block diagram of the processing of the character recognition device according to the first embodiment of the present invention. An image input unit 201 reads a document document as image data. The image input unit 201 includes a scanner 106, an external storage medium 103, and a communication I / F.
Through the step 104, the image data is stored in the memory 102. Reference numeral 202 denotes an area dividing unit that separates respective areas into text blocks, figure blocks, table blocks, image blocks, and the like from one page of document image data in which text, figures, and tables are stored in the memory 102. ((A) of FIG. 10 is an image of a document,
(B) shows the area data obtained as a result of area division by 2 and the types of each area. ). Reference numeral 203 denotes a language determination unit that determines whether the text in the text block to be subjected to character recognition is in Japanese or English. Reference numeral 204 denotes an English character recognizing unit that recognizes English texts, such as alphabets, numbers, and symbols.
Recognize only characters that appear in English sentences. Reference numeral 205 denotes a Japanese character recognition unit that recognizes all characters that appear in Japanese sentences, such as English, Kanji, and Kana. The text data obtained as a result of character recognition is displayed on the display in font, and then
Output to a file, network destination computer, or printer.

FIG. 3 shows the overall processing according to the embodiment of the present invention.
Will be described along the flowchart of FIG.

In step S301, an image is read via the scanner 106, the external storage medium 103, or the communication I / F 104, and stored in the memory 102. In S302, after pre-processing such as noise removal and inclination correction, in S303, each rectangular area such as a text block, a figure block, a photograph block, a table block, and an image block is converted from one page of image data by the area division processing. Is extracted, and area information indicating the position of each area and the type of the area (text, figure, photograph,
Attribute information representing tables, images, etc.) is stored in the memory 102. In S304, a process of determining the language type is performed on each of the text blocks among these blocks, and the obtained result is associated with the region information and the attribute information of the block as the language attribute of the block. It is stored in the memory 102. In S305, character recognition of each text block is performed. In S306, the result of the character recognition and the block data (area information,
Layout data including attribute information and language attribute information).

The details of the character recognition process in step S305 in FIG. 3 will be described with reference to the flowchart in FIG. A block counter B for managing each block subjected to the processing of the flowchart of FIG. 4 is initialized to 1.

In step S401, the block data of the text block specified by the counter B is read. In step S402, the language attribute of the block is referred to. If the language attribute is English, the process advances to step S403 to perform character recognition on the image in the block using the English character recognition device. If the language attribute is other than English, that is, Japanese and the language is unknown, character recognition by Japanese character recognition is performed. S40
At 4, it is determined that the counter B matches the number of text blocks. If all text blocks have been processed, this character recognition process is terminated. If it is determined that text blocks remain, Counter B is set to 1 in S406
The value is incremented and returns to S401, and the processing of S401 to S404 is repeated for the next text block.

The details of the language type determination processing in step S304 in FIG. 3 will be described with reference to the flowchart in FIG. This language type determination processing is performed on all blocks determined to be text in S303.
The flowchart of FIG. 5 explains the processing performed for one block.

In step S501, block data of a text block to be processed is read. In step S502, it is determined whether the text is written vertically or horizontally. The determination method compares the variance of the vertical and horizontal projections of the black pixels in the text block, and if the variance is greater in the horizontal direction, the vertical writing is performed. If it is determined that the text is vertically written, it cannot be English, so the process proceeds to step S503, where the language attribute of the text block is set to Japanese and the language attribute information is stored in the memory 102.
finish.

If horizontal writing is determined in S502, the process proceeds to a process of determining whether the writing is Japanese or English. First, S504
The line region is cut out by taking the horizontal projection of the black pixel in the text block. Next, in step S505, a character region is cut out by taking the vertical projection of black pixels in each line region. Both processes are performed using the breaks of black pixel projection. The results are obtained as follows, as the coordinate values of a circumscribed rectangle that surrounds the pixels of each line and character. These data are stored in the memory 102 as line area information and character area information in association with the block data.

[0034]

[Outside 1]

In S506, the cut out rows are classified. For each line, if the number of characters extracted from the line is equal to or greater than a threshold or the line height is within a predetermined range,
The line is identified as a line that can be used for language type determination in steps 509 to S515. The threshold of the number of characters and the range of the line height used for the identification are stored in the memory 102 in advance as parameter data.

In S507, if no line usable for determination is identified, the flow advances to S508 to store the language attribute information in the memory 102 as the language attribute of the text block is unknown, and the process ends. If there is a row that can be used for the determination, the process proceeds to S509.

In step S509, the inclination is statistically obtained from the set of character rectangles, and the coordinate value of the rectangle is corrected with the calculated inclination. Specifically, (x ₀ ^{i, j} −X ₀₎ for all character rectangles c _{i, 1}
ⁱ , y ₀ ^{i, j} −Y ₀ ⁱ ) to obtain a least square curve from the distribution,
The coordinate value of the character rectangle is corrected using the inclination d.

After S510, S510, S512,
When the linguistic features can be extracted from within the text block, the language type of the block is determined by using the three determination methods of S513 in stages. Details of each determination method will be described later.

In step S510, it is checked whether or not Japanese character can be extracted from the text block using the character width feature. If the Japanese attribute is obtained by the determination,
Proceeding to 511, the language attribute of the text block is stored in the memory 102 as Japanese, and the process ends. If not, the process proceeds to S512. Details of the determination processing based on the character width are shown in the flowchart of FIG.

In S512, it is checked whether or not Japanese character can be extracted from the text block by using the horizontal line degree of the character. If the Japanese attribute is obtained by the determination,
Proceeding to 511, the language attribute of the text block is stored in the memory 102 as Japanese, and the process ends. If not, the process proceeds to S513. Details of the determination processing based on the degree of horizontal line are shown in the flowchart of FIG.

In step S513, the English language or the Japanese language is extracted from the text block by the judgment based on the reference line. If the English attribute is obtained by the determination, S5
Proceeding to 14, the language attribute of the text block is set to English, and the processing ends. If the Japanese attribute is obtained, the process proceeds to S511, where the language attribute of the text block is set to Japan and stored in the memory 10
2 and end. If neither language attribute is obtained, the process proceeds to S515, where the language attribute of the text block is stored as unknown in the memory 102, and the process ends. Details of the determination process using the reference line are shown in the flowchart of FIG.

The details of the language type determination processing using the character width characteristics in step S510 in FIG. 5 will be described. This is a process of assigning Japanese attributes to a text block from which Japanese-like features have been extracted by using the fact that the width of Japanese characters and English characters can be statistically classified.
Specific processing contents will be described with reference to the flowchart in FIG.

In step S601, the values of various counters are cleared. The processing from S602 is performed for each line in the text block. In S602 of character rectangles in a character row L _i width ^{_{{(x 1 i, j -x}} 0 i, j +1)} arithmetic mean w of
_Ask for _i .

In step S603, it is determined whether or not the number of characters in the line is equal to or greater than a certain threshold, and whether or not the line is appropriate for determination is identified. If it is less than or equal to Nmin, u that counts the lines for which the counter language could not be determined in S604 is incremented by +
Do one. If there is a character of Nmin or more, the process proceeds to S605.

In S605, the character pitch between the adjacent rectangles, the mode of {(x ₀ ^{i, j + 1} −x ₀ ^{i, j} +1)} and the character spacing {(x ₀ ^{i, j + 1} − x ₁ ^{i, j} +1)} is determined, and the difference is defined as p _i . is a value corresponding to the width of the characters in the p _i row L _i.

Note that the numerical sequence {s ₁ , s ₂ ,. . . , S _n ｝
Is obtained from the coordinates at which the maximum point of the following distribution function D (s) is obtained (FIG. 11).

[0047]

[Outside 2]

[0048] At S606, to compare the value divided by the height of the row of p _i with the threshold value Pth. If p _i is smaller than Pth, the process proceeds to S607, in which the line is regarded as English, and e for counting the lines determined to be English is incremented by one.
If it is larger, the process proceeds to S608, in which the line is estimated to be Japanese, and j that counts lines determined to be Japanese is incremented by one.

In S609, the average value of the character width w _i is calculated.

[0050]

[Outside 3] Compare with the threshold value Wth. If w _i is large, the process proceeds to S610, where c is incremented by one. This means that the current line contains many character rectangles by touching characters, and c is the number of the line.

In step S611, the line counter L is compared with the number of lines. If it is determined that all the lines have not been processed yet because they do not match, L is incremented by one in step S620, and step S60 is performed.
Return to 2. If it is determined that all have been processed, S61
Proceed to 2. In S612, it is checked whether or not the ratio of c to the total number of rows is larger than the threshold value Vc. If it is larger than Vc, the process proceeds to S613, and for the text block,
An attribute indicating that there is a lot of character contact is given, and the process ends without performing language determination for this block. If the ratio is equal to or lower than Vc, the process proceeds to S614.

In S614, the ratio of the number j of lines regarded as Japanese to the total number of lines is compared with a threshold value Vj. If the ratio is greater than Vj, the process proceeds to S615, where the text block is stored in Japanese as the language attribute information in the memory 102, and the processing ends. If the ratio is small, the process ends without giving any attribute.

Next, a description will be given of the determination processing using the characteristic of the horizontal line degree of the character in step S511 in FIG.

A character is vertically cut at an arbitrary position, and the number of character lines crossed by the cross section at that time is counted. The number of character lines cut at the position where this numerical value is maximum will be referred to as the horizontal line degree of the character.

In the case of English text characters such as alphabets and numbers, the degree of horizontality is 4 or less for many fonts, and none of them is 5 or more. On the other hand, there are many Japanese characters, such as kanji such as "country", in a number of five or more. That is, if the character has a horizontal line degree of 5 or more, it can be estimated that the character is a Japanese character.

In step S511 in FIG. 3, this horizontal line degree is obtained for all the character rectangles in the text block, and if a character rectangle that seems to be a Japanese character is found, the Japanese attribute is added to the text block. This is the process of giving.

Specific processing contents will be described with reference to the flowchart of FIG.

By managing all the character rectangles in the text block by the counter ch, the horizontal linearity is determined for all the characters in S701.

The image of the character is divided into bands having a width of 2 dots.
The columns of the horizontal 2-dot pixel values in each of the bands are {b ₁ , b ₂ ,. . . , B _m } ({b _k } = {00, 0
1, 10, 11}). Also, 0 at the beginning and end of the column
0 is supplemented (b ₀ = b _{m + 1} = 00).

At this time, between b _{0 and} b _{m + 1} , 00 (00/01/10) * 11 (11/10/01)
The number of times the pattern arrangement of * 00 appears is counted, where (a | b) is either a or b, and * means 0 or more repetitions of the immediately preceding pattern. That is, 0
Except for 1 and 10, the number of times of 00 → 11 → 00 is obtained.

The number of occurrences of the pattern described above is checked for all the bands in the character (FIG. 12), and the maximum value is used as the character horizontal line degree. In this method, the processing 2
Since the dot width is provided and the repetition of the 01 and 10 patterns mainly due to the noise at the time of binarization is ignored, the horizontal linearity can be counted stably with respect to the noise.

In step S702, the horizontal linearity c is compared with a threshold value Cth. If c is larger than Cth, the process proceeds to S704, where the Japanese attribute is stored in the memory 10 in association with the text block, and the process ends. Otherwise, the process proceeds to S703. S7
At 03, ch is compared with the total number of characters, and when they do not match, it is determined that all characters have not been processed, and at S705, ch is set to 1
Increment and repeat from S701 for the next character. If it is determined that all characters have been processed, the process ends without giving a language attribute.

As a third determination process, the determination process using the reference line in step S513 in FIG. 5 will be described.

As shown in FIG. 13, in the English line, apart from the upper and lower ends of the character line, the reference line where the character rectangle is aligned
Exist. On the other hand, no such reference line is found in the Japanese line. Utilizing this difference, the English attribute is given if such a reference line like an English line can be extracted from each line in the text block, and the Japanese attribute is given if a reference line not like English can be extracted. Is the determination process.

The specific processing will be described with reference to the flowchart of FIG.

After resetting various counters in step S801, the following steps S802 to 811 are performed on all lines in the text block.

In step S802, the number of characters in the line is checked. If the number is smaller than the threshold value, the counter u is incremented by one in step S803. If it is equal to or greater than the threshold value, the process proceeds to S804.

In S804, the distribution of the upper end position and the lower end position of the character in the line is obtained. At this time, two types of methods are selectively used.

One method is to obtain a distribution using the coordinates of the upper and lower ends of a character rectangle in a line. In particular,
Numerical sequence {E ₀ ⁱ } = {(y ₀ ^{i, j} −Y ₀ ⁱ ) / () in which the distance from the upper end of each line to the upper end position of the character is normalized by the line height for the characters in line L _i. Y ₁ ⁱ −Y ₀ ⁱ +1) |
j = 1,. . . , N _i }, and a numerical sequence {E ₁ ⁱ } = {(y ₁ ⁱ −Y ₁ ^{i, j} ) / (Y ₁ ) in which the distance from the lower end of the line to the upper end position of the character is similarly normalized by the line height. ⁱ −Y ₀ ⁱ +1) |
j = 1,. . . , N _i }, the distribution function of FIG. 11 is applied. These are referred to as distributions D ₀ ⁱ and D ₁ ⁱ , respectively.

Another method is to form a character rectangle horizontally Ndo.
In this method, the distribution is obtained by using the coordinates of the uppermost end and the lowermost end of the black pixel in each of the divided t, and using a set of numerical values obtained in the same manner as described above.

The character rectangle used in this processing is a rectangle obtained by dividing a line at the break of the vertical projection of black pixels. A rectangle may actually contain more than one character.

At this time, by using this method, the upper and lower ends of each character can be more correctly extracted.

The number of dots to be divided is set to a bit number of 1 byte, for example, a multiple of 8, for speeding up, and a value close to the average width of the character is selected according to the height of the line of interest.

As described above, the selection of the two methods uses the presence or absence of the contact character attribute given in step S613 of FIG. 6 for each text block. FIG. 9 shows step S in FIG.
FIG. 6 is a flowchart showing a portion for obtaining a distribution 804 in more detail. In step S901, if the text block does not have the contact character attribute, the process advances to step S802 to calculate the distribution from the upper and lower coordinates of the character rectangle. If there is a contact character attribute, S9
Proceeding to 03, the distribution is obtained from the coordinates of the uppermost and lowermost ends of each of the bands obtained by dividing the character rectangle for each horizontal Ndot.

Returning to the processing flow of FIG. 8, in S805, each of the distributions D ₀ ⁱ and D ₁ ⁱ is classified into one of the distributions shown in FIG.

The classification in FIG. 16 will be described. Reference positions G ₁ and G ₂ are provided at 0.2 and 0.125, respectively, with respect to the horizontal axis of the distribution function D (FIG. 15), and the number of peaks having a height exceeding a certain value in the distribution, and Depending on whether G ₁ or G _{2 is on} the left or right, 1-4 in FIG.
Classify into either 8 or X.

For example, when there is only one peak, it becomes one of 1, 2, and 6 depending on the position. If the peak there are two, if the right peak is to the left in G ₁ is
It is classified into 3 or 4 depending on which of the heights of the left and right peaks is higher. If the right of the peak is to the right in G ₁ is
It is classified into 6 or 7 according to the height of the left and right peaks.
However, in the presence of peaks at the right of the G _1, if further exists a peak to its right is classified as X.

Here, the thick arrow pointing one peak in each distribution column in FIG. 16 corresponds to the position of the reference line in the row of interest. However, if the classification code is 1
In the case of, the reference line is not at the peak but at the position of 0. The criteria for determining these distributions are stored in the memory 102 in advance, and coding is performed based on the criteria in S805.

In S806, the distance Wb between the upper and lower reference lines determined from each distribution is determined.

[0080] In S807, from the combination of the classification code of the distribution D 0 i, D 1 i, for the line L i, J (Japanese line),
Either E (English line) or M (I do not know) attribute is given. This relationship is shown in FIG. The table of FIG. 17 is stored in the memory 102 in advance.

If the attribute given to the row is M, S80
Then, the counter j for counting the lines determined to be Japanese is incremented by one. If the attribute given to the row is J, the process proceeds to S809, and the counter e for counting the rows determined to be English is incremented by one. When the attribute given to the line is E, the process proceeds to S810, and the counter m for counting the line for which the language type is determined to be unknown is incremented by one.

In step S811, the counter L is compared with the total number of rows. If they do not match, it is determined that all rows have not been processed, and L is incremented by one in step S830, and step S80 is performed.
The process returns to 1 and the processes of S802 to 811 are repeated for the next row. If L matches the total number of lines in S811, S81
Proceed to 2.

At S812, counters e and j are both 0.
, It is determined that the language of the text block cannot be specified, and the process ends without giving the language attribute. Otherwise, the process proceeds to S813.

In step S813, if the counter e is 0 and j is greater than 0, the text block is considered to be Japanese, and the flow advances to S814 to store the Japanese attribute in the memory 102 in association with the text block. To end. Otherwise, the process proceeds to S815.

In S815, the average value of the distances Wb between the reference lines of all the rows to which the E attribute is given is calculated to be W.

In S816, the counter j 'is reset, and the counter e' is given the number e of English attribute lines. The processing of S817 to 820 is performed on all the rows having the J or M attribute.

In step S817, the difference between W and Wb is obtained. If the difference is smaller than the threshold value, the flow advances to step S818 to set the counter e 'to +1. This is a process of adding a current line of interest to a set of English-like lines when the current line of interest has a reference line interval of the same width as a line group of the E attribute already specified.
If the difference is equal to or larger than the threshold value, the process proceeds to S819, where the counter j ′ is incremented by one, and is counted as a line that is not English-like.

If there is a line with the J or M attribute that has not been processed, the process proceeds to S817-820 for the next J or M attribute line.
Repeat. If all have been processed, the process proceeds to S821.

In S821, the number of e ′, that is, the ratio of the English attribute line to the total number of lines is obtained. If this ratio is larger than the threshold, the flow advances to step S823 to give an English attribute to the text block and terminate. Otherwise, the process proceeds to S822.

In S822, the ratio of the number of j's to the total number of rows is determined. If the ratio is larger than the threshold, the process proceeds to S814, and the text block is given a Japanese attribute, and the process ends. Otherwise, the language determination is not possible, and the process ends without giving the language attribute.

As described above, according to the present invention, a sentence written in a text block in a document is a sentence consisting of only English characters which can be coded at a higher recognition rate by the English character recognition unit, or a Japanese character. It is automatically determined that the sentence includes Japanese characters that can be recognized only by the recognition unit. as a result,
In a character recognition device for a document in which two languages, Japanese and English, are mixed on a page basis or within a page, the user can specify two highly accurate recognition units for English and Japanese without specifying each language. Since they can be automatically used depending on the recognition target, labor saving and high recognition accuracy can be obtained at the same time, and the practicality of the character recognition device can be greatly improved.

In the above description, the language discrimination between Japanese and English has been described as an example, but the present invention is not limited to this.
It may be used for discriminating between Japanese and other alphabet languages, for example, French and German, or for discriminating between Chinese and alphabet languages. Also in this case, the recognition unit can be automatically used according to the recognition target without having to specify the language each time, so that labor saving and high recognition accuracy can be obtained at the same time. Practicality can be improved.

In the above description, an example has been described in which the processing of language determination and character recognition is selectively performed on a text block. However, these processing is performed in a table block, a figure block, or an image including characters. It may be performed for blocks or photo blocks.

[0094]

As described above, according to the present invention,
Extracting a plurality of character regions from an image, determining a language type for each of the extracted plurality of character regions, and separately performing a character recognition process suitable for each character region according to the determined language type. Thereby, the operator can visually recognize the language type and can obtain a high recognition rate without giving an instruction, so that a great effect can be obtained in both the operability and the recognition rate.

As described above, according to the present invention, characters are extracted from the image in the character area, and the language type of the character area is determined based on the width of each character.
Since the language type is automatically determined using information having different characteristics depending on the language type such as the character width, the accuracy of the language type determination is improved.

As described above, according to the present invention, the average value of the horizontal width of the character for each line is obtained, and the language type is determined based on comparison with a predetermined threshold value. Since the language type can be determined in consideration of the characteristics of the entire region, the accuracy of determining the language type is improved.

As described above, according to the present invention, as a result of comparing the average value of the width of the character for each line with the threshold, the average value is calculated based on the ratio of the number of lines exceeding the threshold to the total number of lines. By performing the above-described language type determination, the language type can be determined in consideration of the characteristics of the entire character area, so that the accuracy of the language type determination is improved.

As described above, according to the present invention, the mode of the pitch and spacing of characters in a line is obtained from the peak of each distribution, and the difference is used as the average value for each line. Since the language type can be determined in consideration of the characteristics of the entire character area, the accuracy of determining the language type is improved.

As described above, according to the present invention, characters are extracted from an image in the character area, and the language type of the character area is determined based on the number of horizontal lines included in each character. Since the language type is automatically determined using information having different characteristics depending on the language type, such as the character pitch and spacing, the accuracy of the language type determination is improved.

As described above, according to the present invention, the image of the extracted character is divided into two dots in width and scanned vertically, and the number of the horizontal lines is determined based on the result of counting the specific arrangement pattern. , The number of horizontal lines can be accurately determined without being affected by noise or blurring of lines.

As described above, according to the present invention, a line is extracted from the image in the character area, and the language type of the character area is determined based on the reference line of each line, whereby the entire character area is determined. Since the language type can be determined in consideration of the characteristics,
The accuracy of the language type is improved.

As described above, according to the present invention, the reference line of the line is a line representing the upper and lower ends of the characters included in the line, so that the upper and lower ends of the characters are different depending on the language type. Since the language type is determined based on the language type, the accuracy of the automatic determination of the language type is improved.

As described above, according to the present invention, characters are extracted from the image in the character area, the image of each character is divided into a fixed width, and the upper and lower ends of the black pixels detected in each are divided. By determining the reference line of each line from the distribution of, it is possible to deal with contact characters.

As described above, according to the present invention, by providing a plurality of the language type discriminating methods, it is possible to appropriately cope with a case where the shape, layout, and font of a character are different.

As described above, according to the present invention, the line direction of the extracted character area is determined, and if the determined line direction is a specific direction, the language type determination is not performed. 2. The character processing method according to claim 1, wherein the language type is determined in advance.

As described above, according to the present invention, the information indicating the position of the extracted character area is stored in association with the information indicating the determined language type. Character recognition processing suitable for the language type can be easily performed.

As described above, according to the present invention, by outputting the image information and the result of the character recognition by the same output device, it becomes easy to confirm the image of the character recognition target and to confirm the recognition result. .

[Brief description of the drawings]

FIG. 1 is a hardware configuration diagram of an apparatus according to the present invention.

FIG. 2 is a functional block diagram according to the present invention.

FIG. 3 is a flowchart of an overall process according to the present invention.

FIG. 4 is a flowchart showing details of a process in step S305 in FIG. 3;

FIG. 5 is a flowchart showing details of a process in step S304 in FIG. 3;

FIG. 6 is a flowchart showing details of a process in step S510 in FIG. 5;

FIG. 7 is a flowchart showing details of a process in step S511 in FIG. 5;

FIG. 8 is a flowchart showing details of processing in step S513 in FIG. 5;

FIG. 9 is a flowchart showing details of a process in step S613 in FIG. 8;

FIG. 10 is a diagram illustrating region division.

FIG. 11 illustrates a distribution function.

FIG. 12 is a diagram illustrating character horizontal line degree.

FIG. 13 is a diagram illustrating a reference line of a character line.

FIG. 14 is a view for explaining extraction of character end positions in italics

FIG. 15 is a diagram illustrating the distribution of character end positions.

FIG. 16 is a view for explaining the classification of the distribution of character end positions.

FIG. 17 is a diagram for explaining correspondence between upper and lower character end positions and line attributes;

Claims (35)

[Claims] 1. A plurality of character regions are extracted from an image, a language type is determined for each of the plurality of extracted character regions, and a character recognition process suitable for each character region is performed according to the determined language type. A character processing method characterized by processing separately. 2. The character processing method according to claim 1, wherein characters are extracted from the image in the character region, and a language type of the character region is determined based on a width of each character. 3. An average value of the width of the character for each line is obtained,
3. The character processing method according to claim 2, wherein the language type is determined based on a comparison with a predetermined threshold. 4. A method for determining the language type based on a ratio of the number of lines whose average value exceeds a threshold value to the total number of lines as a result of comparison between an average value of the width of the character for each line and a threshold value. The image processing method according to claim 3, wherein: 5. The pitch and spacing of characters in a line,
4. The character processing method according to claim 3, wherein a mode value is obtained from a peak of each distribution, and a difference between the modes is set as an average value for each line. 6. The character processing according to claim 1, wherein a character is extracted from an image in the character area, and a language type of the character area is determined based on the number of horizontal lines included in each character. Method. 7. The method according to claim 6, wherein the image of the extracted character is divided vertically into two dots in width, scanned vertically, and the number of the horizontal lines is determined based on a result of counting a specific arrangement pattern. Character processing method described in. 8. Extracting a line from an image in the character area,
2. The character processing method according to claim 1, wherein a language type of the character area is determined based on a reference line of each line. 9. The character processing method according to claim 8, wherein the reference line of the line is a line representing upper and lower ends of characters included in the line. 10. Extracting a character from an image in the character area, dividing the image of each character into a fixed width, and determining a reference line of each line from a distribution of upper and lower ends of black pixels detected in each character. The character processing method according to claim 9, wherein: 11. The character processing method according to claim 1, further comprising a plurality of language type determination methods. 12. A line direction of the extracted character area is determined, and when the determined line direction is a specific direction,
2. The character processing method according to claim 1, wherein the language type is not determined, but is determined as a predetermined language type. 13. The character processing method according to claim 1, wherein information representing the position of the extracted character area and information representing the determined language type are stored in association with each other. 14. The character processing method according to claim 1, wherein the image information and the result of the character recognition are output by the same output device. 15. The character processing method according to claim 1, wherein one of the determined language types is Japanese. 16. The character processing method according to claim 1, wherein one of the determined language types is English. 17. A character area extracting means for extracting a plurality of character areas from an image; a language type determining means for determining a language type for each of the plurality of character areas extracted by the character area extracting means; According to the language type determined by the means,
A character processing apparatus comprising: character recognition control means for separately processing character recognition processing suitable for each character area. 18. The method according to claim 17, wherein the language type determining unit extracts a character from an image in the character region and determines a language type of the character region based on a width of each character. Character processor. 19. The language type determining means obtains an average value of the horizontal width of the character for each line, and determines the language type based on a comparison with a predetermined threshold value. 19. The character processing device according to 18. 20. The language type discriminating means, as a result of comparing an average value of the horizontal width of the character for each line with a threshold, based on a ratio of the number of lines whose average value exceeds the threshold to the total number of lines. 20. The image processing apparatus according to claim 19, wherein the type is determined. 21. The character processing according to claim 19, wherein a mode value is obtained from a peak of each distribution for a pitch and an interval of characters in a line, and a difference between the modes is an average value for each line. apparatus. 22. The language type determining unit extracts characters from an image in the character region, and determines the language type of the character region based on the number of horizontal lines included in each character. Item 18. The character processing device according to item 17. 23. The method according to claim 22, wherein the image of the extracted character is divided into two dots in horizontal width and vertically scanned, and the number of the horizontal lines is determined based on a result of counting a specific arrangement pattern. 3. The character processing device according to 1. 24. The method according to claim 1, wherein the language type determining unit extracts a line from an image in the character region, and determines a language type of the character region based on a reference line of each line.
8. The character processing device according to 7. 25. The line according to claim 24, wherein the reference line of the line is a line representing upper and lower ends of characters included in the line.
3. The character processing device according to 1. 26. Extracting a character from an image in the character area, dividing the image of each character into a fixed width, and determining a reference line of each line from a distribution of upper and lower ends of black pixels detected in each character. 26. The character processing device according to claim 25, wherein: 27. The character processing apparatus according to claim 17, wherein said language type determining means includes a plurality of language type determining methods. 28. A line direction determining means for determining a line direction of a character area extracted by the character area extracting means; and if the line direction determined by the line direction determining means is a specific direction, 18. The character processing device according to claim 17, wherein a language type is determined without performing language type determination. 29. The character according to claim 17, further comprising storage means for storing information indicating the position of the extracted character area and information indicating the determined language type in association with each other. Processing equipment. 30. An apparatus according to claim 1, further comprising output means for outputting the image information and the result of the character recognition.
8. The character processing device according to 7. 31. The character processing apparatus according to claim 17, wherein one of the determined language types is Japanese. 32. The character processing device according to claim 17, wherein one of the determined language types is English. 33. A character processing apparatus according to claim 30, wherein said output means is an ink jet printer. 34. The character processing device according to claim 17, further comprising a scanner for reading the image information. 35. A control program for extracting a plurality of character regions from an image, a control program for determining a language type for each of the extracted plurality of character regions, A computer-readable storage medium storing a control program for separately performing a character recognition process suitable for a character area.