JPH0199189A - Character segmenting system - Google Patents

Abstract

PURPOSE:To accurately segment character, and to improve the recognition rate by executing searches for vacant null and null dot processings in both up-and-down directions and right-and-left direction from an inner auxiliary frame to an outer auxiliary frame. CONSTITUTION:In image data obtained by scanning an input sheet, the outer auxiliary frame 12 which is generated by extending a logical frame 11 corresponding to a writing frame outward for a prescribed number of dots, as well as the inner auxiliary frame 13 generated by shrinking inward for a prescribed number of dots, are set, and null lines are searched in up-and-down direction and right-and-left direction from the frame 13 to the frame 12. When a null line is searched, the image data from this line to the frame 13 is processed to be null, and image data surrounded by the frame 13 is segmented as the pattern of one character. As a result, even if a part of a character pattern to be recognized overflows from the logical frame 11, the character can be segment without any lacked part only if the character is within the outer auxiliary frame, also, even if the neighboring character intrudes the logical frame, it is eliminated if the intruding is no longer deeper than the inner auxiliary frame, hence an accurate segmenting of a character can be executed.

Description

DETAILED DESCRIPTION OF THE INVENTION (A) Field of Industrial Application The present invention relates to a handwritten character recognition device, and more particularly to a character cutting method for cutting out characters one by one.

(b) Conventional technology In devices that recognize handwritten characters, generally a character entry frame is printed in a dropout color on the input sheet for each character in order to inform the writer of the position and size at which the character should be entered. ing.

Conventionally, in order to cut out characters from image data obtained by scanning an input sheet, a theoretical frame corresponding to the entry frame is set based on timing marks, etc., and the image data surrounded by this theoretical frame is was cut out as a single character pattern. Such technology is known, for example, from Japanese Patent Publication No. 5B-3162.
It is disclosed in Publication No. 8.

(C) Problems to be Solved by the Invention When writing characters by hand, it is impossible to completely avoid protruding from the writing frame or entering adjacent characters, and in this case, it is impossible to completely avoid writing characters outside the writing frame or entering the adjacent characters. This results in a character pattern in which part of the character to be recognized is missing, or a character pattern including part of unnecessary adjacent characters.Therefore, as in the past, image data surrounded by a theoretical frame is If the pattern is cut out as a single character, accurate cutout cannot be performed, resulting in a decrease in the recognition rate.

Also, even if you handwrite the characters so that they do not protrude into the writing frame, if the input sheet itself is tilted, part of the character pattern to be recognized may protrude from within the theoretical frame, or some adjacent characters may This phenomenon occurs, and in this case too, the recognition rate decreases with the conventional extraction method.

(d) Means for Solving the Problems The present invention provides image data obtained by scanning an input sheet with an outer auxiliary frame that extends outward by a predetermined number of dots from the theoretical frame corresponding to the entry frame, and an inner auxiliary frame. An inner auxiliary frame narrowed by a predetermined number of dots is set to the inner auxiliary frame, and a blank line is searched in each direction from the inner auxiliary frame to the outer auxiliary frame. The image data up to the frame is blanked out, and the image data surrounded by the outer auxiliary frame is cut out as a pattern of one character.

(N) Function In the present invention, even if a part of the character pattern to be recognized protrudes from the theoretical frame, as long as it falls within the outer auxiliary frame, the character can be cut out without being lost. Even if adjacent characters enter, as long as they are outside the inner auxiliary frame, these unnecessary adjacent character patterns can be eliminated and accurate character extraction can be performed.

(v) Embodiment Figure 2 is a block diagram showing the overall configuration of a character recognition device that realizes the character extraction method of the present invention, in which (1) the entry frame (2) for each character is printed in dropout color. A black timing mark (3) is placed at the left end of each line on the fully printed input sheet, (4) is the character observation unit that scans the input sheet to obtain black and white binary image data, and (5) is the input sheet ( An image memory 1 which has a capacity corresponding to the plurality of characteristics of 1) and stores the obtained image data. and (6) is a character cutting unit which has a character memory (7) for storing image data for one character and cuts out a character pattern one character at a time from the image data of the image memory (5), The extracted character pattern is sent to the recognition unit (8), where characteristic patterns are extracted and compared with standard patterns stored in advance in the standard pattern memory (9) to perform character recognition. .

Hereinafter, the processing contents of the character cutting section (6) will be explained in detail with reference to the flowcharts of FIG. 1 and FIGS. 3 to 4.

First, in the image data in the image memory (5), the coordinates (To, ■,) of the read timing mark (10) of the line to be recognized are determined. Next, input sheet (1)
Relative distance from timing mark (3) to entry frame (2) 1. , 1. , and the size of the entry frame (2), that is, the width w1 and height, and the coordinates (1
,,T,), the theoretical frame (
11) Find (solid line). Furthermore, a predetermined outer X
Using the direction margin n and the outer y direction margin n, an outer auxiliary frame (12) (dotted chain line) is set that is expanded outward by a predetermined number of dots from the theoretical frame (11), and this outer auxiliary frame Frame (1
Move the image data surrounded by 2> to the character memory (7).

Here, the coordinates of the four points A, B, C, and D of the outer auxiliary frame (12) on the image memory (5) are (T, +1. 1
1. t T, +1. nt) F (T, +1゜1
1+ * IF"IF" nt + My) + (
Tz+1. + nI+ W, , T, +1y-n
t ) + (T t ” 1 x + n 1 +
W, + T, 1y + n, + W, ), and point A is the origin on the character memory (7).

Next, on the character memory (7), using the predetermined inner X-direction margin m A supplementary frame (13) (broken line) is set, and blank line search and blank processing are performed. As detailed in the flowchart of FIG. 4, this search and processing begins with the upper line (0+ ax +
m@l ) ~ (w, +2n+ 1, n, +m
2-1>, the upper line (0,
0) to (%I, +2n, -1,0), check each dot line to see if it is a blank line, and if it is a blank line, check all dot lines from that blank line to the upper line of the outer auxiliary frame. The blank is left blank, and the same process is performed for the lower, right, and left sides of the inner auxiliary frame (13).

The image data in the character memory (7) subjected to the above processing is cut out as a pattern of one character and sent to the recognition unit (8>), and the same processing is repeated for the next character in the same line.

When all the characters in one line have been cut out, the process moves to the next line and starts the same process from detecting the coordinates of the timing mark in that line.

FIG. 5 is a diagram showing an example of actually cutting out characters using the method of the present invention. Similar to FIG. 1, the solid line indicates the theoretical frame (11
), the dashed line is the outer auxiliary frame (12), and the broken line is the inner auxiliary frame (13).

Not only when the characters fit neatly within the theoretical frame (11) as shown in Figure 5(a), but also when the characters fit neatly within the theoretical frame (11) as shown in Figure 5(b).
Some of the letters may protrude from the theoretical frame (11) or
Even if part of the adjacent character falls within the theoretical frame (11) as in C), the character is accurately cut out.

(G) Effects of the Invention According to the present invention, if characters are entered incorrectly or due to the inclination of the input sheet, some of the characters may protrude from the theoretical frame corresponding to the entry frame, or some of the adjacent characters may Even if the parts get stuck in, the characters can be cut out accurately.
Recognition rate improves.

[Brief explanation of the drawing]

FIG. 1 is an explanatory diagram for explaining the cutting method of the present invention,
2nd UgJ is a block diagram showing the overall configuration of a character recognition device using the method of the present invention, FIG. 3 is a flowchart showing the processing contents of the method of the present invention, and FIG. 4 is the main processing contents of the method of the present invention. FIG. 5 is a flowchart illustrating an example of the extraction method of the present invention. (1)...Input sheet, (2)...Entry frame, (3
)...Timing mark, <5)...Image memory, (6)...Character cutting section, (7>...Character memory.

Claims (1)

[Claims] (1) In a character recognition device that uses an input sheet on which a writing frame for writing characters is printed in advance, the image data obtained by scanning the input sheet is added to the outside of the theoretical frame corresponding to the writing frame. Setting an outer auxiliary frame that is expanded by a predetermined number of dots and an inner auxiliary frame that is narrowed inward by a predetermined number of dots,
A search for a blank line is performed in each direction from the inner auxiliary frame to the outer auxiliary frame, and when a blank line is found, the image data from the blank line to the outer auxiliary frame is blanked, and the image data from the outer auxiliary frame is A character cutting method characterized by cutting out image data surrounded by an auxiliary frame as a single character pattern.