JPH0981679A - Optical character reader - Google Patents

Abstract

PROBLEM TO BE SOLVED: To eliminate reading impossibility and erroneous reading even when a slip is zigzag. SOLUTION: A mechanism part 1 carries and feeds the slip 2-2 corresponding to the instruction of a mechanism control part 5, an optical part 2 performs irradiation by a lamp 2-1 disposed at the read position of the carried slip 2-2 and images on the slip ch are formed through a lens 2-3 on the CCD sensor of a photoelectric conversion part 3. The photoelectric conversion part 3-1 amplifies and A/D converts the images on the slip formed on the CCD and stores the images of written characters in an image memory 4-1. The photoelectric conversion part 3-2 amplifies and A/D converts the images on the slip formed on the CCD and stores the images of character frames and the written characters in the image memory 4-2. A pre-processing part 26 calculates the inclination of an upper edge, corrects a read field from the inclination and segments a character. When rejection is performed, the address of the character frame of the rejected character is calculated and the character is segmented again corresponding to the address.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image acquired when a form is meandering in an optical character reader (hereinafter abbreviated as OCR (Optical Character Reader)).
The present invention relates to correction of character positions for cutting out characters.

[0002]

2. Description of the Related Art FIG. 2 is a conceptual diagram of a conventional OCR. 3A and 3B are conceptual diagrams of the OCR sheet feeding / reading mechanism showing an example of the configuration of the mechanical unit 1 and the optical system 2 in FIG. 2, and in particular, FIG. 3A is a plan view. The same figure (B) is a side view. Hereinafter, referring to FIG. 2 and FIG. 3, the conventional O
The operation of the CR will be described. As shown in FIG. 3, the form bundle 1-3 is set on the hopper 1-1 in the mechanism unit 1 in FIG. 2 along the hopper reference surface 1-2. When reading is started, first, the suction roller 1-4 descends to feed one form 2-2. When multiple sheets are about to be fed at the same time, the reversing roller 1-6 causes the second and subsequent sheets 2-2 to be fed to the hopper 1.
Since it is returned to -1, it is always fed one by one. Form 2-passed through paper feed roller 1-5 and reversing roller 1-6
Reference numeral 2 designates drive rollers 1-7, 1-8, 1 located on the way of the traveling path.
-9 It is conveyed toward the reading position R by other rollers. The form 2-2 is illuminated by the lamp 2-1 arranged at the reading position R, the image of the form is reflected by the mirror 2-4, is condensed by the lens 2-3 in FIG. The image is formed on the CCD sensor. The image of the form 2-2 photoelectrically converted by the CCD sensor is stored in the image memory 4 as multivalued information (usually 8 to 32) of the entire form 2-2.
It is stored as.

The preprocessing unit 6 cuts out the contents of the image memory 4 for each read field based on the form format information, and finally cuts out as an image of each character unit, and sends it to the feature extraction unit 7. In the feature extraction unit 7,
The feature amount of the cut out image is calculated, and the identifying unit 8
The character is identified by referring to the identification dictionary 9 based on the characteristic amount, and is output to the upper WS (workstation) through the IF control unit 10. When the recognition of one form 2-2 is completed, the common control unit 11 instructs the mechanism control unit 5 via the system bus 12 to feed the next form. The mechanical unit 1 feeds and conveys the next form to the optical system 2 in the same manner as described above. In the conventional device, the reference plane of the form running path is provided on the extension of the hopper reference plane 1-2 in FIG. 3, and the form 2-2 is provided by an aligner mechanism (a mechanism that presses the form against the reference plane). Was running along the reference plane of the form running path, and a form image without inclination was acquired. In recent years, it has become possible to reduce the size and weight of the device, and to reduce the amount of forms to be read (for thin forms of 40 kg or less, etc.). Instead of the aligner mechanism,
The inclination of the form has been corrected in terms of electric potential. For this reason, the reference plane of the form traveling road disappears, and it should also be called the right end wall 1-10 of the form traveling road at a place considerably distant from the virtual traveling road reference plane (referred to as a side frame). There is also). The fact that there is no reference surface for the road and there is no aligner mechanism means that thin forms are prevented from being pressed against the reference surface by the aligner mechanism and breaking or wrinkling.
For some reason, the suction roller 1-4 and the paper feed roller 1-5 have a problem that the form 2-2 that is inclined and fed is conveyed in the traveling path while being inclined. is what happened.

FIGS. 4A and 4B show the image memory 4
It is a conceptual diagram of the image of the form in the. As shown in FIG. 4, the image memory 4 in FIG. 2 corresponds to the form,
It has a virtually planar structure. Addresses are provided in the horizontal direction X (X0 to Xn) and the vertical direction Y (Y0 to Yn). Actually, this is a multi-valued information memory in which a plurality of memories (the number of gradations) of this planar structure are stacked, but here,
For the sake of clarity, the description will be given with a plane structure of X and Y. In FIG. 4A, A1, B1, C1, and D1 are the four corners of the form 2-2, (X0, Y0) is the origin based on the upper and left sides of the form, (XA1, YA1), (XB1). , YB1) is
These are the addresses A1 and B1. A2 in FIG. 4 (B)
B2, C2, D2 are the four corners of the form 2-2, (X0, Y
0) is the origin based on the top and left sides of the form, (XA
2, YA2) and (XB2, YB2) are the addresses of A2 and B2. FIG. 4A is a diagram showing a case where α1 = β1 where the upper side U has a slope α1 and the lower side has a slope β1, and the form 2-2 that has been conveyed to the reading position R with a slope α1 rising to the left. The lower side D also shows the image of the form that has passed the reading position R with the same inclination α1 as the upper side U. In this case, it can be seen that the reading field F1 on the upper side U (marked with 110), the reading field F2 on the lower side D (marked with 220) F2, and the reading field at any other position can be accurately cut out. ing.

[0005]

However, in the above-mentioned conventional OCR, as shown in the case of α2 ≠ β2 shown in FIG. 4B, the sheet is conveyed to the reading position with an inclination to the left (upper side inclination = α2). When the received form 2-2 meanders during reading (passing through the reading position) and passes through the reading position with a slope β2 on the lower side D, which is the reverse tendency of the upper side U, and is an upward slope. , If the inside of the form is cut out using only the inclination α2 of the upper side U as the correction coefficient, the reading field (1
F1 (marked as 10) can be accurately cut out, but reading field F2 (marked as 220) on the lower side D
Sometimes it was not possible to cut out exactly. That is, FIG.
In (B), with respect to the reading field F2 close to the lower side D, even if an attempt is made to cut out the inside of the form using only the inclination α2 of the upper side U as a correction coefficient, the position of the image is displaced, and there is a problem that it cannot be accurately cut out. there were. This tendency of meandering of the form should be avoided because it is related to the dimensional accuracy, assembly accuracy, and other variations of mechanical parts during mass production of the device, the size of the form, and the amount of continuous material, and environmental conditions (temperature, humidity). It is a problem that cannot be passed.

[0006]

In the first invention, in order to solve the above-mentioned problems, the OCR feeds and conveys a form in which characters are written or printed in a character frame printed in dropout color. And the optical system that obtains an optical signal by irradiating the conveyed form by the lamp, and the optical signal output from the optical system. It is provided with a first photoelectric conversion unit for converting into an electric signal to obtain an image of a written character or a printed character in the character frame. Then, a first image memory that stores the image obtained by the first photoelectric conversion unit, and an optical signal output from the optical system that amplifies a minute signal that is close to the white leher included in the dropout color.
It is provided with a second photoelectric conversion unit for extracting and converting into an electric signal to obtain the image of the character frame, and a second image memory for storing the image obtained by the second photoelectric conversion unit. . Further, the inclination of the upper side of the form from the reference line is measured, and the inclination and the information of each field which is an area including one or more of the character frames beside the form,
And a first pre-processing unit that cuts out each character pattern in the reading field based on the form format information that represents the information of each character frame in each field, and the cut-out character is determined to be unreadable by the identification unit. Then, check the cutout status of the unreadable character pattern, and if the top or bottom of the character pattern touches the frame of the cutout range, the cutout position is judged to be inappropriate and the unread Secondly, the address of the character frame in the read field is calculated from the image stored in the second image memory, and the unreadable character pattern is re-cut out according to the address of the character frame. And a discriminating unit for recognizing the characters cut out by the first and second pre-processing units. According to the first invention, as described above, the OC
Since R is configured, when the cut-out character in the reading field becomes unreadable by the identification unit, the address of the character frame to be included in the reading field which becomes unreadable is obtained from the second image memory, and Re-cut out the character pattern according to the address. Therefore, the above problem can be solved.

[0007]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS First Embodiment FIG. 1 is a block diagram of an OCR according to a first embodiment of the present invention, in which elements common to those shown in FIG. I am doing it. The OCR of the first embodiment is a conventional OCR
3 is different from the photoelectric conversion unit 3 for reading the character frame.
2 and an image memory 4-2 that stores the image of the character frame
And the preprocessing unit 26 measures the inclination of the upper side of the form from the reference line, corrects the inclination of the reading field based on the inclination, and cuts out the character pattern in the reading field. For the reading field that is unreadable by the processing unit and the identification unit 8, the address of the character frame that should be included in the reading field is calculated, and the unreadable character pattern is re-cut out based on the address. And a second pre-processing unit for As shown in FIG. 1, in this OCR, the mechanical unit 1, the optical system 2, the first photoelectric conversion unit 3-1, the second photoelectric conversion unit 3-2, the first image memory 4-1, and the second image memory 4-1 are used. Image memory 4-2, mechanism control unit 5, pre-processing unit 26, feature extraction unit 7, identification unit 8,
It is composed of an identification dictionary 9, an I / F controller 10, a common controller 11, and a system bus 12.

The optical system 2 has a form 2-for the mechanical unit 1.
2 are arranged in the transport direction. The output side of the optical system 2 is
The photoelectric conversion unit 3-1 is connected. Photoelectric conversion unit 3-1
The image memory 4-1 is connected to the output side of, and the image memory 4-2 is connected to the output side of the photoelectric conversion unit 3-2. The preprocessor 26 is connected to the output sides of the image memories 4-1 and 4-2, and the feature extraction unit 7 is connected to the output side of the preprocessor 26. An identification unit 8 is connected to the output side of the feature extraction unit 7. The input / output side of the identification unit 8 is connected to an identification dictionary 9 that stores character feature amounts. An I / F control unit 10 is connected to the output side of the identification unit 8. An upper WS is connected to the output side of the I / F control unit 10. Image memories 4-1, 4-2, mechanism control unit 5, preprocessing unit 26, feature extraction unit 7, identification unit 8, I
The system bus 12 is connected to the / F control unit 10 and the common control unit 11. The mechanical unit 1 does not have a reference surface of a form traveling path and an aligner mechanism, and feeds and conveys one form 2-2 from a form bundle set in a hopper, and the optical system 2 Form 2 by the lamp 2-2 at the reading position
-2 is irradiated and an image is formed on the CCD sensor of the photoelectric conversion unit 3-1 through the lens 2-3. The photoelectric conversion unit 3-1 uses a CCD sensor to form 2-2.
The above image is photoelectrically converted and further amplified and A / D converted to read ordinary written characters and printed characters. The photoelectric conversion unit 3-2 amplifies and A / D converts the electric signal photoelectrically converted by the CCD sensor disposed in the photoelectric conversion unit 3-1, and is a character frame printed in dropout color. Is to be read.

The image memory 4-1 includes a photoelectric conversion unit 3-
1 is a memory for storing a normal image of the form 2-2 obtained in 1 and the image memory 4-2 is a character frame printed in the drop-out color of the form 2-2 obtained in the photoelectric conversion unit 3-2. Is a memory for storing the image of. Preprocessing unit 26
In the first preprocessing unit, characters are cut out, and the second preprocessing unit recuts unreadable reading fields. The feature extraction unit 7 extracts the feature amount of the cut out character, and the identification unit 8 identifies the character. The identification dictionary 9 is a dictionary that stores character feature data for identifying characters. The mechanism control unit 5 controls the paper feeding operation of the mechanism unit 1, and the I / F control unit 10 is for interfacing with a higher-level WS that performs knowledge processing of characters recognized by the recognition unit 8. Common control unit 11
Is for controlling the operation of the mechanism control unit 5, the preprocessing unit 26, the feature extraction unit 7, the identification unit 8, and the I / F control unit 10.

FIG. 5 is a diagram showing an example of the configuration of the photoelectric conversion unit in FIG. As shown in FIG. 5, the photoelectric conversion unit 3-1
Is a CCD sensor and photoelectric conversion unit 13, an amplifier circuit 14-
1, the A / D conversion circuit 15-1, and the photoelectric conversion unit 3-2 includes an amplification circuit 14-2 and an A / D conversion circuit 15-2.
It is composed of CCD sensor and photoelectric conversion unit 13
The output side of is connected to the amplifier circuits 14-1 and 14-2. The A / D conversion circuit 15-1 is connected to the output side of the amplification circuit 14-1. The output side of the amplifier circuit 14-2 is connected to the A / D conversion circuit 15-2. CCD
The sensor and photoelectric conversion unit 13 inputs the light condensed by the lens 2-3 and converts it into an electric signal, and the amplifier circuits 14-1 and 14-2 amplify the electric signal. . The A / D conversion circuit 15-1 includes an amplifier circuit 14-1.
The electrical signal output from the dropout color is converted into a digital signal between a reference voltage slightly lower than the signal level of the dropout color and a reference voltage near the black level.
The A / D conversion circuit 15-2 converts the electric signal output from the amplification circuit 14-2 into a digital signal between a reference voltage slightly lower than the signal level of the dropout color and a white level reference voltage. It is a thing.

FIG. 6 is a diagram showing an example of a character frame of a form. As shown in FIG. 6, an entry character frame is printed in advance on the form, and characters are to be entered in the entry character frame. The writing character box has a writing character frame FR1 which is an independent character frame for each character, a writing character frame (tabular form) FR2 in the form of a table in which each character frame is vertically and horizontally attached,
Character frame F for printing that writes multiple characters in one character frame
There is R3 etc. These character frames are printed in a dropout color close to the white level, and are used as a guide when writing or printing. Below, referring to these figures,
The operations (a) to (j) of the OCR of FIG. 1 will be described.

(A) Mechanism Unit 1 In the mechanism unit 1, under the control of the mechanism control unit 5, the form 2-2 is fed from the form bundle set in the hopper by the suction roller and the paper feed roller, It is conveyed in the direction of the optical system 2 by the conveying roller. (B) Optical system 2 The form 2-2 conveyed by the conveying rollers is irradiated on the lamp 2-1 at the reading position R, and the image on the form 2-2 is condensed by the lens 2-3, An image is formed on the CCD sensor 13 of the photoelectric conversion unit 3-1 in FIG. (C) Photoelectric conversion unit 3-1 FIGS. 7A and 7B are operation conceptual diagrams of the photoelectric conversion unit, and in particular, FIG. 7A is a diagram showing a part of the form 2-2. 5B is a diagram showing output signals of the amplifier circuits 14-1 and 14-2 in FIG. 5A of FIG. 5A at the reading position R. F1 to F6 in FIG. 7A are vertical character frame portions F
R is shown and C1 to C4 represent the written character portion CH. In this example, 110 entry characters CH are entered in the three character boxes FR. In the amplifier circuit 14-1 in the photoelectric conversion unit 3-1, the CCD sensor and the photoelectric conversion unit 1 in FIG.
The electric signal obtained from No. 3 is amplified (for example, 0 to 5 V) as shown in FIG. In the A / D conversion unit 15-1, first, the signal amplified by the amplifier circuit 14-1 is set to a level (+) which is slightly lower than the level of the dropout color (for example, 4.5V). )
C, and a reference voltage (-) close to the black level (for example,
0.5V) B2, and level conversion is performed to a voltage between the (+) reference voltage C and the (−) reference voltage B2. Then, the area between the (+) reference voltage C and the (-) reference voltage B2 is converted into a digital value in N (for example, N is 8 to 32) stages, and written into the image memory 4-1. As a result, the character frame is processed as white, and the entered characters C1 to C4
Only are identified.

(D) Photoelectric conversion section 3-2 In the amplifier circuit 14-2 in the photoelectric conversion section 3-2, the electric signals obtained from the CCD sensor and the photoelectric conversion section 13 in FIG. Amplification (for example, 0 to 5 V) is performed as shown in FIG. This is because the thin line printed in the dropout color slightly drops from the white level W1 even in the photoelectric conversion waveform. In the A / D converter 15-2, first, the signal amplified by the amplifier circuit 14-2 is set to a level slightly lower than the level of the white level (+) reference voltage W1 and the dropout color. Compared with the (−) reference voltage C, the (+) reference voltage W1
The level is converted into a voltage between the reference voltage C of (-). Then, the interval between the (+) reference voltage W1 and the (-) reference voltage C is converted into digital values in N (for example, N is 8 to 32) steps and written into the image memory 4-2. As a result, in the image memory 4-2, the characters to be written are mixed with light characters such as dropout colors and signals of character frames.

(E) Image memories 4-1 and 4-2 FIGS. 8A and 8B are image memories 4-1 and 4-2.
It is a figure which shows the example of a part of image of the form stored in. As shown in FIG. 8A, the character frame is treated as a white level in the image memory 4-1 and the character frame information is not stored therein. On the other hand, as shown in FIG. 8B, the image memory 4-2 contains information on character frames and written characters. (F) Pre-Processing Unit 26 FIG. 9 is an operation explanatory diagram of the pre-processing unit 26 of the first embodiment, and shows an example of an image stored in the image memory 4-1. In this example, 110 characters are entered in the upper left character box, and 220 characters are entered in the lower right character box. A3, B3, C3, D3 indicate the four corners of the form 2-2, and the addresses of A3, B3 are (XA
3, YA3) and (XB3, YB3). (X
0, Y0) is the origin with the left and right sides of the form as reference sides, (X0 to Xn) is the horizontal direction X, and (Y0 to Yn) is the vertical direction Y. First, in the first preprocessing unit, if the reference side of the form 2-2 is the upper side U and the left side L, the image memory 4-
The address of the change point from the upper end of 1 (the area other than the form 2-2 is at the black level) to the white level is sequentially detected in the horizontal direction to detect the upper side U of the form.

Then, from the address (XA3, YA3) of the point A3 at the upper left corner of the form and the address (XB3, YB3) of the point B3 at the upper right corner of the form, for example, the image of the form can be expressed by the following equation (1). The slope α3 of the upper side is calculated. tan α3 = (YB3-YA3) / (XB3-XA3) (1) First, in order to cut out the reading field RF1 on the first line in FIG. 9, the size and position of the character frame of the form 2-2, etc. From the form format information (which can be obtained by fixedly determining the form format information in the system and searching it, or by reading the form ID and obtaining the form ID from the form ID) A cutout range S1 of the first row is set in the vertical direction from the point of the dimension Y1 from U to the center line of the reading field RF1 with an inclination α3. Next, the horizontal range of the read field RF1 is set within the cut-out range S1 of the first line, and each character included in the read field RF1 is cut out and output to the feature extraction unit 7. Next, the reading field RF2 in the second row
In order to cut out, the cut-out range S2 of the second line is set in the horizontal direction from the point of the dimension Y2 from the upper side U with the inclination α3 of the upper side according to the format information. Then, the horizontal range of the read field RF2 is set within the cut-out range S2 of the second row, and each character included in the read field RF2 is cut out and output to the feature extraction unit 7.

Each of these cut-out characters is a feature extraction unit 7
After the features are extracted in step S1, the identification unit 8 recognizes characters. In the reading field RF1, since the cut-out character is corrected based on the inclination α3 of the upper side U, as shown in FIG. 9, each of the cut-out characters in the read field RF1 contains the written character 110. Therefore, the identifying unit 8 correctly identifies the character. FIG.
(A), (B) is a diagram showing an example of a cutout image of the reading field RF2, in particular (A) is an image corresponding to the reading field RF2 on the form,
FIG. 2B shows the read field RF actually cut out.
It is an image of the cutout character of 2. 9 and 10
As shown in (B), the second reading field RF2 can be cut out only in the form in which the upper part of the image of the written character 220 is cut off. This is because the form 2-2 travels in a meandering manner, and even if the inclination α3 (upper left) of the upper side U of the form 2-2, the inclination β3 of the lower side D is the reverse (downward right) of α3. Is.

In such a case, the reading result of the character having the reading field is the recognition unit 8
It is unreadable (rejected) due to: ・ When the cutout image of the character touches the upper or lower part of the frame of the cutout range (the image is cut out improperly because the cutout range is inappropriate), Image memory 4-2 by the second preprocessing unit described in
The character frame inside is accurately cut out, its address is detected, and the correct character is cut out again. As described above, this processing is performed only when the read result is unreadable (reject) because the processing speed of the entire apparatus is kept as high as possible. If the processing is performed while detecting all the character frames in the configuration of FIG. 1, the processing speed will be significantly slowed down. In the second preprocessing section, the second read field R
The character frame to be included in F2 is searched laterally from the image memory 4-2 in order from the top, and the address (X10 ', Y10') of the character frame in which the black dot first appears is detected. From this, it is determined whether the character frame is descending to the right (the character frame at the left end is detected) or is descending to the left (the character frame at the right end is detected). In this example, the leftmost character frame is detected first,
It is judged that it is falling to the right.

Among the character frames to be included in the second reading field RF2, if the character frame is in the lower right direction, the address of the black dot (X20 ', Y20') that maximizes the X address.
If the character frame is rising to the right, the address (X20 ', Y20') of the black dot that minimizes the X address is detected. The approximate inclination β3 ′ of the reading field RF2 is calculated by the following equation (2). tan β3 ′ = (Y20′−Y10 ′) / (X20′−X10 ′) (2) The inclination correction of the inclination β3 ′ is performed, and the character frame to be included in the reading field RF2 is obtained as follows. Calculate the address accurately. 11A to 11C are diagrams showing an example of character box search.

As shown in FIG. 11 (A), the inclination of β3 'is corrected, and the upper and lower portions are set slightly wider than the normal cut-out range from the image memory 4-2 to set the character frame search area AR. Cut out. (X 'in FIG. 11 (A)
, Y ') represents a coordinate axis obtained by rotating (X, Y) by β3' around the origin (X0, Y0) in FIG.
Then, based on the width W of the character frame, the pitch P of the character frame, and the height H of the character frame obtained from the form format information, the character frame search area AR is in the range X1 to X2 (including each character frame) in the X ′ direction. When the number of black spots in the Y'direction for each address is counted and projected, the projection data shown in FIG. 11B is obtained. As shown in FIG. 11B, since the black dots are concentrated in the character frame portion, the X of the character frame of that portion is determined based on the matching / judgment with the information of the width W of the character frame and the pitch P of the character frame. The exact address in the'direction can be specified. When the number of black dots in the X ′ direction for each address in the Y ′ direction in the character frame search area AR is counted in the range of Y1 to Y2, FIG.
The projection data shown in (C) is obtained. As shown in FIG. 11C, since the black dots are concentrated on the character frame portion, it is possible to determine the exact value in the Y ′ direction of the character frame of that portion from the matching / judgment with the information of the height H of the character frame. The address can be specified.

Accurate X and Y addresses of the character frame are obtained from the inclination β3 ', the address in the X'direction, and the address in the Y'direction. Then, according to the address of the character frame of the unreadable character, the character is cut out again and output to the feature extraction unit 7. Furthermore, if there is a read field after this, the slope β3 expressed by the following equation (3)
Then, the inclination of the subsequent reading field is corrected. Then, the accurate inclination β3 of the reading field RF2 is calculated, for example, by the following equation (3). tan β3 = (Y20−Y10) / (X20−X10) (3) where (X10, Y10) is the read field RF2.
The address of the upper left point of the leftmost character box (here, the address of the character box where the first entry character "2" is entered),
(X20, Y20) is the address of the upper left point of the character box at the right end of the reading field RF2 (here, the address of the character box in which the last entry character "1" is entered). Further, the slope of the formula (3) calculated from the address of the character frame obtained from the distribution of the black dots is more accurate than the slope of the formula (2) calculated from the position of the black dots. . Furthermore, if the cut-out character in the subsequent reading field becomes unreadable and the character pattern is in contact with the cut-out range, re-cut out that character pattern and make it unreadable as described above. According to the inclination of the read field, the inclination of the subsequent read fields is corrected and the character pattern is cut out.

(G) Feature Extracting Unit 7 In the feature extracting unit 7, the 1
Various features are extracted from the character image. (H) Discrimination Unit 8 The discrimination unit 8 refers to the discrimination dictionary 9 using the characteristic data for one character transferred from the characteristic extraction unit 7, and selects a candidate character. In the reading field RF2 cut out by the first preprocessing unit, since the cut-out character is over or under (in this example, over), the cut-out character is unreadable (reject). As for the read field RF2 cut out again by the second pre-processing unit, each cut-out area completely contains characters and the direction of the characters is set correctly, so that the identification unit 8 does not reject it. Then, the candidate character is selected. The candidate character is I
It is transferred to the upper WS (workstation) through the / F control unit 10, and after some knowledge processing (word matching, context processing, validity check, etc.), it is determined to be a certain character code.

(I) Common Control Unit 11 In the common unit control unit 11, if the identification result of the identification unit 8 is unreadable (reject), the
Notify the preprocessing unit 26. When the identification of one sheet is completed, the mechanism control unit 5 is instructed to feed the next sheet. (J) Mechanism Control Unit 5 The mechanism control unit 5 instructs the mechanism unit 1 to feed the next form from the form bundle set on the hopper in accordance with the instruction from the common control unit 11. As described above, according to the first embodiment, the reading result becomes unreadable,
When the character pattern is in contact with the cutout range, the address of the character frame of the character pattern is obtained and the character pattern is cut out again.Therefore, cutout or unreadness caused by the form meandering at the reading position It is expected that excellent OCR will be realized without misreading.

Second Embodiment FIG. 12 is a block diagram of an OCR according to a second embodiment of the present invention. Elements common to those in FIG. 1 are designated by common reference numerals. The difference between the OCR of the second embodiment and the OCR of the first embodiment is that the form is vertically divided into a plurality of block areas, the inclination of the form is calculated for each block area, and the inclination is determined by the inclination. That is, the inclination of the read field is corrected for each block area. As shown in FIG. 12, in this OCR, the mechanical unit 1, the optical system 2, the photoelectric conversion units 3-1, 3-2, and the image memories 4-1, 4- are used.
2, mechanism control unit 5, pre-processing unit 36, feature extraction unit 7, identification unit 8, identification dictionary 9, I / F control unit 10, common control unit 1
1 and the system bus 12. The optical system 2 is arranged with respect to the mechanism unit 1 in the conveyance direction of the form 2-2. The output side of the optical system 2 has a photoelectric conversion unit 3-
1 and 3-2 are connected. The image memory 4-1 is connected to the output side of the photoelectric conversion unit 3-1, and the image memory 4-2 is connected to the output side of the photoelectric conversion unit 3-2.

The preprocessor 36 is connected to the output sides of the image memories 4-1 and 4-2, and the feature extractor 7 is connected to the output side of the preprocessor 26. An identification unit 8 is connected to the output side of the feature extraction unit 7. The input / output side of the identification unit 8 is connected to an identification dictionary 9 that stores character feature amounts. An I / F control unit 10 is connected to the output side of the identification unit 8. An upper WS is connected to the output side of the I / F control unit 10. Image memories 4-1, 4-2, mechanism control unit 5, preprocessing unit 36, feature extraction unit 7, identification unit 8, I
The / F control unit 10 and the common control unit 11 are connected by the system bus 12. Hereinafter, the operation (a) of FIG.
~ (L) will be described.

(A) Mechanism Unit 1 The mechanism unit 1 operates in the same manner as the mechanism unit 1 of the first embodiment. (B) Optical system 2 The optical system 2 operates similarly to the optical system 2 of the first embodiment. (C) Photoelectric conversion unit 3-1 The photoelectric conversion unit 3-1 is the photoelectric conversion unit 3- of the first embodiment.
It operates in the same manner as 1. (D) Photoelectric conversion unit 3-2 The photoelectric conversion unit 3-2 is the photoelectric conversion unit 3- of the first embodiment.
It operates in the same way as 2. (E) Image Memory 4-1 In the image memory 4-1, the same as in the first embodiment,
The image of the written character is stored. (G) Image memory 4-2 In the image memory 4-2, as in the first embodiment,
An image of the character frame and the entered character is stored.

(H) Pre-Processing Unit 36 FIG. 13 is an operation explanatory view of the pre-processing unit 36 of the second embodiment, and shows an example of the image stored in the image memory 4-2. A4, B4, C4, D4 in FIG.
Indicates the four corners of form 2-2, and the addresses of A4 and B4 are
(XA4, YA4) and (XB4, YB3). (X0, Y0) is the origin with the left and right sides of the form as reference sides, (X0 to Xn) is the horizontal direction X, (Y0 to Yn).
Indicates the vertical direction Y. As shown in FIG. 13, the slope α4 of the upper side of the form and the slope α4 of the lower side of the form are β4, and α4 ≠ β4
And This shows that this form was being read while meandering. When reading this form, it is first necessary to detect the vertical dimension of this form. This is because it is possible to read the form ID of the form 2-2, know the vertical dimension of the form from the form format information, and from the image of the form 2-2 in the actual image 4-1. Measure the vertical dimension.

In this embodiment, the vertical dimension of this form is 400 mm. First, the form is 100m from the upper side U
m (This 100 mm is not fixed but is set arbitrarily.) Divide into four block areas. First block area A1: upper side to 100 mm Second block area A2: 100 mm to 200 mm Third block area A3: 200 mm to 300 mm Fourth block area A4: 300 mm to 400 mm (lower side) First block area A1 Cuts out while using the upper side inclination α4 as form inclination correction data. For the second block area A2, 10 from the upper side
For a read field that is closest to 0 mm and is as continuous as possible in the horizontal direction (for example, line L200 in FIG. 13), in the same manner as the process of the second preprocessing unit of the first embodiment, the character frame Is detected, the degree of inclination of the form 2-2 at that position is calculated from the address information in the image memory 4-2. Then, using the calculated inclination correction data, characters are cut out in the reading field in the second block area A2.

For the third block area A3, the read field is closest to 200 mm from the upper side and is as continuous as possible laterally (for example, near the L200 line in FIG. 13).
On the other hand, similarly to the processing of the second preprocessing unit of the first embodiment, the character frame is detected, and the degree of inclination of the form at that position is determined from the address information in the image memory 4-2. To calculate. Then, while using the calculated inclination correction data, characters are cut out in the reading field in the third block area A3. For the fourth block area, the read field that is closest to 300 mm from the upper side and is as continuous as possible in the horizontal reading field (for example,
13) (near line L300 in FIG. 13), the character frame is detected similarly to the processing of the second preprocessing unit of the first embodiment, and the address information in the image memory 4-2 is detected. Then, the degree of inclination of the form 2-2 at that position is calculated.

Then, using the calculated inclination correction data, character cutting is performed on the read field in the fourth block area A4. In this way, each 100 mm
The reason why the address of the character frame is detected while keeping the distance is is to keep the overall processing speed as fast as possible. If reading is performed while detecting the addresses of all the character frames, the processing speed will be significantly slowed down. Also,
With a general mechanism, the inclination of the form does not change significantly within 100 mm, so this interval (100 m
(about m), sufficient performance can be obtained by always detecting and following the inclination of the form.

(I) Feature Extracting Unit 7 The feature extracting unit 7 extracts the features of each character cut out by the preprocessing unit 36. (J) Recognition Unit 8 The identification unit 8 uses the feature data for one character transferred from the feature extraction unit 7 to refer to the identification dictionary 9 to select a candidate character. The candidate character is transferred to the upper WS (workstation) via the I / F control unit 10 and, after some knowledge processing (word matching, context processing, validity check, etc.), is determined as a certain character code. (K) Common Control Unit 11 In the common unit control unit 11, if the identification result of the identification unit 8 is unreadable (reject), the
Notify the preprocessing unit 26. When the identification of one sheet is completed, the mechanism control unit 5 is instructed to feed the next sheet. (L) Mechanism Control Unit 5 In accordance with the instruction from the common control unit 11, the mechanism control unit 5 instructs the mechanism unit 1 to feed the next form from the form bundle set on the hopper. As described above, according to the second embodiment, the inclination of the reading field is obtained for each of the plurality of block areas, and the inclination of the reading field is corrected. Therefore, there is the same advantage as the first embodiment.

Third Embodiment FIG. 14 is a block diagram of an OCR according to a third embodiment of the present invention. Elements common to those in FIG. 1 are designated by common reference numerals. The OCR of the third embodiment is different from the OCR of the first embodiment in that for each reading field, the character box processing unit 41 that calculates the addresses of all the character frames of the reading field is provided, and the preprocessing unit 46. That is, the character pattern in the character frame is cut out according to the address of each character frame. As shown in FIG.
In CR, the mechanism unit 1, the optical system 2, and the first photoelectric conversion unit 3-
1, second photoelectric conversion unit 3-2, first image memory 4
-1, second image memory 4-2, mechanism control unit 5, character processing unit 41, preprocessing unit 46, feature extraction unit 7, identification unit 8, identification dictionary 9, I / F control unit 10, common control Part 1
1 and the system bus 12.

The optical system 2 has a form 2-for the mechanical unit 1.
2 are arranged in the transport direction. The output side of the optical system 2 is
The photoelectric conversion unit 3-1 is connected. Photoelectric conversion unit 3-1
The image memory 4-1 is connected to the output side of, and the image memory 4-2 is connected to the output side of the photoelectric conversion unit 3-2. The output side of the image memory 4-1 is connected to the preprocessing section 46, and the output side of the preprocessing section 26 is connected to the feature extraction section 7. The character frame processing unit 41 is connected to the output side of the image memory 4-2, and the preprocessing unit 46 is connected to the output side of the character frame processing unit 41. An identification unit 8 is connected to the output side of the feature extraction unit 7. The input / output side of the identification unit 8 is connected to an identification dictionary 9 that stores character feature amounts. The output side of the identification unit 8 has an I / F control unit 10
Is connected. An upper WS is connected to the output side of the I / F control unit 10. Image memory 4-1, 4-
2, mechanism control unit 5, character box processing unit 41, pre-processing unit 46,
A system bus 12 is connected to the feature extraction unit 7, the identification unit 8, the I / F control unit 10, and the common control unit 11. The operations (a) to (l) of FIG. 14 will be described below.

(A) Mechanism Section 1 The mechanism section 1 operates in the same manner as the mechanism section 1 of the first embodiment. (B) Optical system 2 The optical system 2 operates similarly to the optical system 2 of the first embodiment. (C) Photoelectric conversion unit 3-1 The photoelectric conversion unit 3-1 is the photoelectric conversion unit 3- of the first embodiment.
It operates in the same manner as 1. (D) Photoelectric conversion unit 3-2 The photoelectric conversion unit 3-2 is the photoelectric conversion unit 3- of the first embodiment.
It operates in the same way as 2. (E) Image Memory 4-1 In the image memory 4-1, the same as in the first embodiment,
The image of the written character is stored. (F) Image memory 4-2 In the image memory 4-2, as in the first embodiment,
An image of the character frame and the entered character is stored.

(G) Character Box Processing Unit 41 FIG. 15 is an operation explanatory diagram of the character frame processing unit 41 of the third embodiment, and shows an example of an image stored in the image memory 4-2. A5, B5, C5 and D5 are the four corners of the form, (X0, Y0) is the origin based on the upper and left sides of the form, and the addresses of A5 and B5 are (XA5,
YA5), (XB5, YB5). As shown in FIG. 15, the inclination α5 of the upper side of the form and the inclination α5 of the lower side of the form are β5, and α5 ≠ β5. This shows that this form was being read while meandering. First, in the character box processing unit 41, the inclination α5 of the upper side is calculated, and from the inclination α5 of the upper side, each address of the character frame in the first line is processed in the same manner as the processing of the second preprocessing unit of the first embodiment. Is calculated and passed to the preprocessing unit 46. Next, each address of the character frame on the second line is calculated in the same way as the process of the second preprocessing unit of the first embodiment, and is passed to the preprocessing unit 46. Hereinafter, similarly, for each of the third line, the fourth line, ..., The seventh line, and the last line, each address of the character frame is set to be the same as the process of the second preprocessing unit of the first embodiment. Calculated and passed to the preprocessing unit 46. In this way, since the address of the character frame is calculated for each line, even if the image of the form is bent, it will always follow because of meandering running in the middle of the form.
For example, as shown in FIG. 15, even if the vehicle is meandering largely in the opposite direction with respect to the fourth line to the fifth line and the sixth line and the inclination α5 of the upper side U, it is possible to reliably follow and control.

(H) Pre-Processing Unit 46 In the pre-processing unit 46, the character pattern within the character frame is cut out according to the address of each character frame output from the character frame processing unit 41, and the character pattern is extracted by the feature extraction unit 7 Pass to. (I) Feature Extraction Unit 7 The feature extraction unit 7 extracts the features of each character cut out by the preprocessing unit 36. (J) Recognition Unit 8 The identification unit 8 uses the feature data for one character transferred from the feature extraction unit 7 to refer to the identification dictionary 9 to select a candidate character. The candidate character is transferred to the upper WS (workstation) via the I / F control unit 10 and, after some knowledge processing (word matching, context processing, validity check, etc.), is determined as a certain character code.

(K) Common Control Section 11 In the common section control section 11, if the identification result of the identification section 8 is unreadable (reject), the
Notify the preprocessing unit 26. When the identification of one sheet is completed, the mechanism control unit 5 is instructed to feed the next sheet. (L) Mechanism Control Unit 5 In accordance with the instruction from the common control unit 11, the mechanism control unit 5 instructs the mechanism unit 1 to feed the next form from the form bundle set on the hopper. As described above, according to the third embodiment, in addition to the same advantages as the first embodiment, the character box processing unit 41 is provided. Is performed separately, it is possible to obtain sufficient performance in terms of speed. Note that the present invention is not limited to the above embodiment, and various modifications are possible. For example, there are the following modifications.

(1) In the first embodiment, the address of a character box that has become unreadable may be obtained and the character may be cut out again. , The leftmost character frame and the rightmost character frame are calculated to obtain the inclination shown in Expression (3), the inclination is corrected with the inclination, and the unreadable character may be cut out.
(2) The block area to be divided in the second embodiment may be appropriately determined according to the number of fields of the form, and the size of the block area may be different in one form. .

[0038]

As described in detail above, the first to third embodiments
According to the invention, the unreadable character pattern is divided into a plurality of block areas in which the address of the character frame is calculated and the character pattern is re-cut out, and the inclination of each block area is obtained. Since the characters are cut out by correcting the inclination of the read field or by obtaining the address of each character frame, no misreading or misreading due to a cutout failure due to the meandering of the form does not occur.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an OCR according to a first embodiment of the present invention.

FIG. 2 is a configuration diagram of a conventional OCR.

FIG. 3 is a conceptual diagram of an OCR sheet feeding / reading mechanism.

FIG. 4 is a diagram showing an image of a form in the image memory 4.

5 is a diagram showing an example of a configuration of a photoelectric conversion unit in FIG.

FIG. 6 is a diagram showing an example of a character frame.

FIG. 7 is an operation conceptual diagram of a photoelectric conversion unit.

FIG. 8 is a diagram showing an example of images stored in image memories 4-1 and 4-2.

FIG. 9 is an operation explanatory diagram of the preprocessing unit 26 of the first embodiment.

FIG. 10 is a diagram showing a cutout image of a reading field RF2.

FIG. 11 is a diagram illustrating an example of character box search.

FIG. 12 is a configuration diagram of an OCR according to a second embodiment of the present invention.

FIG. 13 is an operation explanatory diagram of the preprocessing unit 36 according to the second embodiment.

FIG. 14 is a configuration diagram of an OCR according to a third embodiment of the present invention.

FIG. 15 is an operation explanatory diagram of the character box processing unit 41 according to the third embodiment.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Mechanism part 2 Optical system 3-1 and 3-2 Photoelectric conversion part 4-1 and 4-2 Image memory 5 Mechanism control part 26,36,46 Pre-processing part 7 Feature extraction part 8 Identification part 9 Identification dictionary 10 I / F control unit 11 common control unit 41 character box processing unit

Claims (3)

[Claims] 1. A mechanism unit for feeding and transporting a form in which characters are written or printed in a character frame printed in dropout color, and a lamp is provided, and the form conveyed by the lamp is provided. And a first photoelectric conversion unit that obtains an image of a written character or a printed character in the character frame by converting the optical signal output from the optical system into an electric signal, A first image memory that stores an image obtained by the first photoelectric conversion unit; and an optical signal output from the optical system that amplifies a minute signal that is close to a white leher included in the dropout color. A second photoelectric conversion unit that obtains an image of the character frame by extracting and converting into an electric signal; a second image memory that stores the image obtained by the second photoelectric conversion unit; Upper standard lie Information of each field that is an area including one or a plurality of the character frames beside the form and the form, and form format information indicating the information of each character frame in each field First, cut out each character pattern in the reading field based on
When the cut-out character is determined to be unreadable by the pre-processing unit and the identifying unit, the cutting-out status of the unread character pattern is checked,
When the upper or lower part of the character pattern is in contact with the frame of the cutout range, it is determined that the cutout position is inappropriate, and the address of the character frame in the reading field of the unread character is set to the second value. A second pre-processing unit, which calculates from an image stored in an image memory and re-cuts out unreadable character patterns according to the address of the character frame; and the first and second pre-processing An optical character reading device, comprising: an identification unit for recognizing characters cut out by the unit. 2. A mechanism part for feeding and conveying a form in which characters are written or printed in a character frame printed in dropout color, and a lamp is provided, and the form conveyed by the lamp is provided. And a first photoelectric conversion unit that obtains an image of a written character or a printed character in the character frame by converting the optical signal output from the optical system into an electric signal, A first image memory that stores an image obtained by the first photoelectric conversion unit; and an optical signal output from the optical system that amplifies a minute signal that is close to a white leher included in the dropout color. A second photoelectric conversion unit for extracting and converting into an electric signal to obtain the image of the character frame; a second image memory for storing the image obtained by the second photoelectric conversion unit; and the form. Multiple blocks Divide into areas, for the top block area, measure the inclination from the reference line of the top side of the form, for the other block areas,
Based on the information of each field that is an area including one or a plurality of the character frames beside the form and the form format information that represents the information of each character frame in each field, the second image memory The address of the character frame included in one field included in each block area is calculated from the image, the inclination of each block area is measured from the address of the character frame, and based on the measured inclination of each block area, The reading field in each block area is tilt-corrected to cut out a character pattern in the reading field, and a preprocessing section for recognizing the characters cut out by the preprocessing section are provided. An optical character reader. 3. A mechanism unit for feeding and conveying a form in which characters are written or printed in a character frame printed in dropout color, and a lamp is provided, and the form conveyed by the lamp is conveyed. And a first photoelectric conversion unit that obtains an image of a written character or a printed character in the character frame by converting the optical signal output from the optical system into an electric signal, A first image memory that stores an image obtained by the first photoelectric conversion unit; and an optical signal output from the optical system that amplifies a minute signal that is close to a white leher included in the dropout color. A second photoelectric conversion unit for extracting and converting into an electric signal to obtain an image of the character frame, a second image memory for storing the image obtained by the second photoelectric conversion unit, and each character frame Form that represents information A character frame processing unit that calculates the addresses of all the character frames from the image of the second image memory based on the format information, and a pre-processing for cutting out the character pattern in the character frames according to the address of the character frame. An optical character reading device comprising: a unit; and an identification unit that recognizes the character cut out by the preprocessing unit.