JPS60164879A - Character separating device - Google Patents

Abstract

PURPOSE:To separate a character row image into a single character by calculating the dispersion of the distance between separated characters, and calculating a dispersed evaluation value consisting of plural dispersion maximum likelihood linear sum to identify a character pitch. CONSTITUTION:A scan device 1 optically scans a character row image described on paper, converts it into an electric signal and quantizes it into binary to write it in a character row image memory 2. A character string extraction device 3 sequentially extracts a character string from said image, and stores start point position and size of each of said strings in a character string register 4. A controller 7 calculates the distance between the character strings from the start and end positions to be transferred, and increments the number of frequencies which are the contents of address position of said corresponding distance of a frequency distribution table 6. In such a way, the device forms the frequency distribution of said distance of the table 6 constituted of a memory.

Description

[Detailed Description of the Invention] (Technical Field) The present invention relates to a character separation device, and particularly to a character separation device that is suitable for separating a character line image written on paper into individual characters and identifying the character pitch. This relates to a separation device.

(Prior Art) When recognizing a series of characters with a device that optically reads various printed characters (hereinafter referred to as OCR), each character is
It is necessary to separate each character and send it to the character recognition unit. The information required to separate each character t-i is the character pitch, but this can be provided in advance if the size and type of printed matter to be read by OCR 9 is limited.

However, in recent years, OCR has begun to read items such as mail and documents with unspecified character pitch, such as floor desks, etc.
・-'Pf- Can't you have the power to know? Therefore, it becomes necessary to estimate the character pitch from the character line image on the paper. Conventionally,
For example, an average value has been used as a method for estimating character pitch. However, when the width of individual characters differs greatly depending on the font or character category, such as in the case of English printed characters, or when the number of characters that touch each other increases, the above average value and the actual character pitch are separated by one character. The errors that sometimes occur cannot be ignored. Therefore, for example, if you use the above average value to separate an entire character line image that includes many characters that are in contact with each other, it is possible that the number of characters in contact may be incorrect or that characters may be cut at incorrect separation positions. There was a problem. Also, under the OCR reading conditions mentioned above,
Usually, not only uniform pitch print data but also variable pitch print data in which the character pitch is variable, and sometimes handwritten characters may also be included. Therefore, for example, if contact occurs between characters in European variable pitch print data, there is a second problem in that it becomes difficult to employ the same character separation process as in uniform pitch print data.

This means that the character separation process is different between a print data set represented by variable pitch European characters and a uniform pitch print data set.

Therefore, it is important to identify the character pitches described above before performing normal character-by-character separation processing. That is,
The character separation device detects the character pitch for separating one character and also determines how much variation the character line image to be separated has with respect to the estimated character pitch. It is also required to detect a variance evaluation value indicating whether the Therefore, if such a dispersion evaluation value is calculated, for example, when using a "character separation device" of an invention filed on December 20, 1982 by the same applicant (hereinafter referred to as the "earlier invention"). , it can be used to determine parameters for separating individual characters by using the characteristics of the character pitch and character pattern image, and furthermore, it can be used to determine the parameters for separating characters into individual characters. Even when there are mixed data sets that are difficult to separate characters based on character pitch, such as handwritten characters written unconsciously, it is possible to identify them using the variance evaluation value described above. It is possible to realize a character separation device that can be applied to general-purpose 6CR character reading objects. (Objective of the Invention) The present invention has been made to solve the above problems.
The purpose is to detect the best character pitch and at the same time calculate the variance evaluation measure that shows how much variation the character line image is printed with respect to the detected character pitch. The goal is to understand the characteristics and provide adaptive character separation processing. Another object of the present invention is to stably detect the best character pitch even under various character reading conditions, such as when there is contact between characters or when one character is separated into two or more characters. The object of the present invention is to provide a possible character separator.

<amo configuration) According to the present invention, in a character separation device that scans a series of character line images and separates them into single characters, means for sequentially detecting a plurality of character chunks from the series of character line images; means for calculating distances between character clusters from a plurality of character clusters and storing a frequency distribution of the distances between character clusters; means for predicting an interval in which a character pitch exists; and means for dividing the frequency distribution into a plurality of regions, means for detecting a candidate character pitch included in the existing interval in which an estimation error evaluation scale constituted by a maximum likelihood linear sum of the character pitch estimation errors in a plurality of regions is the minimum; Calculate the variance of the distances between the character blocks included in the plurality of divided regions, and configure by the maximum likelihood linear sum of the plurality of variances.
A character separation device is obtained, comprising means for calculating a variance evaluation value, and means for identifying a character pitch based on the variance evaluation value and separating the character line image 'f: one character.

(Example) Next, the present invention will be explained with reference to the drawings.

FIG. 1 is a diagram showing an example of a character line image for explaining the distance between six character blocks used in the character pitch detection device of the present invention. In the figure, a separable character image, that is, a block of characters, is shown by a rectangular area shown by a broken line on a white background shown by diagonal lines. Here, reference symbol TJt, t+t (however, i -1, L-6) HafjJ! , i-th.

The distance between the character blocks is defined as the distance from the start of the character block to the start of the i+1th character block, and the reference symbol U'1
-+□ (where ''' 1.a...6) is the distance between character blocks from the end of the first character block to the end of the i+1th character block.

All of these can be used as observed values constituting a frequency distribution of distances between character blocks, which will be described later as distances between character blocks. Similarly, reference symbol U, , (where i=1.2.・
...6, i (j), in the figure U□, 3 and U09.
) is the first character block l = x, z・
a , i < j teashi, in the figure U', 3 (7). Normally, the larger the number of observations, the more statistically stable it is, so the number of observations that make up the frequency distribution of the distance between character clusters, which will be described later. It can be used as

Note that the above-mentioned distance between character blocks, which is an observed value for configuring the frequency distribution, is, for example, the distance between character blocks UI.

j (however, j=i+1, i+2.i+3), distance between character blocks U'1, 3 (j=i+1, i+2,
It is possible to limit the use of up to i+3). Also, for character images such as periods (・), commas (1), etc. (none of which are shown), the width and height of the character blocks are checked using, for example, the average height Hl of multiple character blocks. In particular, it may be removed from the observed amount of distance between character blocks. Furthermore, even if a blank space larger than the above-mentioned average height Ho is detected, the distance between character blocks including the blank space may not be used as the observed amount.

Next, FIG. 2 is a diagram showing an example of the frequency distribution of distances between a series of character blocks in equal pitch data as shown in FIG. 1. In the same figure, the horizontal axis U of the frequency distribution is the distance between character blocks U
, , , and the vertical axis NUM indicates the frequency at a given distance between character blocks.

Next, Figures 3 and 6 are diagrams showing first and second examples of frequency distributions of distances between a series of character blocks in constant pitch data and Roman variable pitch data, respectively. Figure 3 shows constant pitch data, and Figure 3 shows variable pitch data in Roman languages.

Next, the principle of the present invention will be explained using FIGS. 2 and 3. The average height H□ is calculated from the heights of multiple sentence blocks, and the character pitch existence interval (α1·H□) is calculated using the coefficients α□ and α2 (however, α1<−α,).

α2・H, )'t-set. Here, all the distances between character clusters in the existence interval (α, ・HHtα2・H,) described above may be taken as the candidate character pitch P1, but if the process described below is used, the number of candidate character pitches Pl can be It is possible to improve the processing time by reducing the amount of time required. That is, the existence interval (α□・Hm) mentioned above.

The distance U between character clusters that takes the mode within a certain allowable width Δτ (however, Δτ ≥ 1) within α2・H, ) (1) Calculated from the t-frequency distribution, and MAX (9
− α1・Hm, (1−α3)・U(1)) (however, coefficient α3
satisfies O≦α3≦1), the upper limit value of the character pitch existence section MIN (α2・Hm, (1+α3)・U (1)
). In Fig. 2, section C1 is the section where character pitches exist, and the distance between character blocks included in section C1 is the candidate character pitch PI (however, MAX (α1 phrase H,, (
1-α3)・U(1))≦PK≦MIN(α2・HrI
, , (1+α3)−U(1)).

Next, using an arbitrary candidate character pitch P, as a base quantity, the frequency distribution is in the area f'l+f2+... as shown by the broken line in the figure.
It is divided into f p. Here, each area f'k (where = 1
．． 2. -n) boundary point 8 (-1+f'k at f') exists between the region f4-1 center (k-1) P and * ps at the center of the region f', and each Boundary point S in area f'
(fJ, , flya□) is −Pl at the center of the area f(,
and the center of +1 (k+1)·PI in region f'. Therefore, for example, the boundary point S (−1+'' for f'
k) is the center of -1 (k-1) Pl in the area f' and the area f
The center of l, k - P, and the midpoint kf (center of k - P,
and the center of +1 in area f' (k+1) 10- Next, nk (n, ≧0) character blocks existing in area f'k (where = 1.2...n) Calculate the average value P (k, nk) of the distance between
) by dividing the candidate character pitch P by an integer.
An amount corresponding to the character pitch observed from the area f'k when 1 is used as a reference is detected. Therefore, the candidate character pitch P minimizes the deviation between the candidate character pitch P and the candidate character pitch P over all the regions f'k.

It is effective to apply a maximum likelihood linear estimation method in which P is the estimated character pitch. Therefore, for example, in the estimation error evaluation, which is the evaluation standard for performing maximum likelihood estimation, the coefficient C(k,
nk) is the number of samples nk and the integer k (however, -1, 2,
・It is a function of n), and satisfies ""xCk=1 (Jnk)=1. The specific coefficient C(k.

"5c) (D-For example, C(k, nk)-k -n
k/Σ k2·nk can be used. Note that the estimation error evaluation scale T, which is equal to 1 above, is the amount of variance of the estimation error, but the absolute amount of deviation is 1-・P (k + ``k)
An evaluation criterion based on -P+I can also be used. In this way, the optimal estimated character pitch P can be estimated, and the frequency distribution of distances between character blocks can be clustered at the same time.

Next, the character pitch identification method according to the present invention will be explained using FIG. 3 and @. As shown in Figure 3, the frequency distribution of the distance between character blocks is different between uniform pitch print data and variable pitch print data. This difference is between each character block from twice the estimated character pitch P in each region fk (where = 1.2...n) clustered by the optimal character pitch P shown in Figure 2. It can be identified by evaluating the distance variance σ'' (fsc) over all regions f1, f2, .

Therefore, each clustered region fk (−1,
2,...n) can be used. Here, the coefficient C(k, nk) in the value ε2 (the above-mentioned variance amount) can be used. It is a function of the number of samples nk of the distance between clusters and an integer, and
Ck=1 (k, nk)=11 is satisfied. The specific coefficient C(k.

nk) - For example, C(k, nk) - has 2・nk/
) k2·nk can be used. Note that it goes without saying that the error evaluation value 6 can be used instead of the above-mentioned variance evaluation value ε2.

Next, if the variance evaluation value ε2 or the error evaluation value e is larger than the preset threshold α4, character separation based on the above-mentioned estimated character pitch P is possible, that is, equal pitch print data is used. If the threshold value α4 is also smaller than the above-mentioned threshold value α4, it is found that the above-mentioned character separation based mainly on the estimated character pitch cannot be applied, that is, the data is European variable pitch print data, and the nature of the character pitch can be identified. In this way, if it becomes possible to identify the characteristics of the character pitch, if character separation is possible based on the character pitch, for example, the characters can be stably separated using the "character separation device" of the prior invention. Separation can be performed, and other well-known character separation techniques other than those described above can also be used. On the other hand, if character separation based mainly on character pitch cannot be applied, character separation can be performed using, for example, a known character separation technique using blank spaces in character line images that can be easily realized.

FIG. 4 is a logical block diagram showing a specific embodiment of the character separation device of the present invention. A scanning device 1 optically scans a character line image written on a paper surface, converts it into an electrical signal, and writes it into a character line image memory 2 after binary quantization. The character block extraction device 3 sequentially extracts character blocks from a plurality of character line images stored in the character line image memory 2, and stores the starting position and size of each character block in the character block register 4. Note that the size of a character block represents the width and height of the character block. Further, such a character block extraction device is disclosed in, for example, Japanese Patent Application No. 56-2751 filed by the same applicant.
The technique shown in the specification of No. 2 can be used.

The control device 7 sequentially calculates the distance between character blocks from the start and end positions of the character blocks transferred from the character block register 4, and calculates the address position of the corresponding distance between character blocks in the frequency section 14-cloth table 6. Increment the frequency number that is the content.

In this way, a frequency distribution of distances between character blocks as shown in FIG. 2 is generated in the frequency distribution table 6 constructed from memory. It is assumed that the frequency distribution table 6 is initially initialized to O. Next, the average height Hl is calculated by the control device 7 from the heights of the plurality of character blocks stored in the character block register 4, and is transferred to the existing section detection unit 50. The constant register 19 stores the coefficient α□ shown in FIG.
The constants α2 (however, α□×α2), α3, and a constant allowable width Δτ (however, Δτ≧1) are stored. Existence section detection unit 50
First inputs the coefficients α□ and α2 from the constant register 19, and sets α□·Hm and α2·Ho as the lower limit value α1·Hm and upper limit value α2·Ho of the character pitch existing interval. Next, the existence section detection unit 50 sequentially reads the frequency values of the distances between character blocks that belong to the lower limit α1·Ho and the upper limit α2·H of the existence interval from the frequency distribution table 6 via the control device 7, Distance between character blocks U (1) that has the most frequent value with the constant allowable width Δτ (however, Δτ≧1) transferred from the constant register 19? calculate. Existence interval calculation unit 51 uses the distance between character blocks U(1) determined by existence interval detection unit 50 and coefficient α3 input from constant register 19 to calculate the values (1−α3)・U(1) and ( 1+α
, )・U(1), MAX(α,・Hl,l
l, (1-α3)・U(1) )? The lower limit value PL,MIN(α2・H,n,(
1+α3)·U(1)) are respectively stored in the existence interval register 8 as the upper limit pU of the character pitch existence interval. Note that when the above-mentioned distance between character blocks U(1) is not detected by the existing area detection unit 50, the average height Hl is set as the distance between character blocks U(1) in the existing area detection unit 50. shall be taken as a thing. Next, the lower limit value PL of the character pitch stored in the existence interval register 8 is set in the counter 14 by the control device 7. The counter 14 counts up after the operations described below are completed sequentially from the lower limit value PL to the upper limit value PU, and transfers the count value Pl (PL≦P,≦Ptr) to the average value calculation unit 9. The average value calculation unit 9 calculates the count value transferred from the counter 14, that is, the candidate character pitch p, 'tl-
n area ranges of base units, base quantities, and frequency distributions, that is, area f
k (However, by referring to the frequency distribution table 6, the number nk of distances between character blocks belonging to the area fk and the 1,000-degree value P(k, nk) of the distances between character blocks belonging to the area fk are calculated.The above processing is carried out over n areas, and the candidate character pitch P and the number of distances between character clusters nk (however, *=1.z... n) and the average value P(k, nk), (where = 1.2.·n) are transferred to the estimated error evaluation value calculation unit 10.The estimated error evaluation value calculation unit 10 is The value Tf is determined by calculation using the information transferred from the average value calculation unit 9. The minimum estimated error evaluation value T is stored in the evaluation value register 12. The maximum likelihood character pitch register 15 17- It is assumed that a very large value is set as the preliminary estimation error evaluation value.In the comparator 11, the estimation error evaluation value T output from the estimation error evaluation value calculation unit 10 and the evaluation value register 12, and if the output value of the estimated error evaluation value calculation unit 10 is smaller than the content of the evaluation value register 12, the output value of the estimated error evaluation value calculation unit 10 is written to the evaluation value register 12, and ! The candidate character pitch PIO value is written to the maximum likelihood character pitch register 15, and the control device 7'f
: The counter 14 increments by one. on the other hand,
In the comparison section 11, if the output value of the estimated error evaluation value calculation section 10 is larger than the content of the evaluation value register 12, the control device 7t is turned off and the counter 14 is counted up by one.

The above operation is performed until the value of the counter 14 is the upper limit value P of the character pitch.
By repeating until reaching U, the optimum character pitch P
is stored in the maximum likelihood character pitch register 15. Variance evaluation value calculation unit 13, maximum likelihood character pitch register 1
When the estimated character pitch P that is most suitable for the number is set, the estimated character pitch is read from the maximum likelihood character pitch register 15 or 18-, and the estimated character pitch P? By referring to n area ranges of the frequency distribution as a base quantity, that is, area fk (where = 1° Table 6), the number of samples nk of the distance between character blocks belonging to area fk and the distance between character blocks belonging to area f can be determined by referring to table 6. The variance σ2(fk) at −P is calculated for the value of , and the variance σ3(fk) detected in the table variance evaluation value 62=dispersion evaluation value calculation unit 13 as described above is calculated in the average value calculation unit 9. The variance evaluation value C2 calculated by the variance evaluation value calculation unit 13 is stored in the variance evaluation value register 16.The threshold value register 18 stores a threshold value α4t-storage for identifying the character pitch. The comparison unit 17 identifies the nature of the character pitch by comparing the variance evaluation value ε2 stored in the variance evaluation value register 16 with the threshold value α4 stored in the threshold value register 18. That is, the character pitch characteristic is identified by comparing the variance evaluation value ε2 stored in the variance evaluation value register 16 with the threshold value α4 stored in the threshold value register 18. If the value ε2 is larger than the threshold α4, it is determined that the character separation position determining method based mainly on the estimated character pitch P cannot be applied, and if the variance evaluation value ε2 is smaller than the threshold α4 by υ, the method for determining character separation positions based on the estimated character pitch P is determined as the main method. It is determined that the character separation position determination method described above is applicable.

Reference numeral 20 is a one-character separation position determination unit, and the known techniques as described above or a combination thereof can be used. In other words, it is possible to use, for example, the "character separation device" of the invention of the prior application and character separation using blank spaces, which can be realized most easily.

That is, the above-mentioned variance evaluation value ε3 in the comparing section 17
When it is determined that character separation based on the estimated character pitch P is applied, the control device 7 uses the estimated character pitch P stored in the maximum likelihood character pitch register 15 and the variance stored in the variance evaluation value register 16. Evaluation value ε! is transferred to the one-character separation position determination unit 20, which is configured, for example, as the "character separation device" of the prior invention. On the other hand, if the comparison unit 17 determines that the character separation based on the estimated character pitch Pt cannot be applied, it considers the plurality of character block images stored in the character block register 4 as one character, and
This is the output value of the character separation position determining section 20. Note that it goes without saying that the single character separation position determining section 20 can be constructed using a known technique other than the above-described embodiments.

(Effects of the Invention) As described above, according to the character separation position of the present invention, even if the character pitch is not known in advance or even if a character image including contact of character blocks or separation between characters is included,
The effect is that it is possible to accurately measure character pitch, grasp the characteristics of character pitch, and easily realize a highly adaptable character separation device.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing an example of a character line image of the distance between character blocks used in the character separation device of the present invention, FIG. 2 is a diagram showing an example of the frequency distribution of the distance between a series of character blocks in equal pitch data, In Figure 3 (g), @ indicates the first frequency distribution of a series of distances between character blocks in constant pitch and variable pitch data. The second example and FIG. 4 are logical block diagrams showing a specific embodiment of the character separation device of the present invention. In the figure, 1... scanning device, 2......
Character line image memory, 3...Character block extraction device, 4...Character block register, 50...Existence section detection unit, 51...Existence section calculation Department, 6・
... Frequency distribution table, 7 ... Control device, 8 ... Existence interval register, 9 ... Average value calculation unit, 10 ... Estimation error evaluation value calculation unit,
11.17...Comparison unit, 12...Evaluation value register, 13...Dispersion evaluation value calculation unit, 14
... Counter, 15 ... Maximum likelihood character pitch register, 16 ... Variance evaluation value register, 1
8...Threshold value register, 19...Constant register, 20...1 character separation position determination unit. 22-

Claims (1)

[Claims] A character separation device that scans a series of character line images and separates them into single characters, comprising means for sequentially detecting a plurality of character blocks from the series of character line images, and calculating distances between character blocks from the plurality of character blocks. means for storing the frequency distribution of the distance between character blocks; means for predicting the existing interval of character pitch; and means for dividing the frequency distribution into a plurality of regions and calculating the maximum of the character pitch estimation error in the plurality of regions. means for detecting a candidate character pitch included in the existing interval in which an estimated error evaluation measure constituted by a likelihood linear sum is minimized; means for calculating the variance of the distance between the character blocks and calculating a variance evaluation value constituted by the maximum likelihood linear sum of a plurality of said variances; and means for identifying the character pitch based on the variance evaluation value and generating the character line image. A character separating device comprising means for separating into one character.