JPS63136179A - Character segmenting device - Google Patents

Abstract

PURPOSE:To accurately segment characters even though a document is titled by using a background area, a circumscribed frame area and a center area to form a character segmenting window and extracting various feature values. CONSTITUTION:A character segmenting window W consists of a center area A1 and a circumscribed area A2 and a backgound area A3. A character detecting part 4C reads binary data in a reduction memory 4B out of the window addresses X and Y of an address generating part 4D which produces the window W. Then the part 4C counts the character pattern area value within the area A3 and each character pattern area of a square area in a circumscribed frame and secures projections of each character pattern in the X and Y directions in the area A1 for counting the projection length in each direction. Then each obtained value is compared with the set allowable value and the presence/ absence of a character pattern is decided. When the coincidence conditions are satisfied, the position coordinates are stored in a character position memory 4E.

Description

[Detailed Description of the Invention] [Industrial Application Field] The invention is a method for arranging characters in a document using a character cutting window created based on the external shape characteristics of the characters (character size is constant). The present invention relates to a character extraction device for extracting characters 't-1 from a character string.

[Conventional technology]

Generally, as this type of character extraction method, a method using projection is known.

[Problem that the invention seeks to solve]

However, this method has the disadvantage that if the document is input at an angle, the characters cannot be accurately cut out. For this reason, a method of detecting the inclination of the character string and performing rotation conversion according to the inclination may be considered, but there is a problem that the borrowing is complicated and expensive.

Therefore, it is an object of the present invention to provide a character extraction device that can accurately extract characters even when a document is input at an angle.

〔problem? Means to solve]

For the binarized image obtained by capturing a character string and binarizing it pixel by pixel, three types of character extraction windows consisting of a center area, a circumscribing frame area, and a background area according to the size of one character are placed at the center of the image. window generation means that generates a window while sequentially moving the position; area calculation means that calculates the character pattern surface 8t in the background area and circumscribed frame area; and projection length calculation means that calculates the projected length of the character pattern for the central area. and a determining means for comparing and determining the area value and projected length calculation with predetermined settings, respectively, and determining the center position of each sum when Asahi satisfies the predetermined relationship with respect to the set value. The character cutting position is determined using it.

[Effect]

Create a character extraction window that corresponds to the character size in the document, calculate the following feature values for the character pattern in that window, determine whether the character pattern overlaps the window accurately, and determine the character extraction position. Determine a.

(1) Character pattern area in the background around the character: AB
^ (2) Character pattern area of the four sides of the character: AT, AB, AL, AR (3) Projection length within the character frame (horizontal and vertical): LPW/LPH In other words, when the window and character pattern overlap correctly, the background area A The sake is zero, and the quadrilateral area is A1 (top).

AB (bottom) l AL (left) #AR (right) is a predetermined value that is not zero, and the horizontal and vertical projection length Lpw t
Since the LPH should match the character width Wo and the width Ho, the character cutting position [e is determined depending on whether these conditions are satisfied or not.

In other words, the position of a character pattern can be determined based on the following conditions.

(a) Background area: An? 0 (b) Four sides MR: min (AT, AB, AL
JAR)≧ALOWER(C) Projection length: LPw, WO
, LPHtH8 Then, the character extraction window is sequentially moved over the 2111 data of the document to search for a position that satisfies the character pattern determination condition. Thereby, the position @ of a character pattern of a predetermined size can be detected.

Here, as a method of moving the cutting window, for example, when the quadrilateral area and the projected length are zero, the window is skipped by half a character, and when a character is cut out, the window is skipped by one character. Try to move faster.

In short, since the position at of an isolated character pattern is measured using the cutting window, accurate cutting can be performed even if the character arrangement is tilted.

〔Example〕

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

In the figure, 1 is a document, 2 is an image scanner such as a television camera, 3 is an image memory, and 4 is a reduction circuit 4A.
, a reduced memo IJ4B, a character detection section 4C1, a cutout window address generation section 4D, a character position memory (latch circuit) 4B, a character size setting section 4F, and the like.

Document 1 is converted into a 21 mouse image by image scanner 2 and stored in image memory 3. Since the original image data is too large to cut out characters, it is reduced to an appropriate size here. That is, the original data is reduced by the reduction circuit 4A and stored in the reduction memory 4B. Of course, these circuits can be omitted if they are not necessary. The character size (width Wo pixels, height Ha pixels) in the document is known in advance, and can be set in the character size setting section 4F from an external device such as an OCR (optical character reader). The font size setting section 4F sets the address generation section 4 of the cutout window based on the set font size.
For D, specify the size of the extraction window.

The cutout window is moved over the reduced image on the reduced memo IJJB, and the character detection unit 4C determines whether it accurately overlaps the character pattern. If the window and character are at the same position, the position coordinates are stored in character position memo 1J4EK. Based on the character position information, the external device cuts out the character pattern data 't-1 characters from the image memory 3 and processes the data.

That is, the document Ha input by the image scanner 2
t, for example, as shown in FIG. 2, is stored in the image memory 3 as 2 volumes of data. At this time, the font size in the document is limited to several types. Also, if the size is large if you leave image memory 3 as is, you can move it to 3rd row.
A two-dimensional local memory 5 as shown in (A) and a two-dimensional local memory 5 as shown in FIG.
3×3 pixels are reduced to 1 pixel using an OR circuit 6 as shown in (b).

FIG. 4 shows the shape of the character cutting window W determined from the character size (W(,XH,): see FIG. 2). The center coordinates (Xc=Yc) of the window change as the window moves. The cutting window W is composed of three rectangular areas. In other words, the width Wc
These include a central area A1 of x height Hc, a circumscribed frame area A2 of width Ws x height H, and a background area A3 of width WB x height HB. Here, the central area A1 is rectangular, but the circumscribing frame area A2 and the background area A3 are band-shaped areas. The size of each area is determined from the character size (WQXHQ: see FIG. 2), and the circumscribed frame area A2 is determined from the maximum and minimum values of the character pattern as follows.

WS= max (W) , W(=min (W)Hs
-max(H), f(c-min(Hl) However, W and H mean the width and height actually measured for a character pattern of the same character size. Also, the size of the background area A3 is the space between characters 5o Determine from (SOW/SOH).

WB-W8+SowHB-H8+SoH In other words, the area around the characters where no character pattern can exist is the background area A.
Introduced as 3. Furthermore, the central area A of the character pattern
Since the character pattern exists in l, its projected length directly becomes the width Wc and height Hc of the central area.

Using the cutout window W as described above, the position of the character pattern is detected as follows.

FIG. 5 shows a concrete sequence of the character detection section 4C. The address generation unit 4D that generates the character extraction window w2 generates a window (left apex (Xs)) as shown in FIG.

Ys) and the right bottom point (X'12 t YE )) area address (x, y) and area signal (Al
s A2 # A3) Output. Binary data C in the reduced memory 4B from the address (X, Y) of the address generator 4D.

is read out, and the background area counter circuit 4 is read out using the background area signal ■, the circumscribed frame area signal ■, and the center area signal ■.
1. Select the quadrilateral area counter circuit 42 and the projection length counting circuit 43, respectively. The background area counter circuit 41 is
Count the characters butter/area AB in the background area. The four side area counter circuit 42 counts the area of each character pattern for each of the four side areas within the circumscribed frame. The four side areas (circumscribed frame areas) are the top side, bottom side, left side, and right side, and each area overlaps at the four corners. The areas of the four side regions are represented by AT, AB, AL, and AR, respectively. Here, the four-side area signal is ■A# nB *IIL r
Four IIHs are output from the address generator 4D.

Further, the projection circuit 43 projects the character pattern in each of the X and Y directions within the central region, and counts the projection length LPW I LPHe in each direction.

For each 8 obtained in this way, first, the background area AB
A is the allowable background area from the setting device 47 + m (upper limit)
ABU is compared with the comparator 44. Then, the comparison result Jx't-is expressed as in the following equation.

In addition, the quadrilateral area AT, AB, AL, human, is the value of the allowable quadrilateral surface B' from the setting device 48 (lower limit A) ATL
#ABL #ALL t ARL and comparator 45
are compared. The comparison result J2 is expressed by the following equation.

or <ALL or AR<ARL Furthermore, the projection length LPW I LP□ is compared with the allowable projection lengths LwL and LHL from the setting device 49. Comparison result J
3 is expressed as follows.

These comparison results J1 eJ2 vJ3 are AND circuit 4
0, it is determined whether there is a character pattern or not, and if the matching condition is satisfied, the position coordinates are stored in the memory 4E.
Store in.

Since the character extraction window is defined as described above and character position detection is determined based on the three conditions, character patterns can be detected regardless of the inclination of the document. [Effects of the Invention] This invention According to the authors, the character extraction window is configured with three types of areas: background area, circumscribing frame area, and center area, and each type of feature quantity is extracted, so that not only can isolated character patterns be detected, but also it can be This provides the advantage of being able to cut out individual characters even if there is a skew.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the present invention, and FIG. 2 shows 2 WL 'JBJ fl! stored in the image memory. FIG. 3 is an explanatory diagram for explaining an example of data; FIG. 3 is an explanatory diagram for explaining a specific example of a reduction circuit; FIG. 4 is an explanatory diagram for explaining the entire character extraction window used in this invention; FIG. 5 is a block diagram showing a specific example of the character detection section shown in FIG. 1. Code explanation 1...Document, 2...Image scanner, 3...Image memory, 4...Character cutting circuit, 5...2 Dimensional local memory, 6...
...OR circuit, 4A...reduced tJz circuit,
4B: Reduced memory, 4C: Character detection section, 4D: Address generation section, 4E:
・Character position memory (latch circuit), 40...
AND circuit, 41...Area counter circuit for background area, 42.Area counter circuit for music/four-side area, 43.
- Song projection circuit, 44-46... comparator,
47-49...Setting device, W...Window, Al...Central area, A2...
Circumscribed frame area, A3...Background area. Agent Patent Attorney Akio Namiki Agent Patent Attorney Kiyoo Matsuzaki +vh 14 Figure

Claims (1)

[Claims] A character extraction device for extracting characters one by one from a character string recorded in a document, which extracts one character from a binary image obtained by imaging and binarizing a character string. Consisting of a center area, a circumscribing frame area, and a background area according to the font size.
window generation means for generating different types of character cutting windows while sequentially moving their center positions; area calculation means for calculating character pattern areas in each area for the background area and the circumscribed frame area; and the center Projection length calculation means for calculating the projected length of a character pattern for a region; Judgment means for comparing and determining the area calculation value and the projection length calculation value, respectively, with predetermined setting values, and these calculated values are determined by the respective settings. A character cutting device characterized in that a character cutting position is set at a center position when each value satisfies a predetermined relationship.