JPS58222381A - Character stubbing system - Google Patents

Abstract

PURPOSE:To stub correctly a separated character pattern, by detecting and processing sets of isolated adjacent patterns satisfying a prescribed condition with respect to the character pattenr width and pattern interval as the candidate of the separated character pattern. CONSTITUTION:Contrast information in the unit of picture element read at a reading section 1 is converted 2 into a binary white-black digital picture and inputted to a row stubbing section 3, from which the picture data of one character is stubbed and stored in a row picture storage section 7. The projection of the data of the storage section 7 is extracted at a projection extracting section 4, and the data such as the width of projection at each section of the projection, the center position, and adjacent center intervals are stored in a projection storage section 8. The printing pitch satisfying the prescribed condition from the storage section 8 is discriminated and classified at the discriminating section 9 and given to a character stubbing and discriminating section 6 and a separating character pattern candidate detecting section 10 according to the separation. An isolation pattern stubbing section 5 refers to the content of the storage section 8, the isolated pattern is detected from the storage section 7 and gives the result to the discriminating section 6 and a base line extracting section 11. The discriminating section 6 stubs one of the candidates of the detecting section 10 from the position of the extracting section 11 to output the result.

Description

DETAILED DESCRIPTION OF THE INVENTION Technical Field The present invention relates to a technique for cutting out individual character patterns from an input image in an optical character reading device or the like.

Conventional technology Basically, a method involves cutting out character lines from an input image, extracting vertical projections for each character line, searching for projection breaks, and cutting out isolated patterns corresponding to the range of each projection as character patterns. Characters can be extracted by However, with this method, there is a problem in that a separated character pattern such as "ha" is erroneously cut out as two separate patterns.

According to the technique disclosed in Japanese Patent Application Laid-Open No. 55-25105, a separated character pattern can be cut out as one character pattern based on the relationship between the pattern gap and the printing pitch.
The above problems can be dealt with. However, this only applies to a fixed printing pitch. In the case of proportional printing in which the printing pitch changes, the standard printing pitch is not fixed, so errors in cutting out separated character patterns occur.

OBJECT It is an object of the present invention to provide a character cutting method that can correctly cut out separated character patterns in proportional printing.

Summary The present invention provides a character extraction method that creates a projection of image data on a character line in a direction orthogonal to the character line, and cuts out an isolated pattern corresponding to the range of each isolated projection as one character pattern. h, pattern width,
A set of adjacent isolated patterns that satisfy a predetermined condition for width and pattern spacing is detected as a separated character pattern candidate, and from the positional relationship between the bottom position of each isolated pattern in each detected separated character pattern candidate and the baseline position. It is characterized in that it is determined whether the separated character pattern candidate is one separated character pattern, and the separated character pattern candidate for which the judgment result is positive is treated as one separated character pattern and is cut out at '1'::1. be.

Example FIG. 1 is a block diagram showing one embodiment of the present invention.

2 is a reading unit that optically scans the document, reads the grayscale information on the document by decomposing it into pixel units, and outputs it as an analog image signal.

The binarization unit 2 binarizes this analog image signal using a certain threshold value and converts it into white/black binary digital image data. At this time, processing is also performed to remove isolated black pixels surrounded by white pixels as noise. Note that the reading section l is C
Although it is constituted by a known means such as a CD scanner, here it is assumed that the original document is scanned at a scan line density of 8 lines/mm.

The digital image data is manually input to the line cutting section 3, where line cutting is performed. Various methods are known for this row cutting, and any of these methods may be used. For example, the presence or absence of a black pixel is detected for each scanning line, and the image is projected in the character line direction (main scanning direction).

It is possible to find a shadow and, if the width of the projection or the height of one character line is appropriate, cut out the image data on the scanning line within the range of the projection as image data of one character line. The image data of one character line thus cut out is temporarily stored in the line image storage section 7.

When the extraction of one character line is completed, the projection extraction unit 4 extracts the character line image data in the line image storage unit 7 in the vertical direction (sub-scanning direction).
Start extracting the projection of . For example, image data is rask-scanned from the left end to the right end of a character line with the vertical direction as the main scanning direction, the projection is extracted by checking the presence or absence of black pixels on each scanning line, and each The width of the projection, the center position of the projection, and the center distance between adjacent projections are obtained as projection data and stored in the projection storage section 8.

In parallel with the projection extraction described above, the print pitch identification section 9 operates. First, from among the projection center intervals stored in the projection storage section 8, only those suitable for identifying the printing pitch are selected. In other words, projection center intervals corresponding to collocated character patterns and spaces (spaces between words, spaces before and after character lines) are excluded as inappropriate data. In this embodiment, three types of printing pitches are to be identified: "inch 1O", 1 inch 12j, and "proportional". Therefore, the data on the projection center spacing regarding projections having a width of 21 pixels or more and the projection center spacing larger than the intra-path prime are removed as unnecessary data. Next (2. Examine the distribution of the data of the appropriate projection center spacing selected in this way,
Identify print pitch. The specific steps are, for example, 5
One set of projection center spacing data with pixels greater than or equal to 14 pixels, one set of projection center spacing data in the range of 20 to 24 pixels, and one set of projection center spacing data in the range of 15 to 19 pixels. Classify into each category. When the total number of data for each set of l, n, l reaches a predetermined value N, l.

If, check the number of data in one set. If the number of data classified into one set is 2 or more than 75% of the total number N of data, it is determined that the number is "inch 10". If the number N3 of data in one set is 75% or more of the total number N of data, it is determined to be "inch 12".・Also, the number of each data of l, l, one set N1. N
2. If any of N3 exceeds 25% of the total number N of data, it is determined to be "proportional". This determination result is sent to the character cutout determining section 6. Further, if it is determined to be "proportional", the separated character pattern candidate detection section 10 is activated.

When the projection extraction section 4 completes the projection extraction, the isolated pattern cutting section 5 is activated. That is, while referring to the projection data in the projection storage section 8, isolated patterns within the range of each projection are sequentially extracted from the row image storage section 7, and then manually sent to the character segmentation determination section 6 and the baseline extraction section 11. do.

The baseline extraction unit 11 detects the bottom position of the isolated pattern, and sequentially extracts the baseline position while referring to the bottom positions of the preceding and following isolated patterns.

Various methods of extracting the baseline position are known, but in this embodiment, the extraction method is executed according to the procedure shown in the flowchart of FIG.

In other words, the bottom part [P] of the isolated pattern (current pattern) of interest. , the bottom position P of each of the 8 patterns before and after
±1. P±2. Examine P±3 (block 51).

Total of 7 patterns of bottom position P1 (i=o, ±1.±2.±
3) On the back of Q, find one bottom position that is statistically the median value (
Median value) PM is determined (block 51).

Next, the bottom position P of the current pattern. and PIr2. PM
+2 is determined (block 52).

Note that if the current pattern is at the beginning or end of a character line, the patterns before or after it are insufficient. In this case, blocks 50 and 51 are created by excluding the missing putter 7.
, 52 are performed.

If the condition of P -2≦Po≦PM+2 is satisfied in the determination at block 52, the process proceeds to block 53, and if the condition is not satisfied, the process proceeds to block 54. In other words, if the bottom position P of the current pattern is within the range of two pixels above and below the median PM, then the bottom position P. is set as the baseline position of the current pattern. Otherwise, the baseline position of the previous pattern (this is held in a register etc.)
is set as the baseline position of the current pattern.

In this way7. . . . The extracted baseline position and bottom position (Po) are sent to the character cutout determination section 6. Note that a buffer is provided in the character cutout determining section 6 to temporarily store data of a predetermined number of isolated patterns, bottom positions, and baseline positions.

In the case of "proportional", the separated character pattern candidate detection unit 10 operates in synchronization with the isolated pattern extraction operation, and refers to the projection width (pattern width) and the projection center interval (pattern interval) in the projection storage unit 8. , to detect separated character pattern candidates. In other words, if the pattern width is narrower than the predetermined value 2
If two isolated patterns are successively cut out at a pattern interval narrower than a predetermined value, the set of isolated patterns is determined to be a separated character pattern candidate, and the character cutout determination unit 6 is notified.

The character cutout determination unit 6 uses the printing pitch identification unit 9 to determine [inch I
If it is determined that the fixed printing pitch is OJ or 12J inches, it is determined whether the width of the isolated pattern input from the isolated pattern cutting unit 5 is appropriate for the corresponding printing pitch, and if it is appropriate, the isolated pattern is finally It is extracted as one character pattern and output to the outside together with the corresponding baseline position data.

If the pattern width is too large, it is determined that it is a connected pattern of two or more characters, the input isolated pattern is forcibly divided based on the corresponding printing pitch, and finally cut out as a pattern of two or more characters.

In the case of proportional printing, unless the input isolated pattern is a dispersed character or pattern candidate, the character cutting determination section 6 ultimately cuts it out as a single character pattern. However, for a set of separated character/turn candidates and isolated patterns determined by the separated character pattern candidate detection unit 10, the bottom position and baseline position of each isolated turn (both are given by the baseline extraction unit 11, Check the positional relationship with the internal] If the conditions are not met, each isolated pattern in the set of isolated patterns is finally cut out as a separate character pattern.

For example, suppose that a character line of proportional printing including a character string as shown in FIG. 8(a) is read. The double quotation mark after the character string 1'-TAXJ (the same goes for the previous one) is cut out by the isolated pattern cutting section 5 as two consecutive isolated patterns. This double coach
/Yonal's enlarged pattern is as shown in Figure 8 (b), and the pattern width of each isolated pattern is 5 (pixel number conversion value)
Since the pattern interval is also narrow at 7, it is determined to be a separated character pattern candidate. Therefore, the character cutout determination unit 6 examines the difference between the bottom position of each isolated pattern and the baseline position. In this case, the bottom positions of both isolated patterns are sufficiently higher than the baseline position, and the difference therebetween is greater than the determination threshold value TH=5. Therefore, this separated character pattern candidate is determined to be a double coach pattern, which is a separated character pattern, and is cut out as one character pattern.

On the other hand, the set of isolated patterns (Figure 2 (B) ('- shows the enlarged pattern) of the period "・" and 1 comma r = J after the character string "Ltd" is also a putter, ) 1'1 width and Since the pattern interval is narrow, it is determined to be a separated character pattern candidate. However, the difference between the bottom position of each isolated pattern and the baseline position is 0. -4, the determination threshold T I (because they are smaller, each character is determined to be a separate *turn and is cut out as a turn).

Effects The present invention, as described in detail in L.J. 2, can correctly cut out separated characters and patterns such as double coach John as one character from a proportionally printed or academic award image.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an inventive example, FIG. 2 is a flowchart of baseline extraction, and FIG. 3 is a diagram for explaining separated character pattern candidates and their extraction. 4...Projection extraction part, 6...Character cutting out part, 7...
・Row image storage unit, 8... Projection storage unit, 9 ・Stamp pitch relation unit, 10... Separation character ・(Turn information detection unit, 1
1...Penis line extraction part... □・[. Figure 2 Figure 3-48:

Claims (1)

[Claims] (1) A character cutting method that creates a projection of image data on a character line in a direction perpendicular to the character line and cuts out an isolated pattern corresponding to the range of each isolated projection as a single character pattern; A set of adjacent isolated patterns that satisfy predetermined conditions for pattern spacing and pattern spacing is detected as a separated character pattern candidate, and the positional relationship between the bottom position of each isolated pattern in each detected separated character pattern candidate and the baseline position is determined. A character cutting method characterized in that it is determined whether a separated sentence or a character pattern candidate is one separated character pattern, and a separated character pattern candidate for which the judgment result is affirmative is cut out as one separated character pattern.