JPH0581474A - Character string extracting method and character area detecting method - Google Patents

Abstract

PURPOSE:To perform stable character string extraction with small storage capacity. CONSTITUTION:A document is scanned in S1 and character elements of the document image are extracted in S2. In S3-S5, whether the relation of 'mutual closeness' is satisfied or not is checked as to all pairs of the character elements to define the equal value relation of 'same level', and the character elements are classified in an equal-value group. In S5, the character elements contained in the equal value group are extracted as elements constituting the same character string. In S6 and successors, the extracted character string is divided into characters, which are matched with standard character patterns and recognized.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for recognizing characters from an image, and more particularly to a method for extracting individual character areas and character strings from an input image.

[0002]

2. Description of the Related Art In a conventional character extraction method, a character is extracted as follows from an image (hereinafter referred to as "original image") input using a means such as a scanner. First, of the horizontal or vertical direction in the original image, the black dots of the image are accumulated in a direction (hereinafter, referred to as “character string direction”) that is closer to the direction in which the actual character string advances. What is a black dot?
It is a point that constitutes a character area. And, the part of the cumulative value is the blank part between the character strings,
Further, the position of the character string is determined by regarding the portion having a large cumulative value as the area where the character string exists. Characters are extracted by dividing the character string obtained in this way.

However, in such a method, when the inclination of the character string of the original image is large, the difference in the cumulative value of the black dots is not so large in the horizontal direction and the vertical direction. Therefore, it is not easy to accurately extract a character string, and it is also difficult to accurately extract individual characters.

On the other hand, a method of extracting character elements or character strings by previously extracting element regions (hereinafter referred to as "character elements") of characters that are considered to constitute individual characters and integrating them can be considered. .. In this way,
As a step before extracting characters, it is necessary to extract in advance the character elements that are considered to constitute individual characters.

As an example of a character element, an example of using a connected area of black dots can be seen. A connected region is a region made up of a set of black dots of an image, and a region of a set of black dots in which any two points in the region can be connected by a curve passing only in the region. In such a case, conventionally, the connected area is extracted from the image, and then the coordinates of the rectangle circumscribing the extracted area are obtained, and the character element is detected thereby. According to such a conventional method, it is necessary to hold information about which connected area each black dot in the image belongs to during the processing. Therefore, when such a method is used, a large capacity memory is required. Although there is dedicated hardware for performing this process, the overall system becomes expensive.

Further, in such a method, if the character elements are insufficiently integrated, a single character or character string is separated, and if the integration is excessive, different characters or character strings are separated. May be accidentally merged into a single character or string. Therefore, optimal integration is not easy.

[0007]

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a character string extracting method and a character area detecting method which can minimize the influence of the inclination of the original image and can obtain a stable processing result. Is to provide.

[0008]

According to a first aspect of the present invention, there is provided a character string extracting method, which comprises a step of reading an image and converting the image into a plurality of continuous binary scanning line signals, and a binary scanning line signal sequentially given. Sequential analysis is performed to obtain circumscribing frame information for specifying the shape of the circumscribing frame that circumscribes the connected region represented by the preceding binary scanning line signal before the preceding first binary scanning line signal, and obtains this information. A step of holding and detecting a connecting line segment area included in a second binary scanning line signal immediately after the first binary scanning line signal, and circumscribing a connecting area having a connecting relationship with the connecting line segment area. While updating the frame information, it detects the presence of a connected area that is not connected to any of the connected line segment areas,
Whether a predetermined equivalence relation is established between the step of outputting the circumscribing frame information, the step of detecting the end of the image and outputting all the retained circumscribing frame information, and the step of outputting the circumscribing frame information. By checking whether or not each of the circumscribing frame information pieces is inscribed, a connected area is classified into an equivalence class, and a connected area belonging to the same equivalence class is extracted as a constituent element of one character string.

According to a second aspect of the present invention, there is provided a method for extracting a character string by extracting a character element from an image and checking whether or not a predetermined equivalence relation is established between the extracted character elements. The method includes a step of classifying elements into equivalence classes and a step of extracting character elements belonging to the same equivalence class as constituent elements of one character string.

According to a third aspect of the present invention, there is provided a method for detecting a character area, which comprises the steps of reading an image and converting the image into a plurality of continuous binary scanning line signals, and sequentially analyzing the binary scanning line signals that are sequentially applied. Obtaining circumscribing frame information for specifying the shape of a circumscribing frame circumscribing a connected region represented by a binary scanning line signal earlier than the first binary scanning line signal, and holding the circumscribing frame information; The connection line segment area included in the second binary scan line signal immediately following the binary scan line signal of 1 is detected, and the circumscribed frame information of the connection area having the connection relation with the connection line segment area is updated. , Detecting the presence of a connected area that is not connected to any of the connected line segment areas and outputting the circumscribing frame information, and detecting the end of the image and outputting all the retained circumscribing frame information. And a step of performing.

[0011]

According to the character string extraction method of the first aspect, in order to obtain the circumscribed frame information of the connected area, two adjacent two
It is sufficient to check the connection relation of the connection line segment regions between the value scanning line signals. It is not necessary to hold the information of the connected area to which each pixel belongs for all images. Further, the circumscribing frame information thus obtained is mathematically classified by a method called equivalence class decomposition to extract the constituent elements of each character string.
Whether or not a predetermined equivalence relation is established is determined without being significantly affected by the inclination of the original image. In addition, since the equivalence relation is checked between each character element, the result does not depend on the processing order.

In the character string extraction method according to the second aspect, by checking whether or not an equivalence relation is established between the extracted character elements, the character elements are converted into character strings by a method called equivalence class decomposition. It is extracted as a component of.
The equivalence relation can be checked for success or failure regardless of the inclination of the original image. Further, the processing result by the equivalence class decomposition does not depend on the processing order.

According to a third aspect of the present invention, there is provided a character area detecting method, wherein a connecting area represented by a scan line signal before the preceding first binary scan line signal and a connecting line between the following binary scan line signals. The connection relation with the segment area can be established by examining the connection relation of the connection line segment area between the first and second binary scanning line signals. Therefore, it is not necessary to store information indicating which connected area it belongs to for every pixel in the image.

[0014]

FIG. 13 is a block diagram of a character recognition apparatus for carrying out the present invention. With reference to FIG. 13, this character recognition device includes an image input unit 1 for scanning an image and converting it into a plurality of continuous digital scanning line signals, and character elements from the image input by the image input unit 1. By extracting the character element extracting unit 2 for extracting the character string, analyzing each extracted character element, extracting the character string, and separating the extracted character string into individual characters. An individual character extraction unit 4 for extracting an individual character, a character standard pattern 8 prepared in advance, and the extracted individual character and the character standard pattern 8 are matched to each other, so that an individual character included in the input image is obtained. Character matching unit 5 for recognizing the character, the recognition result post-processing unit 6 for adding necessary correction and shaping to the character recognition result obtained by the character matching, and the post-processed recognition result To Use person and a final result output unit 7 for outputting the converted into a readable form.

This character recognition device is realized by a computer. Then, the image input unit 1 is
The final result output unit 7 is a display, and the character element extraction unit 2, the character string extraction unit 3, the individual character extraction unit 4, the character matching unit 5, the character standard pattern 8, and the recognition result post-processing unit 6 are all central processing units. It is realized by the program to be executed.

Referring to FIG. 14, the character element extraction unit 2
An image memory 21 for storing an image of at least one line input from the image input unit 1, and an image memory 21.
The current line connected area extracting unit 22 for detecting the connected area on the line currently being processed stored in the current line, and outputting the information about the connected area on the current line, and the current line connected area extracting unit 22. Previous line connected area memory 24 for storing information and temporarily storing it for processing for the next line, information about connected areas on the current line extracted by current line connected area extraction unit 22, and Based on the information about the connected area on the line immediately preceding the line stored in the line connected area memory 24, the connection relationship of the connected area between the previous line and the current line is checked, and the connected area on the line A connection relation checking unit 23 for performing a process of integrating rectangles circumscribing the connection area.
And stores the coordinates for identifying the rectangle circumscribing the connected area on the image, which is detected by the connection relationship check unit 23, and determines that the area is not connected to any other area during the processing. Circumscribed rectangular coordinate memory 2 for outputting information representing a connected region as a character element
Including 5 and.

The character string extraction unit 3 has a character element memory 31 for sequentially storing the information of the character elements given from the circumscribed rectangular coordinate memory 25, and a number equal to the number of character elements stored in the character element memory 31. A label memory 35 capable of preparing an integer array having elements, and a label initial for initializing the value of the integer array {Sn} in the label memory 35 according to the number of character elements stored in the character element memory 31. A conversion unit 33 and an equivalence relation inspection unit 32 for extracting any two combinations of the character element information stored in the character element memory 31 and inspecting whether or not a predetermined equivalence relation is established between the respective character elements. For the character elements determined to have the equivalence relation by the equivalence relation checking unit 32, the value of the label corresponding to each character element stored in the label memory 35 is determined in advance. A label updating unit 36 for updating according to the method and all the pairs of character elements stored in the character element memory 31 are subjected to an equivalence relation check, and then have a label having the same value in the label memory 35. An equivalence label character element extraction unit 34 for determining such a character element as a character element belonging to the same character string and extracting it from the character element memory 31.

The device shown in FIGS. 13 to 15 operates in principle as follows. The image input unit 1 scans an input image, converts it into an image signal composed of a plurality of scanning lines, and sequentially supplies the image signal to the image memory 21. The image memory 21 sequentially stores at least one scanning line signal of the input image signals. The current line connected area extracting unit 22 extracts the connected area on the line being processed based on the image data of one line stored in the image memory 21, and the information indicating the extracted connected area is used as the connection relation checking unit 23. And the previous line connection area memory 24. The previous line connected area memory 24 stores information indicating the connected area of one line before.

The connection relation check unit 23 represents the information indicating the connection region on the current line provided from the current line connection region extraction unit 22 and the connection region one line before stored in the previous line connection region memory 24. Based on the information
It is checked whether or not there is a connection relationship between the connection areas included in the current line and the previous line. If it is determined as a result of the check that there is a connection relationship, a predetermined process is performed on the coordinates of the rectangle circumscribing the connection area of the previous line, which is stored in the circumscribing rectangle coordinate memory 25, and the current line connection area is moved forward. Integrate with the connected area up to the line. Details of this integration processing will be described later. In the middle of the process, the circumscribed rectangular coordinate memory 25 stores 1 in the circumscribed rectangular coordinate memory 25 for the connected regions up to the previous line that are not connected to any of the connected regions on the current line. The information is given to the character element device 31 as one character element. When the reading of all the images and the above processing are completed, the circumscribed rectangular coordinate memory 25 outputs the remaining circumscribed rectangular coordinate stored at that time as the character element information.

The character element memory 31 sequentially stores the character element information given from the circumscribed rectangular coordinate memory 25.
The label initialization unit 33 initializes the value of the integer array prepared in the label memory 35 by a method described below, prior to the character string extraction processing. The equivalence relation inspection unit 32 examines, for all pairs of character elements stored in the character element memory 31, whether or not a predetermined equivalence relation is established between the respective character elements. When the equivalence relation is established, the label updating unit 36 rewrites the value of the label in the label memory 35 prepared corresponding to the character elements having the equivalence relation by a predetermined method. This method will be described later in detail.

After determining whether or not an equivalence relation is established between all pairs of character elements stored in the character element memory 31, the equivalence label character element extraction unit 34 determines the label memory 35. The character elements corresponding to the labels having the same value among the respective labels stored in are determined to be the character elements belonging to the same character string, and the corresponding character elements are extracted from the character element memory 31 to extract the individual character. It is given to the extraction unit 4.

The individual character extraction unit 4 appropriately divides the given character string area to take out individual characters and supplies them to the character matching unit 5. The character matching unit 5 checks the similarity between the standard pattern prepared in advance in the character standard pattern 8 and the character to be processed by matching, and the standard pattern with the highest similarity is input to the input character. Recognize as. The recognition result post-processing unit 6 adds necessary corrections and shaping to the recognized result and outputs the final result output unit 7
Give to. The final result output unit 7 outputs the recognized character in a user-readable form.

The character string extraction method according to the present invention is performed by the character string extraction unit 3, and the character area detection method is performed by the character element extraction unit 2.

The character element extraction unit 2 and the character string extraction unit 3 are realized by computer programs as described above. FIG. 1 is a flowchart of a program for realizing the character string extracting method. FIG. 2 is a flowchart of a process for extracting a character element in an image in the program shown in FIG.

Referring to FIG. 1, in step S1,
A document scan of the original image is performed. Control is step S
Go to 2.

In step S2, a process of extracting a character element in the input original image is performed. This processing is realized by a program as shown in FIG.

Referring to FIG. 2, the program for processing for extracting a character element in an image has the following control structure. In the following description, the input image is assumed to be a rectangular image having m pixels horizontally and n pixels vertically. The x axis is in the horizontal direction and the y axis is in the vertical direction. Below, the coordinates of a rectangle that is recognized in this image with one pair of sides parallel to the x-axis and the other pair of sides parallel to the y-axis are expressed as (minimum x coordinate, minimum y coordinate), (maximum X coordinate, maximum y coordinate). Hereinafter, these coordinates are respectively (sx, sy),
It shall be written as (ex, ey). In this case, each coordinate is
Set the length of each side of 1 pixel in the x-axis direction and the y-axis direction to 1
It was taken as.

In the above image, two points (0, i)
Consider a 1-pixel-wide straight line connecting (m-1, i) and parallel to the x-axis. This straight line will be called line i.

The connected area on the line i is defined as a connection of adjacent black dots included in the line i. In the claims, the connection area on the line i is expressed as "connection line segment area". Further, in the following description, the connected area in the image is an area including a set of black dots included in the image as described above, and any two black dots included in the area are different from other black dots included in the area. It is defined as an area that can be connected to each other by tracing.

(1) Referring to FIG. 2, step S11
At, each connected region ( _{denoted as} CL _{0 j} ) on line 0 of the image is extracted. A rectangle circumscribing the connected region CL _{0 j} on the line 0 and having one pair of sides parallel to the x-axis and the other pair of sides parallel to the y-axis is a rectangle to which the connected region CL _{0 j} belongs, and its coordinates are Store in a predetermined storage area. That is, the sx = CL _{0 j} starting x coordinate ex = CL starting y-coordinate (= 0) of the end point x-coordinate sy = CL _oj the _oj of ey = CL _{0 j} endpoint y-coordinate (= 0). The control proceeds to step S12.

(2) In step S12, an initial value "1" is set to a variable i for controlling the following iterations. The control proceeds to S13.

Hereinafter, the processing of steps S13 to S17 will be described.
This is sequentially performed for i that satisfies 1 ≦ i ≦ m−1.

(3) In step S13, each connected region ( _denoted by CL _ij ) on the line i is extracted. And the connecting region CL _ij, and rectangular connection region CL _ij belongs. That is, the coordinates for specifying this rectangle are as follows.

[0034] sx = CL _ij of the start point x coordinate ex = CL _ij endpoint x-coordinate sy = CL _ij of the start point y coordinate (= i) ey = CL _ij endpoint y-coordinate (= i) the control proceeds to step S14.

(4) With respect to the connection area on the line (i-1) and the connection area on the line i, the connection relationship between them is inspected. The determination as to whether or not the connection relationship is established is made as follows. Connected area CL on line (i-1)
The coordinates of the start point and the end point of 0 are (x00, i-1),
(X01, i-1), the coordinates of the start point and end point of the connected region CL1 on the line i are (x10, i), (x11,
i). In this case, it may be determined that the connected regions CL0 and CL1 are connected when the following relationships are established.

X00.ltoreq.x11 and x10.ltoreq.x01 However, in the above formula, the number of pixels that can be connected to one pixel is limited to the pixels above and below and to the left and right of the pixel. This is the formula that should be applied in some cases. In the case of so-called “8-connection”, in which pixels that are diagonally adjacent to one pixel can also be connected to this pixel, the following formula must be used.

X00 ≦ x11 + 1 and x10 ≦ x01 + 1 In short, the condition that the x coordinate of the start point of the connected region on one line does not become larger than the x coordinate of the end point of the linked region on the other line is satisfied. If this condition is satisfied, the two connected regions are in a connected relationship with each other, and if this condition is not satisfied, it is determined that they are not connected. When there is a line (i-1) and a connected region on the line i which are connected to each other as a result of the inspection, the rectangles to which they belong are integrated. The process of integrating a plurality of rectangles is a process of determining coordinates for specifying the shape of a new rectangle as follows.

Sx = min {integrated rectangular sx} sy = min {integrated rectangular sy} ex = max {integrated rectangular ex} ey = max {integrated rectangular ey} Coordinates determined as described above With (sx, sy),
The rectangle R specified by (ex, ey) is an integrated rectangle. Then, these connected regions are assumed to belong to the common rectangle R thus obtained. The control proceeds to step S15.

(5) In step S15, a rectangle to which each connected region on line (i-1) belongs and which has not changed before and after the process of step S14 is detected. When such a rectangle exists, the connected area in the image circumscribing the rectangle is discontinued at the line i. In the subsequent processing, connected areas on other lines connected to this connected area cannot appear.
Therefore, the coordinates for specifying this rectangle do not change in the subsequent processing. Therefore, this coordinate can be output as the final result. At the time of output, the value i−1 may be substituted for the boundary coordinate ey in the y direction of this connected region.
If there is no black dot on the line i currently being processed, all the rectangles to which each connected region on the line (i-1) belongs satisfy the above-mentioned condition, and the coordinates can be output. Then, thereafter, in the process for obtaining the rectangle circumscribing the connected area, it is not necessary to store the coordinates for specifying these rectangles at all. The control proceeds to step S16.

In step S16, the content of the variable i is incremented by 1. The control proceeds to step S17.

In step S17, the value of the variable i is
It is determined whether the number of pixels in the vertical direction (total number of lines) is equal to m. If the determination result is YES, the control proceeds to step S18, but if not, the control proceeds to S1.
Go to 3. Then, as described above, the processes (3) to (5) are repeated for the corresponding variable i.

(6) In step S18, the coordinates of the rectangle to which the black dot on the line (m-1) belongs are output as the final step. In this case, the coordinate value ey of this rectangle
The value of is m-1.

By using the program that operates according to the flow as shown in FIG. 2 as described above, the coordinates for specifying the shape of the rectangle circumscribing each connected region in the image can be obtained. In this case, the image information necessary for the processing is only the information about each point included in the current line of interest and the line immediately before it. Information about points in the image included in the line before the line two lines before is not necessary at all. for that reason,
According to this method, the storage capacity required for detecting a character area can be made much smaller than that in the case of recording information about all images as in the conventional case.
Further, the circumscribed rectangular coordinates obtained as the final result are sequentially output during the process. Therefore, it is possible to perform other processing in which the final result is input in parallel with this processing, and it is possible to improve the speed of the entire processing.

4 to 6 are schematic views of images showing the progress of the processing shown in FIG. Referring to FIG. 4, on line (i-1), there are connection regions (connection segment regions) A and B on line (i-1). On the other hand, on the line i, the connecting regions (connecting line segment regions) C, D, and
E exists. According to the definition of connection area, connection (line segment)
Regions A and C are connected to each other. The connected (line segment) areas B, D, and E are also connected to each other.

In FIG. 5 and FIG. 6, the rectangle on the line (i-1) to which the separate connection regions belong is indicated by the line i.
The following shows how they are integrated in the processing for. With reference to FIG. 5, the connection regions A and B on the line (i-1) belong to the common rectangle R1. The connected region C on the line (i-1) belongs to a rectangle R2 different from the rectangle R1.

Referring to FIG. 6, it is assumed that a connected (line segment) region D exists on line i. Connection (line segment) area D
Is connected to the connected (line segment) regions B and C on the line (i-1). In this case, the rectangle itself to which the connected (line segment) area D belongs has the same shape as the connected area D itself. According to the processing described above, the connection area B on the line (i-1),
By integrating the rectangles R1 and R2 to which C belongs and the rectangle (D itself) to which the connected region D on the line i belongs,
The rectangle R0 shown in FIG. 6 is obtained. As described above, the connecting regions A and B on the line (i-1) have the common rectangle R1.
Therefore, the connection area A also belongs to the rectangle R0.

In this way, the connection relationship between the connection area on each line and the connection area one line before is checked in order, and the rectangles to which they belong are sequentially integrated, whereby the area to which the character element on the image belongs Are separated and combined. Therefore, each connected area is separated according to the rectangle to which it belongs.
Will be combined.

FIG. 7 shows the result of separating / combining a character image into each connected area which is a character element in this way. As shown in FIG. 7, the character image "system" is divided into each character element.

Referring again to FIG. 1, steps S3 to S5
In, the process of extracting a character string is performed according to the character string extracting method according to the present invention. In this process, each character element extracted in step S2 is classified into a plurality of equivalence classes by a method mathematically called "equivalence class decomposition". For the equivalence class decomposition, in the case of the present embodiment, the relationship “close to each other” is defined as follows for two character elements. However, the relationship of “close to each other” does not necessarily have an equivalence relationship.

It is assumed that rectangles a and b are circumscribing the two character elements, respectively. The distance in the character direction between the center points of the two rectangles a and b in the character string direction is L, and the lengths of the rectangles a and b in the character string direction are s _a and s _b . Further, the distance between the center points of the rectangles a and b in the direction perpendicular to the character string direction is H, and the distances between the rectangles a and b are the rectangles a and b.
The string length and the vertical length of h _a and h _b , respectively.
And In this case, two character elements are defined as “close to each other” when the following expressions are satisfied at the same time.

[0051] _{L ≦ max {s a, s} b} * c0 H ≦ max {h a, h b} * c1 However, c0, c1 are each constant. In this embodiment, the constant c0 is about "4" and the constant c1 is about "0.5". However, this value is just an example.

In step S3, an integer array {Sn}
Be prepared. The number of elements of this integer array {Sn} is
It is the same as the number of character elements. And the integer array {Sn}
The values of the respective elements are initialized so that they are different from each other, that is, if i ≠ j, then S _i ≠ S _j .
For example, if the value of the i-th element Si is i (i = 0, 1,
2, ..., N). The control proceeds to step S4.

In step S4, the process shown in the flowchart of FIG. 3 is performed. In step S4A, two different combinations of character elements are extracted once from the character elements extracted in step S2 in the image. If the i-th character element is represented as Ci, the combination of the two character elements is (Ci, Cj) (i,
j = 0, 1, 2, ..., N, where i ≠ j). In step S4B, this combination (Ci, C
It is determined whether or not j) satisfies the above-mentioned "close to each other" relationship. If the determination result is YES, the control proceeds to step S4D, otherwise it proceeds to step S4C. In step S4C, the character elements Ci, C
The values of the elements Si and Sj of the integer array corresponding to j are rewritten as follows.

(A) If Si <Sj, the value of the element of the integer array {Sn} whose value is equal to the value of Sj is rewritten to the value of Si. In this case, the value of Sj is also rewritten to Si.

(B) If Si> Sj, the value of the element whose value is equal to Si among the elements of the integer array {Sn} is
Rewrite to the value of Sj. In this case, the value of Si itself is also rewritten to Sj.

(C) If Si = Sj, nothing is done. Then, in step S4D, it is determined whether or not the processing has been completed for all combinations (Ci, Cj) of character elements. If the answer to the decision is NO, control returns to step S4A, otherwise the process for this part ends and the process proceeds to step S5 in FIG.

In step S4, the processing according to the above case classification is performed for all combinations (pairs) of character elements. When all combinations of character elements have been processed, the character elements corresponding to the elements having the same value among the elements of the integer array {Sn} may be determined to belong to the same character string. Therefore,
In step S5, the character string can be extracted by collecting the character elements corresponding to the elements having the same value among the elements of the integer array {Sn}.

A little explanation will be added to the above relationship. The fact that the two character elements Ci and Cj are “close to each other” is represented as Ci to Cj. A set of integers n1, n2,
, Nk and m1 and m2, the relationship shown in the following equation holds.

Cm1 to Cn1 to Cn2 ... Cnk to C
m2 This does not necessarily mean that the character elements Cm1 and Cm2 are directly “close to each other”, but by sequentially tracing the character elements that are “close to each other” from one to the other, Indicates that you can reach.
Of course, the case of Cm1 to Cm2 is also included in this. Cm1 and Cm2 having such a relationship are referred to as “having the same label”.

By definition, any combination of two character elements among Cn1, Cn2 and the difference Cnk appearing in the path from Cm1 to Cm2 at this time also has a relation of “having the same label”.

The relationship that the character elements Cm1 and Cm2 "have the same label" means that the integer array {S when the processing in step S4 is finished in the above description.
Of the elements of n}, the elements Sm1 and Sm2 corresponding to the character elements Cm1 and Cm2 have the same value.

Further, a set of integers i1, i2, ..., Im, j
For j, 1, j2, ..., Jn, and i, j, the following 2
It is assumed that two relationships are established.

Ci1-Ci2 -...- Cim-Ci Cj-Cj1-Cj2 -...- Cjm This means that Ci during the process of step S4.
The value of the element of the integer array {Sn} corresponding to 1, Ci2, ..., Cim is equal to the value of the element Si corresponding to Ci, and Cj
, Cj2, ..., Cjm correspond to the value of the element Sj corresponding to Cj.

At this time, assuming that Ci to Cj are known, the following relationship is obtained. Ci1-Ci2 -...- Cim-Ci-Cj-Cj1-C
j2 -...- Cjm This means that in step S4C (see FIG. 3), 2
When two character elements Ci and Cj satisfy the relationship of being “close to each other”, in the integer array {Sn} corresponding to them, if Si <Sj, the strength value having the same value as Sj is set to Si.
, And if Si> Sj, the value of the element having the same value as Si is rewritten to the value of Sj.

As described above, the relationship of “close to each other” is not necessarily an equivalence relationship mathematically, but the relationship of “having the same label” defined by using it is a kind of equivalence relationship.

The relationship "having the same label" defined above is mathematically a kind of equivalence relationship.
A method of classifying character elements according to whether or not this equivalence relation holds between them is a work called equivalence class decomposition in mathematics. The equivalence class decomposition has a remarkable feature that the processing result does not depend on the processing order. In the case of the above-described embodiment, the character elements in the image are “close to each other”.
The result obtained is the same no matter what order the two character elements are taken to determine whether or not.

Then, in step S6, step S
The character string extracted by the processes up to 5 is appropriately divided in the character string direction to perform the process of extracting individual characters. The control proceeds to step S7.

In step S7, the individual character obtained in step S6 is matched with the prepared character standard pattern (see FIG. 3), and the character standard most similar to the character to be processed is set. Processing for obtaining the pattern as a recognition result is performed. The control proceeds to step S8.

In step S8, post-recognition result processing is carried out for the characters recognized in step S7 to make necessary corrections and shaping. The control proceeds to step S9.

In step S9, a process of outputting the final recognition result obtained in step S8 in a user-readable form, for example, a print character or a display display form is performed. Then, the process up to step S9 is performed for all the characters, and the program ends.

FIGS. 8 to 12 are schematic diagrams showing how a character string is extracted by the method according to the present invention when a connected region in an image is taken as a character element. In the following description, the element of the integer array {Sn} corresponding to the character element Ci is expressed as Si.

With reference to FIG. 8, Si given to each character element (connected region) in the image is initialized to a different value. This process is performed in step S3. In FIG. 8, 0 to 1 are added to 11 connecting regions in total.
A value of 1 is assigned to each.

Referring to FIG. 9, the character elements C0 and C3 are close to each other. For confirmation, it is considered that the relationship of “close to each other” according to the above definition is established when the distance between rectangles to which two connected regions belong is smaller than a predetermined value on the image. .. Since the character elements C0 and C3 are close to each other, the values of S0 and S3 are compared according to the processing of step S4 of FIG.
In this case, S0 <S3 holds. That is, (a) is satisfied under the above conditions. Therefore, all Si values having the same value as S3 are rewritten to S0. In this case, there is no Si other than S3 that has a value equal to S3. Therefore, only S3 is rewritten to the value (0) of S0. The result of rewriting is shown in FIG.

FIG. 10 is a diagram showing an intermediate state of the processing to be performed subsequently. Before reaching the state of FIG. 10, it has been confirmed that the character elements C0, C1, C3, and C4 have a relationship of being “close to each other”, and Si is rewritten accordingly. Furthermore, as shown in FIG. 10, it is assumed that it is confirmed that the character elements C4 and C8 satisfy the relationship of being “close to each other”. At this time, S4 = 0, S8 =
7 and S4 <S8 is satisfied. The above condition (a)
According to, all Si having a value equal to S8 (= 7) is rewritten with the value of S4. As a result, as shown in FIG. 11, 0 is assigned as a corresponding integer to each of the character elements included in the character string “system” in the image.

Referring to FIG. 12, "system" in the image
There is no relation "close to each other" between the character element in the character string "" and the character element in the character string "menu". Therefore, by performing the above-described processing for all combinations of arbitrary two character elements in this image, the value of Si as shown in FIG. 12 can be obtained as the final result. All character elements in the image are classified by the integer value associated with the character element as described above. In the examples shown in FIGS. 8 to 12, the respective character elements are classified into those having 0 and 2 as the corresponding integers, and the character strings are “system” and “menu”, respectively.

In step S6 and subsequent steps, a process is carried out in which individual characters are extracted from the resulting character string obtained in FIG. 12 by appropriately dividing the character string, and each character is recognized by matching.

As described above, in the character string extracting method according to the present invention, the character string is extracted by the method called equivalence class decomposition. The processing result by equivalence class decomposition does not depend on the processing order, and a stable processing result can be obtained. Further, between the adjacent character elements, the influence of the inclination of the image on the above-mentioned "close to each other" is very small. The influence of the image inclination on the above-mentioned character string extraction is small. Therefore, as compared with the method according to the related art, there is an effect that a proper character string can be extracted for a wide range of inclination of an input image.

Although the present invention has been described based on the above embodiment, the present invention is not necessarily limited to the above embodiment. For example, the definition of the relationship “close to each other” is not limited to the definition given in the above description, and
It does not generally hold when two character elements belong to different character strings, and may generally be set when two character elements belong to the same character string and are adjacent to each other.

The method of rewriting the values of the elements of the above-mentioned integer array is not limited to the methods described in the case classifications (a) to (c) as described above. For example, (a)-(c)
In the formula given in (1), the symbols “>” and “<” may be interchanged.

Although the present invention has been described in detail based on the embodiments, it is needless to say that the present invention is not limited to the above-mentioned embodiments and can be implemented with various modifications. Absent.

[0081]

As described above, according to the character string extracting method of the first aspect, the information for specifying the frame circumscribing the connected area in the image is included in the two adjacent scanning line signals. It is obtained by examining the connection relation of the connection line segment areas. Unlike the conventional case, it is not necessary to store which connected region all pixels on the image belong to, and the required storage capacity can be greatly reduced. Further, by checking whether or not an equivalence relation is established between the obtained circumscribing frame information, the connected region included in the image is classified into an equivalence class,
Connected regions that belong to the same equivalence class are extracted as elements that form one character string. In this classification process, a constant result can be obtained regardless of the processing order, and a stable processing result can be obtained. As a result, it is possible to provide a character string extraction method that can obtain a stable processing result with a small storage capacity.

According to the character string extraction method of the second aspect, a process of classifying character elements into equivalence classes by checking whether or not a predetermined equivalence relation is established between the extracted character elements. Is performed. Then, the character elements belonging to the same equivalence class are extracted as the elements constituting one character string. The processing of classifying into the equivalence class based on whether the equivalence relation is correct or not can obtain a stable processing result without depending on the processing order.

According to the character area detecting method of the third aspect, the information for specifying the circumscribing frame circumscribing the connecting area in the image is the connecting line segment area included in the two adjacent scanning line signals. It can be done only by examining the connection relationship.
It is not necessary to store the connected area to which all the pixels in the image belong, and the storage capacity required to obtain the circumscribing frame information can be greatly reduced.

[Brief description of drawings]

FIG. 1 is a flowchart of a program for carrying out a character string extracting method according to the present invention.

FIG. 2 is a flowchart of a program for realizing a process of extracting a character element in an image shown in FIG.

FIG. 3 is a flowchart of a program for realizing the process of extracting the character string shown in FIG.

FIG. 4 is a schematic diagram showing a connection relationship between connection regions on lines included in two scanning lines in an image.

FIG. 5 is a schematic diagram showing connected areas on a line and rectangles to which the connected areas belong, which are included in different connected areas.

6 is a schematic diagram showing a state in which the rectangles to which the two connected regions shown in FIG. 5 belong are merged into the rectangle R0 due to the presence of the connected region D on the line i.

FIG. 7 is a schematic diagram showing a connected region extracted from an image.

FIG. 8 is a schematic diagram showing a state in which an integer value assigned to each connected region on an image is initialized.

FIG. 9 is a schematic diagram showing an intermediate process of character string extraction in the case where the relationship “close to each other” is established between the character elements C0 and C3.

FIG. 10 is a schematic diagram of an image when the character elements C4 and C8 satisfy the relationship of being “close to each other”.

FIG. 11 is a schematic diagram showing a state after updating an integer value assigned to each character element when the relationship “close to each other” is established for the character elements C4 and C8.

FIG. 12 is a schematic diagram showing a result of extracting a character string by the character string extracting method according to the present invention.

FIG. 13 is a block diagram of a character recognition device to which a character string extraction method according to the present invention is applied.

FIG. 14 is a more detailed schematic block diagram of the character element extraction unit 2.

FIG. 15 is a more detailed schematic block diagram of the character string extraction unit 3.

[Explanation of symbols]

 1 image input part 2 character element extraction part 3 character string extraction part 4 individual character extraction part 5 character matching part 6 recognition result post-processing part 7 final result output part 8 character standard pattern

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Minako Kuwata 22-22 Nagaike-cho, Abeno-ku, Osaka-shi, Sharp Corporation (72) Kazuhiro Takehara 22-22 Nagaike-cho, Abeno-ku, Osaka City, Sharp Corporation

Claims (3)

[Claims] 1. A step of reading an image and converting it into a plurality of continuous binary scan line signals, and sequentially analyzing the binary scan line signals that are sequentially applied to precede the first binary scan line signal. Obtaining circumscribing frame information for specifying the shape of the circumscribing frame, which circumscribes the connected region represented by the binary scanning line signal earlier, and holds the circumscribing frame information; A connection line segment area included in the second binary scan line signal immediately following the value scan line signal is detected, and the circumscribing frame information of the connection area having a connection relationship with the connection line segment area is updated. Together with the step of detecting the presence of the connected area that is not connected to any of the connected line segment areas, and outputting the circumscribing frame information, and detecting the end of the image, all the held Outputs the information of the circumscribed frame And a step of classifying the connected region inscribed in each of the circumscribing frame information into equivalence classes by checking whether or not a predetermined equivalence relation is established between the step and any two of the circumscribing frame information, And a step of extracting the connected regions belonging to the same equivalence class as constituent elements of one character string. 2. A step of extracting a character element from an image, and a step of classifying the character element into an equivalence class by checking whether or not a predetermined equivalence relation is established between each of the extracted character elements. And a step of extracting the character elements belonging to the same equivalence class as constituent elements of one character string. 3. A step of reading an image and converting it into a plurality of continuous binary scan line signals, and sequentially analyzing the binary scan line signals that are sequentially applied to precede the first binary scan line signal. Obtaining circumscribing frame information for specifying the shape of the circumscribing frame, which circumscribes the connected region represented by the binary scanning line signal earlier, and holds the circumscribing frame information; A connection line segment area included in the second binary scan line signal immediately following the value scan line signal is detected, and the circumscribing frame information of the connection area having a connection relationship with the connection line segment area is updated. Together with the step of detecting the presence of the connected area that is not connected to any of the connected line segment areas, and outputting the circumscribing frame information, and detecting the end of the image, all the held Output the information of the circumscribed frame A character area detection method including a step.