JPS5953983A - Detecting and segmenting method of character - Google Patents

Abstract

PURPOSE:To set a singular point of a cross-cut as a segmenting point, to make a feature of the singular point clear, and to segment easily adjacent characters, by detecting a contact point of the adjacent characters, from a space between the adjacent characters, and an inspecting area which is set so as to cover its circumference. CONSTITUTION:A character picture pattern P read by a photoelectric converting part is stored temporarily in a picture memory 1, and the pattern P is provided simultaneously to a projection accumulating part 7. Subsequently, character position information provided in advance by a control part 6 is compared with projection information from the accumulating part 7, and whether segmenting can be executed by the projection itself or not is decided. In this case, in principle, a separating point of the projection in an area 12 between character frames 10, 11 is segmented. Also, in case when the separating point does not exist, an inspecting area 13 is set by the control part 6, the area 13 is scanned by the control part 6, and a singular point is derived. Subsequently, a singular point of a cross-cut is set as a segmenting point by use of a character recognizing part 4, a normalizing circuit 3, a picture memory 2, etc., and adjacent characters are segmented easily.

Description

[Detailed Description of the Invention] [Technical Field of the Invention] The present invention provides a method for detecting and cutting out characters that are slightly falling in an optical character reading device, and a method for detecting and cutting out characters that are slightly dropped in an optical character reading device, and a method for detecting and cutting out characters that are slightly falling in an optical character reading device, in which characters that are adjacent to each other are in contact with each other due to a force or force. The present invention relates to a cutting method that can be used to cut out images correctly in such cases.

[Technical background of the invention and its discontinuities]

Some of the characters written on the form may be misaligned, the text may be written outside the text frame, or adjacent characters may come into contact with each other. In cases where this is not possible, conventional methods have been used as follows.

■ Forcibly cut out at the center position between adjacent characters.

■ Reject characters that touch.

■ Even if the characters are in contact as a result of calculating the projection, trace the edge of the character line in the λ dimension and check whether it is possible to cut it out.

[Yes] Create a pistogram of black pixel (h) that has a part corresponding to the gap between adjacent characters, and cut out the part with small honeycomb product value.

Among the above methods, method (2) has a relatively good viscosity, but depending on the state of contact, there is a risk that the cutting position may be incorrectly recognized.

[Purpose of the invention]

Therefore, in the present invention, the # edges and edges of adjacent characters touch,
An object of the present invention is to provide a cutting method that can accurately cut out even when cutting is difficult using the method described above.

[Structure of the invention]

In order to achieve the above object, the character detection and extraction method according to the present invention has the following characteristics: (aJ character image obtained by photoelectric conversion) (obtains the projection of a turn). From the projection, it is determined whether there is contact between adjacent character image patterns.

(1)) If there is contact as a result of the 8 words, an inspection area is set that covers the part corresponding to the open space between the @closing cell image patterns. In addition, if there is no contact, the skin can be cut out at the separation point. Determine by voice surveying the outer circumference. Normally, the ice detection area 9 is set in a rectangular shape.

(d) Trace the character line that enters beyond the edges of the inspection area that exists in the prepayment direction of the character image pattern, that is, the left and right sides perpendicular to the character arrangement direction, and determine whether the character line that enters the left and right sides is Find out whether or not to cut it across the board.

+e) Paying attention to the incoming character line that crosses it, find the singular point that exists in the character line. The obtained points are also different points and are used as cutting points.

〔Effect of the invention〕

According to the present invention having the above-mentioned configuration, the contact point of the WIC matching characters can be detected by covering the gap between adjacent characters (or between the character entry frames) and the surrounding area. Since the singular point of the cross section of the cypress is used as the cutting point, if the characteristic of the singular point is clear, it can be cut out without fail. The invention will be described in detail based on illustrative embodiments. FIG. 1 shows the main parts of an optical character reader (OCR) to which a character detection and extraction method according to the present invention is applied.

First, the case of handwritten characters as shown in FIG. 1 will be explained in detail. In Fig. 1, a character image pattern P read by a photoelectric conversion unit (not shown) is sent to an image memory and temporarily stored. It is possible to store as many character image patterns as possible.

On the other hand, the one-character image pattern P is also sent to the projection unit 7, and the projection of the one-character image pattern P is accumulated.

Next, the control unit 6 compares the pre-given Yuji position information with the obtained projection information, and determines whether or not the projection itself can be used for extraction. At this time, as shown in FIG. 1, even if a part of the character extends from one or both of the character frames io and // to the area between the character frames 10 and 71, the extended area / , 2, if there is a separation point in the projection, the break is, in principle, the cutout point. Furthermore, the area/,! When there is noise in the image or there are multiple projection blocks, the process is done using a known method, such as by comparing the area of the noise area with a predetermined reference value. .

On the other hand, as shown in FIG.
li! Cases where the F clitic character has a small letter FU/Kakuto)
, the control unit t sets the inspection area 13. This inspection area /3 is for obtaining information for determining the cutting point, and the cutting mark i// is used to obtain information for determining the cutting point. 1.

The iP B is determined by the control unit 6 according to the width of the character frame and the writing condition of the person filling it in (the width is the printing accuracy in the case of printing), and must be determined in a regular manner. I can't do it, and it's a little tc at least character frame 10, I
10 opposite sides/y, , t! ; The distance of n6, i.e., IP area/, is thicker than the IP A of 7.

Inspection area/, like this? When , the control 1 part 6 scans the four sides of the inspection area /3, and the character frames io and ii
Opposing sides of /'7, /! 5, scan area/, 2
, Check the position where the character line enters into /3 and its end. In the example of Fig. 2, these entry character lines are ■, ■, and ■.
, ■Explain the part.

Once the incoming character line is specified, singular points existing in the incoming character line are determined by scanning or tracing the incoming character line. A singular point is a point of inflection or refraction. In the example of FIG. 2, the point marked ■ becomes the singular point. The way to find the singular point is to track the line ripening, assign the direction of line H to one of the g-direction vectors at certain minute intervals, memorize the change, and later distinguish it from noise from the memorized contents. Judge whether there is a change in degree or not, and then write the letter f, 4.
A general method for finding the edge sign: ■ Scan the entire region /, 2, collect the shift 1j + distance of the change point of the black and white level for each scan, and calculate the differential value (i.e. , travel distance) changes to a bluff or negative value. Other commonly known methods may also be used.

Once a singular point is found, a control signal is sent to the p-team memory/to cut out a character pattern using the singular point as an extraction point. Based on this control signal, the r or / character (e.g., rust, ! in the figure) is cut out at the point (■) in Figure 2, and one character is stored in the accumulator memory 2. . After that, the well-known 1
・It will be dealt with according to the law.

That is, the normalization circuit 3 recognizes the LE-standardized vomiting and the character shift 1 recognition section j recognizes it. j is the memory in which the old pattern is stored.

Next, a case where the characters on the form are printed characters such as printed letters will be described. Figure 3 also shows an example where the male lines are touching. Serif refers to the 11" part attached to the upper and lower edges of the bond, facing cedar, to cover it.

In such a case, if the conventional method uses the histogram of @
If you draw out a line at the point where two trees intersect at right angles, and make a long line or a slanted line vertically as a grinding process, then draw a letter line on the side of it in the bag of However, as shown in Figure 1, it may be difficult to detect a singular point.In such a case, the control section Either the rejection is performed according to the information provided in advance, or forced separation is performed at the virtual center position.

[Brief explanation of the drawing]

1. A block diagram showing the main part of OO'H in which the character cutting method according to the present invention is reversely applied. Figures 1 and 2 are explanatory diagrams showing an example when the target character is handwritten. Figure 1 is an explanatory diagram showing an example of a case where the target character is printed, and Figure 1 is an explanatory diagram illustrating an example in which it is difficult to detect the abnormality when the target character is printed.・Character image pattern, l...field image, ko・
・・Storage memory for 1 character, 3・・Positive digitization circuit, ≠・
・Character recognition department! ... Word W pattern, 6... Control part, 7... Projection V tangent part, 10, //... Character frame, /-! ...9th area, /,?・Control area.

Claims (1)

[Claims] In a method of extracting each character image by detecting the boundaries between adjacent character image patterns among character image patterns obtained by photoelectric conversion, The presence or absence of contact between character image patterns is checked, and if there is contact, an inspection area is set that covers the part corresponding to the gap between adjacent character image patterns, and character lines entering the inspection area from outside are checked. The number and its entry position are determined by scanning the outer periphery of the inspection area, and the character line entering the inspection area is traced to determine whether the entry character line cuts through the inspection area. , find the singular point of the flame and the product line in the case of W1,
A character detection and cutting method characterized in that the 4'\I point is set as the Mi extraction point.