JPH0461395B2 - - Google Patents

Description

[Detailed Description of the Invention] (1) Description of the field to which the invention pertains The present invention provides a method for converting characters written on a form into
It is related to line reading, in which character pattern information is scanned by a photoelectric converter such as a self-scanning semiconductor image sensor and stored in a buffer memory, and character lines are accurately cut out from this information. This invention relates to a character extraction method that extracts a target character line without being affected by the slant even if the input is slanted.

(2) Description of conventional technology Conventionally, this type of character line reading device generally
As shown in the figure, there are many character lines A ₁ , A ₂ , A ₃ ,...
The form 1 on which is printed or handwritten is sent in the Y direction perpendicular to the character line using the paper feed mechanism, or this form is fixed, and the photoelectric converter is fed in the reverse Y direction, and the feeding is stopped. Then, by moving the photoelectric converter along the character line in a direction parallel to the character line, the character line is scanned two-dimensionally, and the character pattern information is photoelectrically converted, as shown in Figure 2. It was configured to write and store data in a buffer memory 2. Note that _Y1 in FIG. 1 indicates the scanning area, and M in FIG. 2 indicates the peripheral distribution.

This type of scanning method requires how to write each character line on a form, how to cut the form with the character lines written on it, how to set the form on the line reading device, how to set the feeding mechanism and scanning mechanism, and how to Unless strict restrictions are placed on the paper quality, etc., when a character line is scanned to be read, an inclination will occur between the target line and the scanning direction, and the target line will be written on the buffer memory 2 at an incline. Therefore, it was often difficult to separate the destination bank from other banks.

Conventionally, in order to eliminate the influence of such slanting of character lines, line marks are added to both ends of each character line, and the character line is determined by assuming that the character line exists on a straight line connecting the line marks. The method used was to cut out the document and cut the document with strict dimensional accuracy to avoid inclination between the target line and the scanning direction X.

However, when the purpose of reading existing documents containing many lines is to add line marks to both ends of each character line, the method of adding line marks to both ends of each character line requires a lot of manpower to enter the line marks, and the form is large. The problem is how to detect line marks when the height changes in various ways. In addition, requiring high precision in cutting makes it inconvenient when using the device, and there are cases where it cannot be applied to OCR.

(3) Purpose of the Invention In order to eliminate these drawbacks, the present invention aims to accurately extract a character line to be read by using the peripheral distribution of character pattern information written in a buffer memory. It is.

Embodiments of the present invention will be described below with reference to the drawings.

(4) Description of structure and operation of the invention FIG. 3 shows an embodiment of the invention, in which 10 is a photoelectric converter, 20 is a pattern memory section, 2 is a buffer memory thereof, and 30 is an instruction for the number of divisions of character pattern information. 40 is a peripheral distribution creation section, 50 is a line cutting area calculation section, 60 is a line cutting section, and 70 is an output terminal.

In operation, the photoelectric converter of the photoelectric converter ₁₀ scans an area Y1 including the destination row _A1 of the form 1 as shown in FIG.
0 buffer memory 2.

When the scanning of area Y ₁ is completed, the division number instruction section 3
0, the character pattern information in the buffer memory 2 is divided by a predetermined number of divisions, or the number of divisions is increased sequentially starting from the number of divisions of 1, that is, the case where no division is performed, and the condition that there is a state of marginal distribution is set. That is, the black pixel row region and the white pixel row region in the peripheral distribution can be detected, and the period of appearance of these regions is approximately equal to the character row period, or the height of the black pixel row region is approximately equal to the height of the character. Instruct to increase the number of divisions until the following condition is satisfied.

The marginal distribution creation unit 40 divides the character pattern information in the line direction according to the instructions from the division number instruction unit 30, creates a peripheral distribution for each divided area, and also calculates the center of the peripheral distribution and the location of the character line. The direction of inclination of the character line is detected from the relationship with .

Figure 4 shows that the buffer memory 2 is divided into two parts P ₁ and P ₂
Then, the marginal distribution M ₁ obtained for each divided area,
An example of _M2 is shown. The marginal distribution is buffer memory 2
The above character pattern information is scanned for each divided area and projected onto the Y-axis, and pixel rows where black points larger than a certain threshold exist are designated as black pixel areas M ₁₂ , M ₁₄ , M ₂₂ ,
M ₂₄ , otherwise white pixel row regions M ₁₁ , M ₁₃ ,
It is composed of two types of regions as M ₂₁ , M ₂₃ , and M ₂₅ .

In addition, the inclination direction is determined by determining whether the character lines forming the marginal distribution M ₁₂ are connected to the character lines forming the marginal distribution M ₂₂ , but are connected to the character lines forming the marginal distribution M ₂₄ . This can be determined by determining the overlap of and the distance between the center points of the peripheral distribution (targeting cases where the peripheral distribution of black pixels does not touch the frame of the buffer memory 2).

In other words, the marginal distribution obtained by projecting onto the Y axis
The center points of M ₁₂ , M ₁₄ , M ₂₂ , M ₂₄ are Y ₁₂ , Y ₁₄ ,
If Y ₂₂ and Y ₂₄ , the marginal distribution M ₁₂ in area P ₁
The character lines forming 2 overlap with this marginal distribution.
The character lines formed by the two marginal distributions M ₂₂ and M ₂₄ are the marginal distributions of the connected character line candidates in the area P ₂ ,
Distance between center points of marginal distribution |Y ₂₂ −Y ₁₂ |, |Y ₂₄
-Y ₁₂ | is examined, and the one with the closest distance is considered to be a connected character line.Thereby, the character line that is connected to the character line forming the marginal distribution M ₁₂ is the character line forming the marginal distribution M ₂₂ . I understand.

As a result, the position of the center point of the peripheral distribution of the target character line is Y ₁₂ in area P ₁ (left side) and Y ₂₂ in area P ₂ (right side), and from the size relationship between Y ₁₂ and Y ₂₂ , the target It can be seen that the character line _A1 slopes upward to the right.

The line cutting area calculation unit 50 includes the division number instruction unit 30
The cutout area is calculated from the information on the number of divisions from , and the information on the peripheral distribution and the inclination direction of the character line from the peripheral distribution creation unit 40 .

Taking Figure 4 as an example, there are white pixel row areas M ₁₁ on both sides.
The black pixel row region M ₁₂ adjacent to M ₁₃ and the white pixel regions M ₂₁ and N ₂₃ on both sides, and the black pixel row regions M ₂₂ and M ₂₄ adjacent to M ₂₃ and M ₂₅ are approximately the same as the character row period. Since they appear at equal intervals, these are detected as candidates for character lines, and when combined with the information that the character line is in the upper right area, it is possible to detect M ₁₂ and M ₂₂ as one complete character line. Become. As a result, the range in which the marginal distribution M ₁₂ was created in the region P ₁ and the range in which the marginal distribution M ₂₂ was created in the region P ₂ become the target extraction regions, and their addresses θ ₁ , θ on the buffer memory 2 ₂ , θ ₃ , θ ₄ and θ ₅ , θ ₆ , θ ₇ , θ ₈ can be calculated.

In addition, two black pixel line areas of complete character line A ₁
Find the centers Y ₁₂ and Y ₂₂ in the Y coordinate of M ₁₂ and M ₂₂ , respectively, and calculate the difference between the centers Y ₁₂ and Y ₂₂ : Y ₂ = Y ₁₂ −
Find Y ₂₂ . In addition, the peripheral distribution on the X coordinate axis of character line A ₁ is N, and the black pixel row areas of areas P ₁ and P ₂ are N ₁₂ and N ₂₂ , respectively, and the centers X ₁₂ and X ₂₂ on the X coordinate of N ₁₂ and N ₂₂ are Find its center x ₁₂ and
Find the difference between X ₂₂ and X ₂ = X ₁₂ −X ₂₂ . Furthermore, Y ₂
When the ratio Z= _{Y 2 /} X ₂ between the character line A ₁ and

The line cutting section 60 cuts out a complete character line from the buffer memory 2 in the pattern memory section 20 based on the information from the line cutting area calculating section 50 and outputs it to the output terminal 70.

In this way, the character line cutting method of the present embodiment can be used even when a skew occurs in a form when scanning the form with a photoelectric converter, and character pattern information is stored in the buffer memory 2 at an angle. Able to read lines of text with high accuracy. Also, the character line
The amount of inclination and the direction of inclination determined for _A1 can be used to correct the inclination of the character pattern when reading the character pattern information from the buffer memory 2 and sending it to the character recognition section.

Note that this method can also be easily applied to a character line reading method in which character lines are cut out after scanning and photoelectrically converting the entire surface of a form.

(5) Explanation of effects As explained above, when reading the required character line by photoelectrically converting the characters on a form, the inclination amount and direction of the character line can be detected with high accuracy, and the target line can be detected with high accuracy. Since it can be cut out without errors, it can be used flexibly to cut out lines when reading characters input using OCR or facsimile, even if a form or a line of characters on a form is input into memory at an angle.

In addition, it is possible to expand the range of objects to be read, and there is no need to add line marks to each character line or to both ends of the character line, and there is more margin in the accuracy of line mark entry and cutting accuracy of forms, etc. Easier to operate.

Furthermore, when scanning the first target line, the amount and direction of inclination of the character line can be detected and used as the amount and direction of inclination for subsequent character lines, so even in documents with a large number of lines. , each line can be read accurately and at high speed without any deviation.

[Brief explanation of the drawing]

Fig. 1 is an explanatory diagram of a general character line extraction method, Fig. 2 is an explanatory diagram showing the state in which character pattern information is stored in the buffer memory and the peripheral distribution of skewed character lines, and Fig. 3 is an explanatory diagram showing the state in which character pattern information is stored in the buffer memory, and Fig. 3 is an explanatory diagram showing the peripheral distribution of character lines with inclination. FIG. 4, a block diagram showing one embodiment of the invention, is an explanatory diagram for explaining cutting out of character lines in this embodiment. In the figure, 2 is a buffer memory, 10 is a photoelectric conversion unit, 20 is a pattern memory unit, 30 is a division number instruction unit, 40 is a peripheral distribution creation unit, 50 is a line extraction area calculation unit, 60 is a line extraction unit, and 70 is an output terminal,
A ₁ and A ₂ are character lines, M ₁ and M ₂ are marginal distributions in the line direction,
N indicates the peripheral distribution in the column direction, Y ₁₂ and Y ₂₂ indicate the center of the black pixel row region of the target row, and X ₁₂ and X ₂₂ indicate the center of the black pixel column region of the target row, respectively.

Claims (1)

[Claims] 1. A function for scanning a form on which a line of characters is written using a photoelectric converter, storing character pattern information converted into a photoelectric conversion signal in a buffer memory, and then detecting a line of characters from this buffer memory. An optical character reading device comprising: a buffer memory that stores a photoelectric conversion signal; and a buffer memory that divides the character pattern information stored in the buffer memory into a predetermined number of divisions in a scanning direction, and is perpendicular to the scanning direction of the entire divided area. Until the condition that the appearance period of the black pixel row region and the white pixel row region in the peripheral distribution described later in the direction is approximately equal to the character row period or the height of the black pixel row region is approximately equal to the height of the character is satisfied, a division number instruction section that instructs to change the number of divisions in the scanning direction; and a black pixel amount for each area obtained by dividing the character pattern information stored in the buffer memory into a plurality of regions in the scanning direction by the number specified by the division number instruction section. At the same time, the center coordinates of the black pixel row region of the peripheral distribution in the direction orthogonal to the scanning direction of the plurality of divided regions are calculated, and the center coordinates of the black pixel row region in the plurality of divided regions are The distance between the center points of the black pixel row regions of the two regions is calculated, and it is assumed that the black pixel row regions having the center point with the shorter distance constitute the same consecutive character rows, and the same character rows are constructed. Calculate the ratio between the difference in center coordinate values in the scanning direction of the black pixel row region in two of the divided regions and the difference in center coordinate values in the direction orthogonal to the scanning direction, and calculate the polarity of this ratio. a marginal distribution creation unit that calculates the inclination direction of the character line from the ratio and determines the amount of inclination of the character line from the value of this ratio, the information on the number of divisions from the division number instruction unit, and the marginal distribution from the marginal distribution creation unit. and a line cutting area calculation unit that creates a cutting area from the information on the inclination direction of the character line; and based on the information from the cutting area calculation unit, cutting out a complete character line from the buffer memory,
A character line cutting device comprising a line cutting section outputting to an output terminal.