JPS62200490A - Optical character reader - Google Patents

Abstract

PURPOSE:To recognize a character even when invasion and projection occur between adjoining rows by cutting out the character pattern of one character unit based upon the reading sight information of respective characters and reading the character. CONSTITUTION:The variable density information of a slip is quantized by a photoelectric converting part 1, converted to binary digital pattern data and inputted to a pattern buffer 2. The pattern buffer 2 stores the pattern data of at least one row of the character row. The stored pattern data are supplied to an upper lower direction reading sight detecting part 3 and the detecting data of an upper lower direction sight scope are supplied from the upper lower direction reading sight detecting part 3 to a character cutting-out part 4. A character cutting-out part 4 executes the character cutting-out from the pattern data based upon the detecting data of the upper lower direction sight scope and supplies the cut-out character pattern to a recognizing processing part 5. The recognition processing part 5 inputs the character pattern, executes the character recognition and outputs a candidate category to the external part as the character recognizing result.

Description

[Detailed description of the invention] (Purpose of the invention) (Industrial application field) The present invention relates to an optical character reading device.

(Prior Art) Conventionally, an optical character reading device (hereinafter referred to as OCR) receives format information from a host in order to read a target character line from among multiple character lines written on a form. The character line position is then set, and an area of a predetermined width is defined with respect to the center position of the character line to read the character line.

That is, the shading information on the form is photoelectrically converted in a preset range centered on the center position of the character line, and the obtained pattern data is stored in a pattern buffer. Character extraction and character recognition are performed on the premise that the pattern data stored in the pattern buffer does not intrude from adjacent character lines or protrude into adjacent character lines.

(Problem to be solved by the invention) However, when the line spacing between character lines is small,
In reality, parts of characters protrude from a character line, or parts of other characters invade from adjacent lines, resulting in frequent rejections and misreadings.

For this reason, conventionally, documents read for OCR have had to have a sufficiently wide character line pitch, and compared to the document size, the number of written characters is smaller than that of a normal sparrow (which reduces the efficiency of character processing by OCR). There was a problem.

The present invention has been made in consideration of the above-mentioned circumstances of the prior art.
Or, to provide an optical character reading device that can correctly cut out characters even if characters from another character line invade from an adjacent character line, and can increase the number of characters per unit area that can be written on a form. The purpose is to

[Structure of the invention]

(Means for Solving the Problems) In order to achieve the above object, the present invention includes a conversion unit that converts shading information on a form into pattern data, and at least one line of characters written on the form. a pattern buffer for storing the pattern data; a reading field detecting section for scanning the pattern data stored in the pattern buffer to detect a reading field of view in at least one direction of each character; A character cutting section cuts out a character pattern in units of characters based on the detection information of the Fi!
It is characterized by being equipped with a recognition section that recognizes the

(Function) According to the present invention, since the optical character reading device is configured as described above, it is possible to determine the correct reading field of view even if the characters protrude.

(Embodiment) An optical character reading device according to an embodiment of the present invention will be described below with reference to FIGS. 1 to 3 of the drawings.

In this embodiment, a case will be described in which a document with a so-called horizontal character arrangement, in which written characters are arranged horizontally with respect to the @ slip, is read.

As shown in FIG. 1, the shading information of the form is photoelectrically converted by a photoelectric converter 1, converted into a-digital data, converted into, for example, binary digital pattern data, and inputted into a pattern buffer 2. The pattern buffer 2 stores pattern data for at least one character line. The reason why at least one character line of the input cover pattern data is stored in the pattern buffer 2 is because this embodiment is designed to read characters in the target character line (hereinafter referred to as a reading line). This is because we are focusing on cases where part of a character intrudes from an adjacent line or part of a character protrudes from a read line into an adjacent line. The pattern data stored in the pattern buffer 2 is supplied to the vertical reading field of view detection section 3, and the detection data of the vertical direction reading field of view is supplied from the vertical direction reading field of view detection section 3 to the character cutting section 4. Here, the vertical direction refers to the vertical direction perpendicular to the character lines (horizontal lines) arranged in the horizontal direction of the form, so to speak, the vertical direction within the plane of the paper. The character cutting section 4 cuts out characters from the pattern data based on the detection data of the vertical visual field range, and supplies the cut out character pattern to the VT recognition processing section 5. ! ! ! ! ! The recognition processing unit 5 inputs a character pattern, performs character recognition, and outputs a candidate category to the outside as a character III result.

The vertical reading field of view detecting section 3 scans the pattern data stored in the pattern buffer 2 and detects the vertical field of view range of the read line characters according to the following procedure. First, the vertical reading field of view detection unit 3 scans the pattern data and detects whether or not the characters in the read line protrude into the vertically adjacent lines (hereinafter referred to as protrusion detection). Next, in the same manner, it is detected whether or not there is an intrusion into the reading line from the upper and lower adjacent lines (hereinafter referred to as intrusion detection). If an intrusion is detected, the intruded portion is deleted from the pattern data (hereinafter referred to as deletion). The vertical viewing range is detected as described above, and the detected data is supplied to the character cutting section 4. Based on this detection data, the character cutting section 4 cuts out characters from the pattern data using a known technique, and supplies the cut out character pattern to the recognition processing section 5. The recognition processing unit 5 recognizes the input character pattern using a known technique and outputs a recognized character code.

Hereinafter, the operation of the vertical reading field of view detection section 3 will be explained in more detail with reference to FIGS. 2 and 3. Figure 2 is an explanatory diagram showing an enlarged view of a part of the reading line of the form to be read, and shows a case where the number "2" is written on the reading line and a diagonal line intrudes from the adjacent upper line. This is an example. That is,
This means that the pattern data stored in the pattern buffer 2 has an intrusion into the read line from an adjacent line.

The vertical reading field of view detection unit 3 performs protrusion detection, intrusion detection, and erasing operations as follows. In addition, in the following explanation, we will explain the case where the upper visual field range is detected when there is intrusion from the adjacent row above the reading row, or when there is protrusion from the reading row into the adjacent row above, but Since the relationship between the adjacent row and the reading row is the same, a description of the detection of the lower visual field range will be omitted.

In the first step, the vertical reading field of view detection section performs protrusion detection.

First, for the first character in the reading line (the number "2" in Figure 2), the upper boundary of the temporary visual field is set as shown in Figure 2.
Set Y=MSKU, X=MSKL to MSKR.

Here, X and Y are X-axis and Y-axis address coordinates provided for the pattern data. Furthermore, MΔSKO is the Y coordinate of the center position of the reading line, and M S K U G,
t M A S K These are tentative visual field upper end coordinates that are assumed to be separated by a preset width from C. Similarly, M
SKL is a temporary character left end coordinate determined by the character pitch information, and MSKR is a temporary character right end coordinate determined by the character pitch information. Note that the center position of the reading line for setting the MASKC, the character pitch information for setting the MSKL, and MSKR are information included in the format information provided from the host to the OCR. For upward protrusion detection, set x to MSKL with Y-MSKUl, :
This is performed by moving from MSKRl to : and performing a black bit search in the right direction in FIG. If no black bit is detected, the temporarily set visual field upper end coordinate MSKLJ is used as the visual field upper end coordinate MASKU. This is because at this time, the character is located below the visual field upper end coordinate MASKU.

However, if a black bit is detected in the above black bit search, it means that there is an intrusion from an upper adjacent row or an overflow to an upper adjacent row.

In this case, the process moves to the second step, in which it is determined whether the object is an intrusion or an extrusion.

First, when a black bit is detected in the first step, contour tracing is performed clockwise from the black bit detection position (Pl in FIG. 2). In the middle of tracking, U is the upper boundary Y coordinate.
If LMT is exceeded (P2 in Figure 2), it is determined that the intrusion has occurred from the upper adjacent row, and the intrusion flag UIV is set.
Set F to stop tracking and proceed to the third step described below.

Here, ULMT is the upper boundary Y coordinate set upward by a predetermined width from the intermediate position between the center position of the upper adjacent row and the reading line center position, and is a boundary coordinate indicating an upwardly protruding boundary. It is what it is. When it is determined that an intrusion has occurred, the process moves to the third step. In the third step, the position U (Pl in Fig. 2) where contour tracking is started is
Perform white bit search from . If a white bit is detected (P in FIG. 2), the black bit search is performed again from that point P3.

When a black bit is detected again by this black bit search, the contour is traced six times from the detected position (P4 in FIG. 2). If the contour tracking returns to the MSKU without reaching the ULMT (P5 in Figure 2), it is determined that an upward protrusion has occurred in the read line character, and the upper end coordinates of this protruding portion are determined. demand. Then, the black bit search is continued again in the direction of MSKR from the position returned to MSKU (P5 in FIG. 2).

If it is determined in the second step that protrusion has occurred, the process moves to the fourth step. In the fourth step, the coordinate of the uppermost end is determined from among the upper end coordinates of each protruding portion, and the coordinate 1 bit above the minimum value of the uppermost coordinate is set as the visual field upper end coordinate MASKU.

It is determined that there is an intrusion in the second step, and the intrusion flag UI is displayed.
If VF is set, the process moves to the fifth step and erases the intruding block.

On the other hand, if it is determined in the above step that there is no intrusion or protrusion, the vertical visual field detection operation ends.

Now, in the fifth step, intrusion detection and erasure from the upper adjacent row are performed. The operation of the fifth step will be explained below with reference to FIG. FIG. 3 is an explanatory diagram illustrating how the reading line is invaded.

(a) shows the case where the intruding object is separated from the characters on the reading line, and (b) shows the case where the intruding object is in contact with the characters on the reading line.

First, for the first character in the reading line (number "2" in Figure 3), set the upper boundary of the visual field as Y=MASKU,
! : Set L, X=MASKL-MASKR. Then, at Y=MASKU, change X from MASKL to MAS
Move to KR and perform a black bit search to the right in Figure 3. Note that what is detected in this black bit search is only the intrusion block as a result of the second step. When a black bit is detected, its position (in Fig. 3(a), P
Contour tracking is performed counterclockwise from l). If the upper intrusion boundary Y coordinate U I LMT is reached in the middle of tracking, or if the tracking length exceeds the threshold CLMT set in advance for the tracking length, intrusion from the upper adjacent line makes accurate reading line character recognition impossible. It is assumed that this is the case, the reject flag is set, and the operation ends. Here, U I LMT is
These are boundary coordinates that are provided a predetermined width above the center position of the reading line and indicate the permissible limit of intrusion from above. On the other hand, even with contour tracking, the Y coordinate does not reach UILMT, or the tracking length is within CLMT,
And if the Y coordinate returns to MASKU (Fig. 3 (a)
Medium, P2), further contour tracking is continued. If the Y coordinate reaches LJLMT without returning to MASKU again (P3 in FIG. 3(a)), the intrusion from above is regarded as a separate intrusion block, and the intrusion portion is erased.

However, by continuing the contour tracing from P2 above, the Y coordinate becomes L.
If it returns to MASKU again without reaching JLMT (P' in Figure 3 (b)), or if the tracking length exceeds the CLMT, or if it reaches U I LMT (Figure 3 ( In b), P4), the characters on the reading line (
It is assumed that the number "2" in FIG. 3(b) is in contact with the intrusion (hatched in FIG. 3(b)), a reject flag is set, and the operation is terminated.

Now, in the above case, if it is determined that the block has invaded from above, and the intrusion is slight and is separated from the read line characters (see Figure 3 (a)), the outline of P-P2 can be traced. The portion surrounded by the point and the straight line P1P2 is erased using a known method. As described above, the vertical reading field of view detection section 3 performs protrusion detection, intrusion detection, or erasure, and the character cutting section 4 performs character cutting processing within the vertical field of view.

Note that the present invention is not limited to the above embodiments.

For example, the reading field of view may be detected only in one direction, up or down. Also, the entries on the form are not limited to the horizontal character arrangement, but can also be written in the vertical character arrangement.

〔Effect of the invention〕

As described above, according to the present invention, pattern data obtained from a form is scanned, the reading field of view of each character in a certain direction is detected, and character patterns are cut out for each character based on the reading field of view information. Since the characters are read, even if the line spacing bits written on the form are narrow and there is intrusion or protrusion between adjacent lines, the characters can be recognized.

As a result, the number of characters that can be written on a form per unit area can be increased, and the processing efficiency of character reading by OCR can be greatly improved.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an optical character reading device according to an embodiment of the present invention, and FIGS. 2 and 3 are explanatory diagrams for explaining the operation of the optical character reading device. 1... Photoelectric conversion unit, 2... Pattern buffer, 3...
. . . Vertical reading field of view detection unit, 4 . . . Character cutting unit, 5 . . . Recognition processing unit.

Claims (1)

[Claims] 1. A conversion unit that converts the shading information on the form into pattern data, and at least one of the characters written on the form.
a pattern buffer that stores the pattern data for a line; a reading field detection section that scans the pattern data stored in the pattern buffer to detect a reading field of view in at least one direction of each character; What is claimed is: 1. An optical character reading device comprising: a character cutting section that cuts out a character pattern in units of characters based on detection information from the section; and a recognition section that recognizes the cut out character pattern.