JP3337592B2 - Mark position detecting device and mark position detecting method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a technique for detecting a mark position used in an optical character reader or the like, and more particularly, to a method for storing a mark position of an arbitrary size on a form set in an arbitrary direction with a small memory. The present invention relates to a mark position detecting device and a mark position detecting method capable of detecting a mark at high speed.

[0002]

2. Description of the Related Art In an optical character reading apparatus, a form on which a mark of a predetermined size is previously written at a reference position of a recognition target area is usually used as a form on which characters to be recognized are described. . Then, in order to specify a character area to be recognized in the form image, a mark serving as a form position reference is first detected from the input form image.

FIG. 11 shows an example of a form. In this example, two form position reference marks 110 are written in the form 100, and a character recognition target area 111 is provided in accordance with the position of the form position reference mark 110. In the conventional optical character reader, these form position reference marks 110 are used.
Mark position is detected on the assumption that it exists in a certain limited area in the input form image.
Under such a premise, the detection speed has been increased.

[0004]

Recently, there is an increasing demand for performing character recognition on a form image received by, for example, a facsimile machine. However, for example, in order to detect a form position reference mark (hereinafter, referred to as a FAX mark) in a form image in order to perform character recognition on a form image received by a facsimile machine or the like, a limitation is imposed due to image expansion or displacement. The mark may not exist in the area, and therefore, the mark may not be detected, and the character may not be readable.

That is, in the prior art, there is a problem that characters cannot be read when a fax mark does not exist in a predetermined area due to expansion or displacement of a received form image. Conversely, if mark detection is performed for the entire form image as a range, there is a problem that the detection speed is reduced.

Therefore, even when a mark exists at a position deviating from a predetermined position in a form image, a technique for rapidly extracting the mark from the input form image is required.

[0007] In addition, a document to be transmitted by a facsimile apparatus may be set and transmitted in an incorrect direction, such as reverse or horizontal, due to an operation error or the like. In this case, a mark is placed at a predetermined position of the received form image. Not detectable. Therefore, it is necessary to determine the orientation of the form image using the FAX mark as a key to enable character recognition even for a form set in an arbitrary direction. In particular, in this case, it is desired to quickly specify the orientation of the form image with a small amount of memory.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above problems and to provide a means for detecting a mark at a high speed even when a form position reference mark is not within a limited range of an input form image. I do. It is another object of the present invention to specify the orientation of a form without using a large amount of memory even when the form is not set in the correct direction.

[0009]

The present invention is provided with each means as shown in FIG. 1 in order to solve the above problems. FIG.
1 is a diagram illustrating the principle of the present invention. In FIG. 1, 1 is a mark position detecting device used in a character reading device, 2 is a thinned-out image generation processing unit, 3 is a mark position detection processing unit, 4 is an image rotation processing unit, 5 is a division rotation processing unit, and 6 is a second rotation processing unit. 1 is a buffer, 7 is data compression processing means, 8
Denotes a second buffer, 9 denotes data decompression processing means, and 11a to
11d represents a form image.

The thinning image generation processing means 2 performs thinning processing in units of a plurality of rows or a plurality of columns of a form image.
This is means for generating a thinned image in which a plurality of rows or a plurality of columns are set to one row or one column. The thinned image generation processing means 2 generates a thinned image by, for example, taking a logical sum (OR) of a plurality of rows or a plurality of columns of a form image.

The mark position detection processing means 3 is a means for detecting a mark position in the form image using the thinned image of the form generated by the thinned image generation processing means 2.

The image rotation processing means 4 is means for rotating a form image. The image rotation processing means 4
As shown in FIG. 1B, the form image is rotated 180 degrees (form image 11b) for the first time with respect to the orientation of the original form image 11a, and the form image is rotated 90 degrees for the second time. Then, the form image is rotated by 270 degrees for the third time (form image 11d).

That is, in order to determine the correct orientation (erect) of a form set in an arbitrary direction, the image of the entire form is sequentially rotated to determine the thinned image generation processing means 2 and Using the mark position detection processing means 3, mark detection processing is performed on the form image after the rotation processing, and the direction of the form on which the mark is detected is set to the correctly set direction (erect).

The rotation of the form image by the image rotation processing means 4 uses the first buffer 6 and the second buffer 8 as temporary work buffers, and uses the divided rotation processing means 5, the data compression processing means 7, and the data decompression. This is performed by the processing means 9.

The division rotation processing means 5 divides the form image to be rotated, performs rotation processing for each divided image,
This is means for storing in the first buffer 6. The data compression processing means 7 is means for compressing the divided image stored in the first buffer 6 by a predetermined data compression method and storing the compressed image in the second buffer 8. The data decompression processing means 9
A means for expanding the compressed form image stored in the second buffer 8 by a data expansion method corresponding to the data compression method used by the data compression processing means 7.

Thus, the first buffer 6 and the second buffer 6
It is possible to rotate the entire image of the form by using a small amount of memory only by using the buffer 8 of FIG.

[0017]

According to the present invention, the speed of mark detection can be increased by reducing the scanning area for mark detection in the form image by the thinned image generation processing means 2 and the mark position detection processing means 3.

Further, the image rotation processing means 4
The speed of the detection process is improved by using the characteristic of storing the image data of the value. This is because the data is sequentially arranged in the memory for storing the form image, so that when the mark is detected from the binary image, the mark is horizontally (90/270 degrees rotated) or reversed (180 degrees rotated). This is because mark detection for an erect (0 degree) image can be processed at higher speed than when mark detection is performed directly from the image.

In addition, taking into account the frequency of operation errors when a form is set, the form image is
Rotate degrees (reverse), 90 degrees (right) for the second time
By rotating the document image 270 degrees (leftward) at the third time, the correct orientation of the form image is specified at high speed. In other words, when identifying the orientation of a form set in the wrong direction, the rotation of the reverse set direction, which is likely to be mistaken by the operator, is given priority over the horizontal set direction, so that a high-speed stochastic form can be obtained. Achieve orientation identification.

Further, divided rotation processing means 5, data compression processing means 7, data decompression processing means 9, first buffer 6
The second buffer 7 performs the rotation processing with a small amount of memory. If the form image is to be rotated directly, there are two memories, one for storing the current form image and the other for storing the rotated form image.
One large memory is required. In the present invention, if the form image is divided into n pieces, the first buffer 6 may have a memory amount of 1 / n of the form image.
Is sufficient for the buffer 8 having a size corresponding to the data compression ratio. The rotation result obtained by expanding the data can be written back to the memory of the original form image. Therefore, when rotating the form image, a memory with twice the size of the entire form image (before and after rotation) ) Is not necessary.

[0021]

An embodiment of the present invention will be described below with reference to the drawings. FIGS. 2 and 3 are diagrams for explaining a thinned-out image generation process using a logical sum (OR) and a mark position detection process in the embodiment of the present invention.

As shown in FIG. 2A, a 1 / n logical OR (OR) thinning process is performed on a binary image of the entire form. That is, a logical sum (OR) is performed on the values of each of the n lines of the form image 11 and output as one line to generate the thinned image 12. As a result, the number of scanning lines for mark detection can be reduced to 1 / n, and high-speed detection can be realized.

Next, a black image having a width similar to the mark width is searched for in the left-right direction (x direction) of the thinned image 12. The black image that does not approximate the mark width is shown in Fig. 2.
As shown in (B), it is deleted from the mark candidates. FIG.
(C) is a diagram showing the size of the mark in the thinned image to be compared. The size of the thinned standard mark is calculated in advance.

When a black image having a width similar to the mark width is detected, a line that is entirely white in the vertical direction (y-direction) is searched by the width, and the height of the black image is calculated. If the height of the black image is close to the 1 / n mark height shown in FIG. 2C, this black image is set as a mark candidate. The three black images shown in FIG. 2D are mark candidates that are similar in mark width and mark height.

As shown in FIG. 2E, for the mark candidate selected in the above-described processing, the width and height of the circumscribed rectangle of the mark candidate are calculated from the image of the entire original form, and the mark Refine candidates further. FIG.
(F) is a diagram showing the size of the actual size mark in the form image 11.

If the calculated width and height of the circumscribed rectangle are similar to the width and height of the mark, the correct width and height of the mark candidate are determined for the image of the entire form by the following method. Find and confirm the mark. The following method is used, for example, when a form is set to be inclined and a mark in the form image 11 is inclined, an error occurs between the width and height of the circumscribed rectangle and the actual mark width and mark height. That's why.

1) As shown in FIG. 3A, the width (x)
A histogram of the number of pixels of each mark is created in the height (y) direction. 2) A normal distribution is obtained for each of the histograms in the x and y directions, and the maximum distribution position is determined.

3) For the longer one of the width (x) and the height (y), it is determined whether the width is representative of the whole. Specifically, if at least 1/2 of the entire width corresponds to the width, it is regarded as a representative.

4) If the total number of pixels of the mark and the area of the representative width / representative height obtained in 1) to 3) are similar, the mark candidate is determined as a mark. FIG. 3B is a diagram showing a pixel distribution of a histogram of the mark shown in FIG. 3A in the x direction. In the case of FIG. 3B, the representative width is 10 pixels. FIG. 3C is a diagram showing the pixel distribution of the histogram in the y direction. FIG. 3 (C)
In the case of, the representative height is 9 pixels.

As a result, the accuracy of mark detection is improved,
Marks can be detected quickly and accurately even from a slanted form image. Although the access efficiency of the memory is slightly deteriorated, the mark detection may be performed by generating a thinned image having a plurality of columns as one column, instead of generating a thinned image having a plurality of rows as one line.

Next, with reference to FIGS. 4 and 5, a description will be given of a division rotation process in which the entire image of a form is rotated at high speed and with a small amount of memory. In order to determine the correct orientation for a form set in an arbitrary direction, the above-described mark detection processing is performed, and when a mark is not detected, the form image is sequentially rotated, and the rotated form image , And the direction in which the mark was found is determined as the correct form direction. To perform this in a small-capacity memory, the following processing is performed. For the rotation processing, the first buffer 6 is used as a temporary working buffer.
And a second buffer 8 are prepared.

As shown in FIG. 4A, the entire form image 11 is divided into parts a, b, and c. here,
To make the explanation easier to understand, 3
Although it is divided into pieces, the number of divisions may be any number. The size of the first buffer 6 is such that one of the divided form images can be stored. First, the form image part a
Is rotated and output to the first buffer 6.

Next, the part a of the form image after the rotation processing stored in the first buffer 6 is compressed and subjected to the second processing.
To the buffer 8. As this compression method, an arbitrary compression method such as an MR encoding method can be used.

Next, as shown in FIG. 4B, the portion b of the form image is rotated and output to the empty first buffer 6. The part b of the form image after the rotation processing stored in the first buffer 6 is compressed and output to the second buffer 8.

As shown in FIG. 4C, the image portion c of the entire form is similarly subjected to rotation processing and compression processing and output to the second buffer 8. When the rotation processing and the compression processing have been completed for all the divided parts, as shown in FIGS. 5A to 5C, from the beginning of the compressed data of the parts a to c stored in the second buffer 8 , And sequentially perform image decompression processing. After rotationally compressing the entire form image, the memory area of the original form image becomes unnecessary, so the decompressed rotated form image is overwritten on the memory area of the original form image.

Thus, the first buffer 6 and the second buffer 6
It is possible to rotate the image of the entire form only by using the buffer 8 of. FIG. 6 is a diagram illustrating a configuration example of a character recognition device using the mark position detection device of the present invention. In FIG. 6, reference numeral 20 denotes a microprocessor for controlling each unit according to a predetermined program; 21, a high-speed mark detection processing unit;
23, a form positioning processing unit, 24, a one-character cutout processing unit, 25, a one-character recognition unit, and 26, a facsimile (FA)
X) Adapter, 27 is a facsimile (FAX) device, 2
8 is an image scanner, 29 is a shared memory for storing images,
Reference numeral 30 denotes a work memory.

The high-speed mark detection processing section 21 is processing means comprising a program for realizing the thinned image generation processing means 2 and the mark position detection processing means 3 shown in FIG. The high-speed form entire image rotation processing unit 22 is a processing unit including a program for realizing the image rotation processing unit 4 shown in FIG. The form positioning processing unit 23 is a processing unit including a program that determines the position and orientation of the form from the marks detected by the high-speed mark detection processing unit 21. One-character extraction processing unit 24 and one-character recognition unit 25
Is cut out one character at a time from the form reading field,
This is processing means comprising a program for recognizing the extracted characters.

The shared image storage memory 29 is a memory for storing a form image input from the facsimile machine 27 or the image scanner 28. Work memory 30
Is a memory used for storing the thinned image generated by the high-speed mark detection processing section 21 and for the first buffer 6 and the second buffer 8 shown in FIG.

Since the character recognition by the one-character cut-out processing unit 24 and the one-character recognition unit 25 can be realized by using a well-known character recognition technology, detailed description is omitted here. FIG.
FIG. 3 is a diagram showing another configuration example of a character recognition device using the mark position detection device of the present invention.

The thinning image generation processing, image rotation processing, image compression processing, and image decompression processing performed by the high-speed mark detection processing section 21 and the high-speed form full-page image rotation processing section 22 are performed, for example, as shown in FIG.
SI32, LSI for rotation 31, LSI for image compression3
3. It is also possible to perform the processing at a high speed by using an image decompression LSI 34 or the like. For each of these LSIs, an existing data processing LSI is used, or a known LSI is used.
Since it can be easily realized from the I technology, individual detailed description is omitted.

FIGS. 8 to 10 are processing flowcharts of the embodiment of the present invention. In FIG. 8, step S1
Then, the form image is input and stored in the image storage shared memory 29.

In step S2, an image is generated by thinning out the form image by 1 / n. Note that this thinned image does not necessarily need to be generated for the entire original form image, but is generated only for a partial area of the form image, for example, the upper fifth area of the entire form image. You may make it.

In step S3, the OR-thinned image is scanned line by line, and the segment (black dot group) is scanned.
To inspect. In step S4, it is determined whether the inspection has been completed for all segments. If the inspection of all the segments has been completed, the process proceeds to step S12 (FIG. 9), and if there is any segment that has not been completed, the process of step S5 is performed.

In step S5, it is determined whether a segment (black) having a width similar to the width of the mark stored in advance has been found. If a segment having a width similar to the mark width is found, the process proceeds to step S6; otherwise, the process proceeds to step S11.

In step S6, 1 / n of the mark height
It is determined whether or not the segment (black) has a height similar to the above. If the segment is close to 1 / n of the mark height, the process proceeds to step S7, and if not, the process proceeds to step S11.

In step S7, the width and height of the circumscribed rectangle of the corresponding segment are calculated from the original form image before thinning. The position of the corresponding segment can be calculated from the position of the segment in the thinned image.

In step S8, the width (x) and the height (y)
For each direction, a histogram as shown in FIG. 3 is created, and the representative width and the representative height are calculated from the normal distribution of the number of black pixels.

In step S9, it is determined whether or not the width / height of the mark and the representative width / height are similar. When the width / height of the mark and the representative width / height are approximate, the process of step S10 is performed, and when they are not approximate, the process proceeds to step S11.

In step S10, an approximate segment (black) is determined as a form position reference mark. In step S11, the process target is set to the next segment, and the process returns to step S4 to repeat the mark detection in the same manner.

In FIG. 9, in step S12, it is determined whether or not the mark has been determined and the orientation of the form has been specified. When the orientation of the form is determined, the process proceeds to step S24 (FIG. 10), and when the orientation of the form is not determined, the process of step S13 is performed.

In step S13, it is determined whether all rotation directions have been tried. If all rotation directions have been tried, the process proceeds to step S30. If there are still rotation directions that have not been tried, the process of step S14 is performed.

In step S14, the form is rotated in an untried rotation direction. The rotation is performed in the order of reverse (180 degrees), right (90 degrees), and left (270 degrees). In step S15, the form image 11 is divided by an amount that fits in the first buffer 6.

In step S16, it is determined whether or not the processing for all the divided images has been completed. If the processing for all divided images is completed, step S
Proceeds to the process of No. 19, and if not completed, proceeds to step S
Step 17 is performed.

In step S17, the divided image is rotated and output to the first buffer 6. Step S18
Then, the divided image rotated and output to the first buffer 6 is compressed by a predetermined data compression method and stacked in the second buffer 8.

In step S19, the rotated and compressed images stacked in the second buffer 8 are sequentially expanded.
In step S20, it is determined whether or not expansion processing of all rotated / compressed images has been completed. If the decompression process of all the compressed images has been completed, the process proceeds to step S23, and if not completed, the process of step S21 is performed.

In step S21, rotation of the unprocessed portion
The compressed image is expanded and output to the first buffer 6. In step S22, the rotated image output to the first buffer 6 is overwritten on the form image in the shared image storage memory 29. The image data may be directly developed in the shared image storage memory 29 without using the first buffer 6.

In step S23, control is transferred to step S2 in FIG. 8 to perform mark detection again when the full rotation of the form image is completed. When the mark is determined and the orientation of the form can be specified, the process proceeds to step S in FIG.
24, the form is positioned based on the determined mark position.

In step S25, it is determined whether character recognition has been completed for all read fields of the form. If all the reading fields have been completed, the process proceeds to step S29, and if not, the process of step S26 is performed.

In step S26, all characters in the reading field are cut out in units of one character. Step S2
At 7, it is determined whether or not the recognition processing of all the characters has been completed. If all characters have been recognized, the process proceeds to step S25, and if not, the process proceeds to step S28.

In step S28, a recognition process is performed for each character. When the character recognition for all the reading fields is completed, the character recognition result is output in step S29, and the reading of the form is completed.

If it is determined in step S13 that the mark has not been determined after all the rotation directions have been tried, the processing is terminated in step S30 (FIG. 9) assuming that the form reading direction has not been determined.

The above embodiment is an example in which the entire form image is rotated. However, the image rotation processing is performed only on an area where the presence of a mark is predicted in the entire form image (hereinafter referred to as a markable area). Is also possible. Hereinafter, the second embodiment will be described.

First, with respect to a markable area of a form image in the image storage shared memory, a thinned image generation process is performed to detect a mark position. If the mark position cannot be detected, only a markable area portion expected after rotation in the entire form image is extracted and rotation processing is performed.

The thinned image generation processing is performed on the rotated mark area to detect the mark position. Thereafter, the rotation is similarly repeated until the mark position is detected.

In the embodiment described above in which the entire form image is rotated, when the mark position is detected,
The form image of the correct orientation exists in the shared memory for storing images. Therefore, character recognition can be immediately performed using the form image. Another advantage is that the processing is simple. On the other hand, in the case of the second embodiment, after detecting the mark position in the markable area, it is necessary to further rotate the entire form image in a correct direction for character recognition. However, each rotation process has the advantage of being faster as the amount of data to be rotated is smaller.

[0066]

As described above, according to the present invention,
Even when the form image is read without the form set correctly, the mark position of the form can be detected correctly and at high speed. In particular, when performing character recognition of a form image received by a facsimile machine,
Even if there is a human error in the form setting direction, a form image misalignment, a tilt, or the like, the direction of the form can be correctly and quickly specified, so that the effect is large. In general, a form image has an enormous amount of data. However, when rotating a form image, it is possible to rotate the form image with a small amount of memory and specify the orientation of the form.

[Brief description of the drawings]

FIG. 1 is a principle configuration diagram of the present invention.

FIG. 2 is a diagram illustrating a thinned image generation process and a mark position detection process.

FIG. 3 is a diagram illustrating a thinned image generation process and a mark position detection process.

FIG. 4 is a diagram illustrating a division rotation process and a data compression process.

FIG. 5 is a diagram for explaining data expansion processing of divided and compressed data.

FIG. 6 is a diagram showing a configuration example of a character recognition device using the mark position detection device of the present invention.

FIG. 7 is a diagram illustrating another configuration example of the character recognition device using the mark position detection device of the present invention.

FIG. 8 is a processing flowchart of the embodiment of the present invention.

FIG. 9 is a processing flowchart of the embodiment of the present invention.

FIG. 10 is a processing flowchart according to the embodiment of the present invention.

FIG. 11 is a diagram illustrating an example of a form.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 mark position detection device 2 thinned image generation processing means 3 mark position detection processing means 4 image rotation processing means 5 division rotation processing means 6 first buffer 7 data compression processing means 8 second buffer 9 data decompression processing means 11 a to 11 d Form image

──────────────────────────────────────────────────続 き Continued on the front page (58) Fields surveyed (Int. Cl. ⁷ , DB name) G06K 9/00-9/82 G06K 7/00, 7/10

Claims (5)

(57) [Claims] 1. A mark position detecting device for identifying a position of a processing target in a form by detecting a position of a mark written on the form, performing a thinning process in units of a plurality of rows or a plurality of columns of the form image. A thinned image generation processing means for generating a thinned image in which a row or a plurality of columns is one row or one column; and a mark position detection processing means for detecting a mark position in the form image by using the thinned image of the form. A mark position detecting device. 2. The mark position detecting device according to claim 1, wherein the thinned image generation processing means generates the thinned image by taking a logical sum of a plurality of rows or a plurality of columns of a form image. Mark position detection device. 3. A mark position detecting device for specifying a position and an orientation of a processing target in a form by detecting a position of a mark written on the form, wherein the mark position is detected within a predetermined range based on an erect direction of the form image. Mark position detection processing means for detecting a mark position in the form image; form image rotation processing means for rotating the form image; and the form image rotation processing means divides the form image to be rotated, Split image
Division rotation processing means for performing rotation processing on a divided image, and a first buffer for storing the divided and rotated divided image.
And dividing the divided image in the first buffer into a predetermined data pressure.
Data compression processing means for compressing the compressed form image by a compression method, a second buffer for storing the compressed form image, and a compressed form image stored in the second buffer.
Data expanded by a data expansion method corresponding to the compression method
When the mark position in the form image cannot be detected by the mark position detection processing means, the form image is rotated by the form image rotation processing means, and the mark position is detected again by the mark position detection processing means. A mark position detecting device for detecting a mark position. 4. The mark position detecting device according to claim 3, wherein the form image rotation processing means rotates the form image clockwise or counterclockwise with respect to the direction of the original form image. A mark position detecting apparatus characterized in that the form image is rotated by 180 degrees, the form image is rotated by 90 degrees for the second time, and the form image is rotated by 270 degrees for the third time. 5. A mark position detecting method for specifying a position of a processing target in a form by detecting a position of a mark written on the form, wherein a form image to be detected is thinned out in units of a plurality of rows or a plurality of columns. A first step of generating an image, a second step of using the thinned image to detect a mark position in the form image within a predetermined range based on the erect direction of the form image, And a third step of rotating the form image by a rotation angle according to a predetermined priority when the mark position cannot be detected, and repeating the first to third steps until the mark position is detected. And a mark position detecting method.