JPH01154296A - Character segmenting method - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

Detailed Description of the Invention [Table of Contents] Overview Industrial Application Fields Prior Art Problems to be Solved by the Invention Means for Solving the Problems Effects of the Invention [Summary This paper relates to a character extraction method for cutting out printed characters on paper sheets that have been converted into characters for character recognition, and aims to provide a method that can improve the accuracy of character recognition. The positions of both ends of the character included in the region are detected, and a rectangular cut-out window with a blank space corresponding to both detection positions is formed on the side determined by both detection positions of the characters at both detection positions. It is characterized in that it is set for image data, and the image data within the set frame is extracted as character cutting data.

[Industrial Application Field] The present invention relates to a character cutting method for cutting out printed characters on paper sheets that have been converted into image data.

Characters printed on a sheet of paper can be recognized by optically reading them using an image scanner or the like, cutting out each character from the image data, and comparing the cut out characters with a dictionary.

[Prior Art] Fig. 8 explains a conventional character cutting method. “A”, symbol “′″”, symbol “
~” are in the same figure (E) and (F).

As shown in (G) and (H), they were cut out with square windows inscribed on all sides.

In this way, since the character to be recognized is inscribed in the entire rectangular cut-out window, stable character recognition is always possible.

[Problem that the invention attempts to solve] However, in the conventional case, the same figure (E) is used.

As can be understood from (F), (G), and (H>, the size and position of the characters are not included in the extracted image data at the time of extraction, and only the shape of the characters inscribed in the window is the recognition material. Therefore, for example, (G), (H>
There was a problem in that it became difficult to distinguish between characters, and as a result, errors occurred in character recognition.

In other words, normal documents contain many characters other than kanji, which can be used to obtain sufficient results with approximately square shapes, and non-kanji characters have their contents indicated by their size and position, such as syllables or symbols. In these cases, almost everything will be recognized incorrectly.

The present invention has been made in view of the above-mentioned conventional problems,
The purpose is to provide a character cutting method that makes it possible to accurately recognize phonograms, symbols, etc. that indicate their content by size, position, etc.

[Means for Solving the Problems] In order to achieve the above object, the method according to the present invention is configured as shown in FIG.

First, in step 10 of FIG. 1, the positions of both ends of a character included in one character area are detected from the image data of that area.

In the present invention, the left and right, top and bottom, or both end positions of the characters are detected, and in the figure, both the top and bottom and left and right end positions are detected.

When the positions of both ends of the character are detected in this way, in the next step 12, a rectangular cutting window is set for the image data, with spaces corresponding to both positions provided outside the character at both positions. Ru.

The side where the blank space is provided is determined by both detection positions,
In the figure, a blank space is provided on the side opposite to the character bias side.

In the final step 14, the image data within the window set in step 12 is extracted as character cutting data, and the extracted data is used to recognize the character.

[Function] In the present invention, both detection positions of a character indicate the size and position of the character, and the blanks provided accordingly serve as information on the size and two positions. characters are recognized accurately regardless of their size or position.

[Example] Hereinafter, a preferred example of the method according to the present invention will be described based on the drawings.

FIG. 2 illustrates a system to which the present invention is applied, in which a printed text on a sheet of paper 20 is optically read by a scanner 22, and a computer 24 performs character recognition of the printed text based on the image data.

When performing this character recognition process, each character is cut out as shown in the figure, and FIG. 3 shows the procedure.

In this embodiment, a horizontal section with a character height of one character is first detected for each line, and thereby the area of one character is determined.

The determining action is explained in Fig. 4, in which ".J, r, J, r-J.

rAJ, rJ, rBJ, r, J, rcJ,
"B" is included.

And the character rAJ with the largest vertical dimension in this line
The upper end position cup and the lower end position cdwn are detected and these are set as the upper end position 1up and the lower end position Idwn,
Since the character area is basically a square, these positions 1u
The distance between p and Idwn is set as the height of one character in the lines and its horizontal interval bc-bc.

Next, determine the top and bottom positions of each character (cup.

cdwn) is detected (step 31), and their vertical center position is determined (step 32).

Roughly, the positions (ctop, cbottOm) separated outward by half the character height from the center position are found (step 33), and these are the top and bottom stroke positions 1up.
, Idwn is exceeded (step 34.
35>, then their positions Iup, ldwn
(step 36.37>).

The vertical position (ctop) determined in this way.

cbottom) starts to move inward (step 38), and it is confirmed that one of them has reached the upper end position cup or the lower end position cdown of the character (YES at step 39), or the amount of movement is equal to the character height. When it is confirmed that it has become half (YES in step 40)
, the movement is stopped (step 41), and the upper stop position ctop and lower stop position cdown at that time are recorded as the cutout position of the character in the vertical direction.

Similar processing is performed for the left and right ends (step 43), and after confirmation (step 44, Y
ES), the image data within the rectangular window determined by the storage position is extracted as character cutout data and output (step 45).

FIG. 5 shows the operation of the embodiment, in which characters with the symbol "," are cut out as shown in FIG. 5(A).

In this case, since this character is small and located on the lower left side, its image data is extracted using a window with blank spaces above and to the right of the character, as shown in FIG. 2B.

In the case of "-", this is located at the center in the vertical direction of a single character area, so the image data is extracted in a window with margins above and below the character as shown in FIG. 4C.

As can be understood from the above description, according to this embodiment, when a character is not inscribed on all sides of the character cutout window, the character is small, and the characters in a single character area are biased depending on the position of the space. Since the position is indicated, using this, the symbols r-J, r J, r, J, r, J in FIG. 4 can be recognized accurately and without error.

The same holds true for kana letters, and as a result, it becomes possible to recognize characters in kana and symbols much more accurately than in the past.

However, if the character is small like the symbol "・" shown in Figure 4 and located in the center of the - character area, the character is inscribed in the window as can be understood from Figure 6. Therefore, in some cases, errors may occur in character recognition such that the symbol is recognized as ",".

In this case, it is preferable to set the window so that the left and right sides of the window touch the left and right ends of the character, respectively, in order to perform stable character recognition.

FIG. 7 explains the method. First, the left end position of the character c+eft and the right end position of the character crclht are detected.

Next, their center position cmid is determined, and left and right positions ch I eft and ch i C1ht that coincide with the left and right boundaries bC of the single character area separated by half the character height are determined.

Note that when they exceed the boundary position bc, they are restricted to the boundary position bc.

Further, the positions ch l left and ch r i g ht are moved inward and when they reach both sides of the character "・", their positions C1 eft and cright are moved inward in the same figure (B
>, which determines the left and right sides of the window.

However, if the character is not placed in a symmetrical position, such as the character ",", it will be circumscribed by either the left or right window or the character.

According to this embodiment, since the characters to be cut out are inscribed on two of the four sides of the window, it is possible to perform more accurate character recognition than in the first embodiment.

[Effects of the Invention] As explained above, according to the present invention, it is possible to check the size of the character to be cut out from the blank space of the character cutout window and the position of the character within a single character area, which greatly improves the recognition of Kana characters, symbol characters, etc. It becomes possible to recognize with high accuracy.

It has been confirmed that the recognition rate for only symbols and characters can be improved from about 60% in the past to 95%, and for kana and symbols from 90% to 95%.

[Brief explanation of the drawing]

Fig. 1 is an explanatory diagram of the principles of the invention, Fig. 2 is an explanatory diagram of the system of the embodiment, Fig. 3 is an explanatory diagram of the procedure of the embodiment, Fig. 4 is an explanatory diagram of the single character area determination operation of the embodiment, and Fig. 5 6 is an explanatory diagram of the operation of the embodiment, FIG. 7 is an explanatory diagram of another horizontal cutting method, and FIG. 8 is an explanatory diagram of a conventional example. 10...Both end position detection 12...Cutout window setting 14...Image data extraction Principle explanatory diagram of the invention Fig. 1 System explanatory diagram of the embodiment Fig. 2 <''--Single character area Fig. 5 Other Figure 7 is an explanatory diagram of the left-right direction cutting method.

Claims (1)

[Claims] The positions of both ends of a character included in one character area are detected from the image data of the area (10), and the side determined by both detection positions of the outside of the character at both detection positions is detected according to both detection positions. A rectangular cutting window with a blank space of the same amount is set for the image data (12).
, extracting image data within a setting window as character cutting data (14).