JPS60132281A - Character separating device - Google Patents

Abstract

PURPOSE:To determine easily and with a high accuracy a character separating candidate example position by calculating a sequence of the character separating candidate example position for minimizing a dispersion of a distance between each separating proposed example position, between each separating candidate example section, and a dispersion regarding a shift quantity of an average distance and a characer pitch, by using a dynamic planning method. CONSTITUTION:An allowable section which can set a character separating position is set by using a character pitch P and threshold levels T1, T2. Subsequently, by using the character pitch P and a threshold level T3, a separating candidate example section (k) (provided that k>=0) is set successively, and with respect to a separating candidate example position (x) (k, ik) (provided that ik>=1) in each separating candidate example section (k), a distance (d) (k, k+1; ik, kk+1) between each separating candidate example position is calculated. Subsequently, a dispersion sigma<2>d of the distance (d) (k, k+1; ik, ik+1) calculated in each separating candidate example section (k), and an evaluating measure U consisting of a double of a double error (mud-P)<2> of a shift of an average value mud and the character pitch P are calculated. Next, a sequence of a separating candidate example position (x) (k,ik) (k>=0) for minimizing its evaluating measure is derived, by which a character separation determining position is determined.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a character separation device that separates a character string image written on paper into individual characters.

When recognizing a series of characters in a device that optically reads various groups of printed characters (hereinafter referred to as 0C), it is necessary to separate each character one by one and send it to a character recognition unit. By the way, when reading mail or a large amount of documents using an OC machine, there are many different qualities and printing styles of printed matter, and it is necessary to handle them as objects to be read.

In such a case, characters may come into contact with each other in the character string image on the paper, or a % 1 character may be separated into two or more characters, and a character separation method that can efficiently handle these situations is required. Conventionally, when input data that has relaxed constraints on the reading target is included, character separation methods have been applied that add functions that are considered effective in each case. However, adding functions corresponding to individual cases as appropriate may reduce the accuracy of character separation, or may require the development of character separation devices with different functions for each individual object. arise.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to sequentially set separation candidate sections using character pitch and blank information, and to determine the distance between each separation candidate position between each separation candidate section in order to solve the above conventional problems. By using dynamic programming to calculate a series of character separation candidate positions that minimizes the variance of the variance and the variance of the deviation between the average distance and the character pitch, it is possible to solve cases where an image of contact between characters occurs. To provide a character separation device capable of easily and accurately determining a character separation candidate position without adding any special function even if one character is separated into two or more characters.

Hereinafter, the present invention will be explained in detail using the drawings.

FIGS. 1 fa), fb), and (C) are diagrams for explaining an example of a separation candidate section setting method for character separation in the present invention using an example of a partial character string image.

Note that when performing character separation in a reading target where there are various print qualities and printing styles, it is necessary to first detect the character pitch. As a character pitch detecting device, for example, there is a character pitch detecting device as shown in Japanese Patent Application No. 160763 (1982) filed by the same applicant (hereinafter referred to as Cited Example 1). Character pitch can be detected from a series of character line images using a conventional device. - Figure 1 (aj shows the character string image including the contact image between characters and the case where one character is split into two characters with diagonal lines, and P in the figure shows the character pitch.
) is vertically projected, the partial character-sequence image in Figure 1 (
The projection distribution shown in b) is obtained, and can be separated into a black background area (shaded area in the figure) and a white background area. The character pitch P may be determined from the character string image shown in FIG. A value may also be used. In order to correctly separate the touching character (amu) and the split character (hl) shown in FIG. 1 fa) into one character position, it is necessary to correctly predict the separation start position of the character. For example, in the figure, in the touching character images 'a' and 'm', a is slightly smaller, so its extraction start position is slightly to the left of the starting edge of the touching character image (amu). Conventionally, corrections for misalignment of touching characters have been carried out, for example, by referring to the character images.As a result, depending on the degree of contact between the character images, etc., it takes processing time to refer to the images. , it may not be possible to determine the separation position correctly.Therefore, the method for determining the character separation position in accordance with the present invention begins with the character line image shown in Fig. 1(a). A white background area (hereinafter referred to as a blank space) and a black background area (hereinafter referred to as a blank area) extracted from the projection distribution etc. shown in b)
The position and size of the character block (hereinafter referred to as a character block) are extracted. For example, the character block width Vi (i==
1. ...4) and blank size Wi (i = l, ・
...4)1 Furthermore, their positional information can be obtained using known techniques. Next, the character pitch P obtained in advance
and thresholds T, , T, to set an interval (hereinafter referred to as an allowable interval) in which a character separation position can be set using, for example, the following conditions (1) and (2). be able to.

Condition (1) A blank section is a permissible section.

Condition (2) At any character block width Vj, Vi>P+T
.

T2 from both ends of one character block width Vi in the black area that satisfies
The section excluding the black area up to is the permissible section.

For example, the permissible intervals that satisfy the above-mentioned conditions (1) and (2) are At , At , As , A4 in Fig. 1 (C) for the character string image in Fig. 1 (al).
．． This is obtained as an interval indicated by As. Note that the threshold values T and T mentioned above may be given as functions of the character pitch P. Further, the threshold T1 may be set based on the estimation error when estimating the character pitch P shown in Cited Example 1 above. Next, character separation positions in the present invention are determined using the character pitch P and threshold Tj within the above-mentioned allowable interval, and sequentially determines the separation candidate interval k (-
However, k≧0) is set, and for a separation candidate position X (k, 1t) within each separation candidate section (however, jk≧1, representing a relative number within the separation candidate section), The distance between each separation candidate position d(k, +1; ik, jk++
), and the distance d(k, +1; jk) calculated in each separation candidate section k.

jh+t) variance σ5 and average value μd and character pitch P
The evaluation scale consisting of the squared error (μd-P)2 of the deviation from
This is done by calculating the separation candidate position x(k, 1k) (k≧0) that minimizes the evaluation scale. Therefore, an example of a method for setting the above-mentioned separation candidate sections will be explained using FIG. 1(C).

The black dots in FIG. 1(C) indicate separation candidate section k (however,
k: 0, 1, 2, 3.4) for each separation candidate position [x(k.

jh), and are sequentially set as positions that satisfy the relationship of equation (3) shown below within the above-mentioned allowable interval.

1d (k, +1; ik, jk+1) P I≦T3
...(1) In equation (1), distance d (k, k
+ 1; "k+"k+1) is the distance x(k+1, ih+x) between the separation candidate position x (k+1, ih+t) of +1 in the separation candidate section and the separation candidate position X (k-"i+) in the separation candidate section x(k': ih).For example, the distance d between the separation candidate position X(1,1) and the separation candidate position X(2,2) in FIG. 1(C)
(1,2; 1,2 ), IP-d(1,2;
1, 2) 1≦P −4−T3 is satisfied. In addition, from the separation candidate position x (OJ) of separation candidate section 0, equation (1
) There are six positions that satisfy the relational expression shown in Figure 1 (C), as shown by the white dots in Figure 1 (C). As shown by the black dots,
The separation candidate position of the separation candidate section l is X(1,
1) and x(1,2).

Note that the threshold value T3 described above can be given as a function of the character pitch P, like the threshold value T1 described above. Next, when determining the optimal separation candidate position using the evaluation scale U, the separation candidate section that becomes the terminal state may be set only within the blank space that is the end of the character line, or the above-mentioned blank space When size Wi is character pitch P, Wi
> T4·P (however, T4 is a threshold value) may be set within an allowable interval. If the latter method is adopted, for example, region E in FIG. 1(C) is detected as a permissible section in which the end state is reached. Furthermore, in calculating the evaluation scale U, the separation candidate interval (k+
1) is set using the above-mentioned formula (1) using the separation candidate section, the separation candidate section (k+1) that satisfies the formula (1)
are not all allowable intervals (in other words, the separation candidate section k can be set as the separation candidate section that becomes the terminal state. On the other hand, the separation candidate section k that becomes the starting state is used to find the optimal separation candidate position using the evaluation scale U). The candidate section can be set based on the position of the separation candidate section that has already been detected as the terminal state.

Note that the separation candidate section setting method for character separation according to the present invention can be applied to one character pitch P even under conditions other than the above-mentioned conditions.

Needless to say, it can be set based on the character block width ■i and the in size Wi.

FIG. 2 is a diagram for explaining the principle for extracting optimal character separation positions in the present invention.

In the figure, the positions indicated by black dots indicate the values of each separation candidate position x(k, ik) of the separation candidate section k (lc=Q, . . . 4) shown in FIG. 1 fc). Further, the case where the character pitch P is 20 is shown. Furthermore, in order to simply explain the principle of the present invention, the character string images to be separated are from separation candidate section 0 to separation candidate section 4 in J (C) of FIG. 1. First, let's define the symbols. Symbol μd(r*ntir*'n) (where l≦r; (n)
is 11-r+1 arbitrarily selected in each separation candidate section from separation candidate position x (r, i,) of separation candidate section r to separation candidate position x (n, 1fi) of separation candidate section n.
separation candidate positions x(r, 1r)eX(r+1.'r
+x), -e x(n, in) n −
r distances d(' * ' + 1; 1, @ir
+1)d(r+1.r+2:ir+t*jr+z)・
-, d(n-1, n; in 1°1n). Symbol σd(rs”:ir+i+) (1≦r≦n
) is arbitrarily selected n − in each separation candidate section from separation candidate position x (r, i,) of separation candidate section r to separation candidate position x (n, 1fl) of separation candidate section n. r-
1-11 Same separation candidate position X(reir)ex(r+1
．． ir+x), ・−, n obtained from x(n, i, ,)
−r distances wjd (r t r + 1 r ir
-r+1) *d (r+1, r+2: j
The average value μd('o
n+'r+in). Therefore, character separation is performed from the separation candidate interval r (r = 0 in Figure 2), which is the start state, to the separation candidate interval n (n = 4 in Figure 2), which is the terminal state. The position is given by formula (2)
Separation candidate position x (
r+ ir)+x(r+1+ ir+1)*...+x
It is obtained by subtracting (n+j11).

Ll (r, n) = βoff: (r, n; i, 1n)
-1-(1-β) ”(μd(r,n; Pass 4.)p)z
(2) The weighting coefficient β in equation (2) satisfies 0≦β≦1.

Here, a more specific method for realizing separation candidate positions that minimizes Equation (2) can be performed without consuming memory capacity using dynamic programming described below. Therefore, any separation candidate position x at +1 in the separation candidate section
In (kr1, ik+t), any separation candidate position X(k, 1k)(
However, from ik = 1.2,..., hk, and h≧1), the separation candidate position x (k
+ 1 + 'k) The optimal separation candidate position x(k
, il) can be calculated using the recurrence formula described later.

First, distance d(k, to +cik, jk+t) (however, i
, = 1.2...hk), the following formula (3-
1) Calculate Equation (3-2) and Equation (3-3).

μd (0, +1; io, ik+1) = ” (k-J
(0, to: jo, jk) krl -4-d (k, to +cjk, jk+i) )...(3
-1) D(kr1)=D*(kl+d2(k, k
r1: ik, ! +x)...(3-2)U(o
, krt ) = β・(Kabuto Shitosu − μ6 (0, k
+ 1 r 'o + ``k)) 2 + 1 + (1 - β) (μd (0, + 1.io, lh) p)
t...(3-3) hk separation candidate positions x(k, l) in the separation candidate section.

...For x(k, hk), find the separation candidate position x(
k, ik) becomes the optimal separation candidate position for an arbitrary separation candidate position x (kr1, ik+x) in the separation candidate section k-1-1.

Here, for each separation candidate position X (0, 1O) in the separation candidate section O which is the starting end state (1o=l in Fig. 2), the optimum average value μd shown in equation (3-1) is calculated. (
0,0,1o,1o)=Q, the squared cumulative sum D of the optimal distance d (1,O;'-1-iO) shown in equation (3-2)
*(0)-0. The above-mentioned optimal average value/4(0+
kr' +io) and the optimal distance d(k-1,
; jk−s+jk) cumulative sum D”(k−1, k
), the optimal separation candidate position X(k, jh) for the separation candidate section at each separation candidate position x (k-1-1, ik+1) of +1 can be found in the next separation candidate section. Note that the first term in equation (3-3) is another way of expressing the variance σM (0, +1; io, jh+t) shown in equation (2).

Next, the calculation process of equations (3-1) and (3-3) will be explained using diagram @2. In the figure, the values shown in parentheses are for each separation candidate section k (k=Q, 1.2, 3.
4) at each separation candidate position x(k, ik), the formula (
3-1) and the average value μ shown by the recurrence formula of formula (3-3)
h O, k * io , 1k) and evaluation scale U (
0, k), and the separation candidate position x(k-1e”k
-i) It is calculated as the optimal value from . In addition, in this explanation, the weighting coefficient β in equation (3-3) is 0.
．． The case where it is set to 5 will be described. Further, each arrow in the figure indicates a series of optimal separation candidate positions. For example, the separation candidate position X (2, 1) is located at position 39, and the distance from the separation candidate position
1) becomes 19.

Therefore, the average value μd(o, 2) at the separation candidate position X(2, 1) passing through the separation candidate position
-1) and the figure, it becomes -.(1X20+19), and the value is 195. Next, at the separation candidate position X (1, 1),
Equation (3-2) D"(1)=20' is stored (not shown in the figure), so using Equation (3-2), D(2)=20"+192 becomes. Therefore,
Separation candidate position x(2,
1) Oker rating scale iU (0,2) haU(o, 2-
)-o, 5 (appended 5v-19,5") + 0.5・(
19,5-20) 2, resulting in a value of 0138. Similarly, separation candidate position X(
The evaluation scale U (0°2) in 2.1) is also calculated (however, the calculation is omitted) and has a value of 1.26. Therefore, two evaluation scales U(0
, 2), the optimal separation candidate position in the previous separation candidate section 1 is x (1, 1),
Also, the average value μtc 0.2 )=19.5 and the rating scale iU (0,2)−0,38 are selected. Below, similar operations are performed using equation (3-1).

Using the recurrence formulas shown in equations (3-2) and (3-3),
By doing as shown in FIG.

Each separation candidate position X (J'h) (k=Or 1.2,3
．． 4) evaluation scale U(0,k) (where k =
0.1.2.3°4) is calculated.

Next, as described above, the separation candidate position X (4, 2) * X (43) ex (
4, 4), the separation candidate position x(4, 2) with the minimum evaluation scale [J(0, 4) is selected as the end position of character separation. Therefore, by tracing the series of optimal separation candidate positions backwards from the character separation end position x (4, 2), x (4, 2) = 81, x (3, 3) = 60, etc.
X(2,2)=20. X(1,1)=20. X(0゜1
) can be taken as 20.

FIG. 3 is a logical block diagram showing a specific embodiment of the present invention. A scanning unit 1 optically scans a character string image written on a paper surface, converts it into an electrical signal, and writes it into a character string image memory 2 after binary quantization. The character block extraction unit 3 sequentially extracts character blocks from the character string image stored in the character string image memo 1J2j, and stores the position, width, and height of each character block in the character block information register 21. Incidentally, such a character block extraction section 3 can be extracted using a known technique. The character pitch detection unit 4 detects the position and character block width of each character block stored in the 1-character block information register 21,
Furthermore, the character pitch P is estimated using the character height and stored in the character pitch information register 22. Incidentally, such a character pitch detection unit 4 is similar to the above-mentioned Cited Example 1 by the same applicant.
This can be done by using the technique shown in the specification of 2007, or if the character pitch P is known in advance, a given character pitch P can be used. The parameter information register 30 stores parameters T11'r, l rl, e'r, I which are various threshold values and weighting coefficients used in the present invention.
Store T5+β. The permissible interval extraction unit 5 extracts a permissible interval that satisfies the conditions (1) and (2) described using FIG. First, the allowable interval for blanks shown in condition (1) is determined by using the positions of a plurality of character blocks and the width of character blocks Vi stored in the character block information register 21. Extracted by a comparison circuit or the like. Next, in the permissible interval within the black background area shown in condition (2), first, each character block width Vi is determined by the character pitch P stored in the character pitch information register 22 and the parameter stored in the parameter information register 30. Sum P+'l'' with T1,
Compare whether it is larger or not, and if it is larger, each character block width V
Excluding both ends of i up to the value indicated by the parameter information register T2, the section including the character block width Vi is extracted as a permissible section. Condition (1) extracted as above
Blank allowable sections that satisfy condition (2) and allowable sections within the black background area that satisfy condition (2) are extracted, and the position and width of each extracted allowable section are stored in the allowable section information register 23. The terminal candidate section extracting unit 6 detects a blank allowable section Wif among the allowable sections sequentially stored in the allowable section register 23 corresponding to a single character line image, and extracts the parameters T4 and 2 stored in the parameter information register 30. By calculating the product T4・P with the character pitch P and comparing the product T4・P with the acceptable interval quality for blanks, M
A permissible interval Wi that is larger than P is detected. Next, from the start of the allowable interval Wi, the parameter T and the character pitch P
The product T, ・P (where T, ≦T, the permissible interval 1Lu is output, and the sum P of the character pitch P and the parameter TI from the starting end of the character block width Vi that exists immediately before the permissible interval Wi -1-T, is calculated, and the logical sum of the above-mentioned two permissible intervals is sequentially stored in the terminal candidate section register U as the terminal candidate section.

FIGS. 4(a) and 4(b) show an example of a terminal candidate section extracted by the terminal 1 bacteria complementary section extraction unit 61. In the case of FIG. 4(a), the terminal section is defined as the section indicated by T3.P in the figure. In the case of Fig. 4(b), the terminal section is the section indicated by the last arrow in the figure, where T, ・P and P+TI
This is the allowable interval that is blank among the logical sums.

The separation candidate section extraction unit 7 extracts each separation candidate position x(k,il) in the separation candidate section as explained using FIG.
) are sequentially extracted using the permissible intervals and parameters stored in the permissible interval information register 23 and parameter information register 301. In addition, each separation candidate position x (0, 1o) of the starting end separation gorge supplementary section O including the character separation start position (however,
io=1.2・"ho) is first determined by the control unit 10 from each separation candidate position within a certain blank tolerance interval set based on the character pitch P from the starting end of the character string image. , optimal separation position information register 2
6. Therefore, the 1-separation candidate section extraction unit 7 extracts the separation candidate section k (where k=O,
1, 2...) separation candidate position x (k, jk) (however,
separation candidate position x (kll h ih+
v) is calculated.

That is, the first separation candidate position x(k
, 1) to character pitch P and parameter information register 30
x(k, 1)+P using the parameter T stored in
Calculate the position of T3, and then use the character pitch P and parameter T to calculate the position of b x (k * tlk) + P + T from the last separation candidate position x (k, hk) in the separation candidate section. do. Above two positions X(kll)+P
-T, , X(k, hk)+P-1-T, among the separation candidate positions in the interval obtained by Each separation candidate position x (kll, jh+z
) (where i*+x = L 2- "kll) and is stored in the separation candidate position information register 25.
When the contents of are input, the optimum separation position information register 26
, each separation candidate position x (o, to) (where i,
= 1...-ho ), x(1,1s) (however, 1
I=1...hl),...x(k, 1k) (however, 1
k=1...hk) are stored. Furthermore, each separation candidate position x (m
, im) (where i□=1...salt), the evaluation scale calculation unit 8 calculates the average value μ (0, m; '(le i ,, ), formula (3-2
), the cumulative sum D*(k) of the squares of the distances calculated from the formula (
3-3) and the optimum separation candidate position x(m-
1°'ny-1) is stored. Note that the control unit IO controls each separation candidate position x(0, io
) is stored, the average value μT (o, o; to, io) is stored corresponding to each separation candidate position x (o, io).
It is assumed that 0 is stored in the cumulative sum D*(0) of the squares of the distances. Therefore, the evaluation scale calculating unit 8 selects the separation candidate section initially stored in the optimum separation position information register 26 at the separation candidate position x (k + 1 #'+c) sequentially transferred from the separation candidate position information register 25. each separation candidate position x(k, 1k) (where, jh =
1.2-hk) distance d(k, +1:i,
, ik+t).

Furthermore, the average value μdcOgksi0.1k) s distance 2
Using the next prefecture product sum D*(k) and the parameter β stored in the parameter information register 30, the formula (3-1) is sequentially calculated.
Recurrence formula μ, 1 (0, +1; io, jk+x) shown in
=” (+1 to k+kll+ik*’),
Recurrence formula LJ (0, +1) shown in formula (3-3) = I・
4-μ'a(+1 to 0゜kll; io e jh+t) )"+ (1-
β)・(μd(0, kll);'o, jk+z)
By calculating P)2, the separation candidate position X(k, 1k) in the immediately preceding separation candidate section (where ik = 1-
hk) Calculate the rating scale 111 (0, +1) for +c.

Next, among the h separation candidate positions x(k, jh),
Separation candidate position x where the evaluation scale U (0, +1) is the minimum
(k, ik) as separation candidate position x(kll, jk+t
) is the optimal separation candidate position x(k, i
k), and furthermore, the minimum value of the evaluation scale [J (0, +1) and the average value μ kichi (0 + k+' * 'o m ik ) at which the evaluation scale tJ (0, +1) is the minimum value,
and the cumulative sum of squared distances D*(kl1) are stored in the optimum separation position information register 26 together with the separation candidate position x (k + I * ik).

When the evaluation scale calculation unit 8 performs the above-mentioned calculation processing on all the separation candidate positions x (kl1, ik++) sequentially transferred from the separation candidate position information register 25, the control unit 10 The extraction unit 6 extracts each separation candidate position x (kl2
, ih+z), and the similar operations described above are repeated. Here, the control unit 10 uses the evaluation scale calculation unit 8 to control the optimal separation position information register 2.
+1 separation candidate position x in the separation candidate section transferred 6i times
It is checked whether (kl1, jk+t) has reached the terminal candidate section stored in the terminal section information register 24, and if it has not arrived, only the above-mentioned request is output to the separation candidate section extraction section 6. On the other hand, separation candidate position x (k
l1, +t) reaches the terminal candidate section, the control section 10 transmits the above-mentioned request to the separation candidate section extraction section 6.
and output each separation candidate position x of +2 to the next separation candidate section.
(kl2, ik+2) (however, ik+z= 1
-hk+z) is evaluated by the evaluation scale calculation section 8 as described above, and then the control section 10 selects the separation candidate position Xck', ik stored in the optimum separation position information register 26.
), the section in which the separation candidate position x (n, in) with the minimum evaluation scale U (0, 1) of the plurality of separation candidate positions x (n, jl) in the above-mentioned terminal candidate section is evaluated. A series of optimal separation candidate positions that are detected as the end point positions in Separation candidate position X (n l , 1n-1) +... x (0, io
) and is stored in the character separation position register 27.

The next control unit 10 controls the character pitch Pi within a certain range that is set within the blank tolerance interval from the end point position x (n, x) to the start of the first detected character block. is stored in the optimal separation position information register 26 as the separation candidate section O, which is the starting point of the partial character string image to be separated next, and commanded to perform the operations described above.

In this way, the character separation position of the character string image stored in the character string image memory 2 is stored in the character separation position register 27, and the height and character of each character block stored in the above-mentioned character block information register 21 are stored. Separation position register 27
By using the character separation positions stored in , it is possible to separate each character.

FIG. 5 is a logical block diagram showing a specific embodiment of the evaluation scale calculating section 8 in FIG. 3.

As mentioned above, each separation candidate position X (kl1゜jk+t
) (However, when tk+t = 1...hk+1) is stored, the control unit IO shown in FIG. +1 is transferred to the predetermined position of the separation candidate section in the stage register 80.
and stored in a predetermined position of the separation candidate position group register 261.

The separation candidate position x (kl1 * ik+
v ) is stored, the control unit 10 selects each separation candidate position X(k, 1k) (where ik=1−·
・hk) are sequentially transferred to the distance calculation unit 81. The optimal separation position information register 26 shown in FIG. 3 is composed of a separation group 264. The distance calculation unit 8 calculates the separation candidate position x
(+1 to kl 1 e') and the separation candidate position that is sequentially transferred d(k, kl1; ik+ ik+x
)=X(kl1+ik+t)-x(k, ik) is calculated.

The statistics calculation unit 82 calculates the above-mentioned equation (3-1) and equation (3).
-2), the average value μd(0,
+1; to, ih + t) and distance 2 prefecture product sum D(k
Calculate l. That is, the average value μd(01kl1110
.

jh+1) is the average value μ (0, ni+io+ik) at the separation candidate position x(k, ik) stored in the read optimal statistical group register 263, and the output d of the distance calculation unit 81
(k, kl1; ih*lk+x), using +1 and k for the separation interval that is the content of the stage register 8o, is calculated by the calculation formula %). On the other hand, the cumulative sum of squared distances D(k+1
) is the sum of the two-prefecture product D*(k) of the distance at the separation candidate position x(k, jh) stored in the read optimal statistical group register 263 and the output d(k,
1; jk, fh+t), it is calculated by the calculation formula % formula %). The average value μd (0, +1; jo, ih+1) calculated by the statistics calculation unit 82 and the sum of products of two prefectures D (k+1) of the distance are stored in the statistics storage register 83, respectively. The evaluation value calculation unit 84 calculates the value of the evaluation scale U (0, +1) based on the above-mentioned formula (3-2). That is, the evaluation value [1(0, k+1) is
Using the character pitch P shown in FIG. 3, the parameter β stored in the parameter information register 301, the contents of the statistics storage register 83, and the contents of the stage register 80, it is calculated by the calculation formula %)) . Next, in the comparison section 85, the evaluation value which is the output of the evaluation value calculation section and the minimum evaluation value register 86
If the output of the evaluation value calculation unit 84 is smaller than the contents of the minimum evaluation value register 86, the output signal 851S of the output signal line 851 is turned ON. Assume that the contents of are initially set to a very large value.

When the output signal 851S becomes 'ON', the gate circuit 53
is opened, and the output of the evaluation value calculation unit 84 is transferred to the minimum evaluation value register 86.

Furthermore, when the output signal 851S becomes ``ON'', the average value μd (0*k+
1: jo m jk + t) and the sum of the products of 2 prefectures D
(k+1) is stored in the minimum statistic register 88 by opening the gate circuit 52.

Further, when the output signal 851S becomes 'ON', the gate circuit 51 opens, and the position information k and 11. at the separation candidate position x (k, ik) in the separation candidate section transferred to the distance calculation unit 81 is It is stored in the connection information register 87.

The above operations are performed to determine the separation candidate position X (k, jk) (however, ik
This is done for nido...hk.

Next, the control unit 10 shown in FIG.
The separation candidate position x (k+1
, ik+t ) as the optimal mean value μ*(0°k+' e ``o e ''k+s ) and the sum of the products of two prefectures D*(k+1) of the optimal distance, the contents of the minimum statistics register wing are converted into the minimum statistics. The contents of the minimum evaluation value register 86 are transferred to the minimum evaluation value group register 264 as the optimal evaluation value of the separation candidate position x (k+1, ik+x), and further, the contents of the separation candidate section
The contents of the connection information register 87 are transferred to the connection information group register 263 as the optimal separation path information to the separation candidate position x (k + 1 e 'k) at .

Next, the contents of the minimum evaluation value register 86 are set to an initial value (a very large value).

By repeating the above operation, add 1 to the separation candidate section.
All separation candidate positions x (k+1 * fh+x
) (where jh+1=1.2...hk+s), an optimal evaluation value and an optimal separation path can be obtained.

It goes without saying that another specific method of implementing the present invention is to use an ordinary microcomputer.

As mentioned above, by applying the present invention,
Even if there is contact between letters, or if one letter is split into two or more letters, it is easy to write -
It becomes possible to perform character-by-character separation.

[Brief explanation of the drawing]

FIG. 1 is a diagram for explaining an example of a method for setting separation candidate sections for character separation in the present invention using an example of a partial character string image, and FIG. FIG. 3 is a logical block diagram showing a specific embodiment of the present invention, and FIG. An example is shown. FIG. 5 is a logical block diagram showing a specific embodiment of the evaluation scale calculating section 8 in FIG. 3. In the figure, 1 is a scanning unit, 2 is a character string image memory,
3 is a character block extraction unit, 21 is a character block information register, 4 is a character pitch detection unit, 22 is a character pitch information register, 30
is a parameter information register, 5 is a permissible interval extraction unit, 23
is a permissible section information register, 6 is a terminal candidate section extractor, 2
4 is a terminal candidate section register, 7 is a separation candidate section extractor,
25 is a separation candidate position information register, 8 is an evaluation scale calculation unit, 26 is an optimal separation position information register, 27 is a character separation position register, 10 is a control unit, 8O is a stage register,
81 is a distance calculation unit, 261 is a separation candidate position group register,
262 is a linked information group register, 263 is an optimal statistics 1 group register, 264 is an optimal evaluation value group register, 82 is a statistics calculation section, 83 is a statistics storage register, 84 is an evaluation value calculation section, and 85 is a ratio axis section. , 86 is a minimum evaluation value register. Year 1 mouth (a) d (f, 2:L2) 4

Claims (1)

[Claims] Using the projection distribution obtained from a series of character line images,
In the character separation device that separates the character line image into individual characters, means for sequentially extracting a plurality of character blocks that can be separated by spaces from the projection distribution, and estimating character pitch using the plurality of character blocks. and the projection distribution and the character pitch in the character line image,
means for setting a separation candidate section; calculating a distance between each separation candidate position in the separation candidate section and the separation candidate position in the adjacent separation candidate section; A character separation device comprising: means for calculating a series of optimal separation candidate positions that minimizes an evaluation scale configured based on an average value of distances and a difference between the character pitches over the character line image. .