JPH0973516A - Slip type identification method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a slip type identification method by which a type can stably be identified even in the case of the overlap of line noise and the cut of a segment through the use of an existed slip as it is and the initial registration of a format is easy. SOLUTION: A slip type identification part corrects the inclination of the input picture of a scanner and the like by a first means and a second means roughly meshes an input means on an unknown format slip. A third means integrates the extracts a background area becoming a closed area and a fourth means generates the circumscribing rectangle of the background area. A fifth means integrates the overlapped circumscribing rectangles. A sixth means checks the geometric relation of the respective circumscribing rectangles, groups them and generates a rectangle block. An eighth means calculates an adaptability with format template data which is read by a seventh means with the rectangle block as a unit and the pertinent format is decided from the adaptability. A format initial registration part executes a processing by the first to sixth means on an unfilled format slip and a ninth means sets obtained rectangle block data to be format template data.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a computer system or the like for efficiently inputting data in a fixed format such as a form, in which line noise is superimposed on the type of the form surrounded by a frame line such as the form. The present invention relates to a form type identification method for stable identification even when there is blurring or cutting in a line segment.

[0002]

2. Description of the Related Art There are roughly two types of conventional form type identification methods.

The first is to use a special mark (symbol, bar code, etc.) that indicates the type of form. In the case of this method, the work of registering the format is easy because it is only necessary to register the correspondence between these marks and the form types. In addition, it is relatively stable against line noise superimposition and blurring.

The second is to extract the frame lines that form the format,
This is a method of identifying the form type by checking the structure of the frame line. This method has the advantage that existing forms can be used as is, and there is great demand.

[0005]

Among the above-mentioned conventional form type identification methods, a dedicated mark (symbol,
With the first method using a barcode, etc., the existing form cannot be used as it is because the dedicated mark is used, and it is necessary to reprint the form with the mark. When there are many, there was a problem that labor and cost would increase.

The second method for identifying the form type by extracting the frame lines forming the format and examining the structure of the frame lines is the case where line noise is superimposed or line segments are cut. Had a problem that the structure of the frame line changed greatly and the identification became unstable. Further, in order to accurately register the structure of the frame line, registration work (drawing) using CAD or the like is required at the time of initial registration of the form, and when there are many types of forms, the burden on the operator is large. However, there is a problem that the initial registration of the format necessary for identifying the format is complicated.

In order to solve the above problems, the present invention can use an existing form as it is without using a dedicated mark, and is stable even in the case where line noise is superposed, line segments are blown, or cut. It is an object of the present invention to provide a form type identification method that enables automatic form type identification and facilitates initial registration of forms.

[0008]

In order to achieve the above-mentioned object, the first invention according to the present invention is to determine the form type of the input form from the form image of the unknown form input by the image input means. A method for identification, in which a rough mesh is applied to the input form image of the unknown format, all background areas that are closed areas are extracted from the image applied with the mesh, and the circumscribing of each extracted background area is performed. A rectangular block generation step of unknown format, which generates a rectangle, examines one or both of the size and geometrical relation of the generated circumscribed rectangle and groups similar circumscribed rectangles to generate a rectangular block; Calculates the conformity with the previously created and saved format template data using the rectangular block data as a unit, and determines the format as the corresponding format when the conformity is high. A document type identification method characterized by comprising the steps, a.

Similarly, in the second aspect of the present invention, in the above-described form type identification method, a step of inputting a blank form image by image input means and a coarse mesh on the input blank form image. , All background areas that are closed areas are extracted from the image to which the mesh is applied, circumscribed rectangles of the extracted background areas are generated, and the size and geometrical relative of the generated circumscribed rectangles are generated. A blank block generation step of generating rectangular blocks by grouping similar circumscribing rectangles by examining one or both of the relationships, and generating format template data that describes the characteristics of the format in the generated rectangular block units. And a format initial registration step of saving and storing the format.

According to a third aspect of the present invention, in the form type identification method described above, in the format determining step, rectangular block data of unknown format and format template data are spatially superposed, and a rectangular block with the unknown format is obtained. When the representative point of is within the range of a certain rectangular block in the format template data, it is assumed that those rectangular blocks have a corresponding relationship, and the overlapping area of all the rectangular blocks having the corresponding relationship is obtained, The present invention is characterized by including a compatibility calculation step of calculating a compatibility between the rectangular block data of the unknown format and the format template data using the value of the obtained area.

In the form type identification method according to the first invention,
The existing form can be used as it is by identifying the form type by matching it with the format template data and without using the dedicated mark. Also, the work load on the operator can be improved by automatically describing the structure of the format surrounded by the fine frame in a large unit called a rectangular block and identifying the form type by matching with the format template data in that unit. Greatly reduce. Also, regarding the superposition of line noise, even if the rectangular block is divided due to line noise,
By performing rough matching in large units called blocks, the divided blocks are associated with the same blocks in the format template data so that the goodness of fit value is almost the same as when there is no line noise. Achieves stable discrimination against noise superposition. Furthermore, regarding the cutting of line segments, small cutting is solved by coarsely meshing, and for locally large line cutting, the structure of a fine format is not described, but it is roughly divided into large units called blocks. Such matching makes it difficult for the influence of the cutting of the line segment to spread to the whole, and realizes stable identification against the cutting or blurring of the line segment.

In the form type identification method according to the second invention,
By automatically describing the structure of a format enclosed by a fine frame in a large unit called a rectangular block and easily registering the initial format, the work load on the operator at the initial registration of the format is greatly reduced.

In the form type identification method according to the third invention,
In matching with the format template of rectangular block unit, even if the rectangular block of the unknown format form is divided due to the superposition of line noise, the correspondence between the blocks is the format template data including the representative points of the rectangular block of unknown format. By assuming that the blocks have a correspondence relationship, the divided blocks are made to correspond to the same block of the format template data, and the fitness value is calculated by using the area of the overlapping portion to calculate the fitness value. Is set to have almost the same value in the case where there is no line noise and the case where there is line noise, thereby realizing more stable discrimination against the superposition of line noise.

[0014]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings for simplification of description.

FIG. 1 is a flow chart of processing showing an embodiment of the present invention. The example of the present embodiment includes two forms, a form type identification unit (FIG. 1A) and a format initial registration unit (FIG. 1B).

The form type identification unit (FIG. 1A) first corrects the inclination of an image input from a scanner or a facsimile (hereinafter, FAX) by the first means 1 for a form of an unknown format. Next, the second means 2 coarsely meshes the input image, then the third means 3 extracts a background area which is a closed area, and then the fourth means 4 extracts a circumscribed rectangle of the background area. And then integrate the circumscribed rectangles by the fifth means 5 and then the sixth means 6
The size and geometrical relationship of the respective circumscribing rectangles are checked to group similar rectangles having a parallel relationship to generate a rectangular block, and then the format template data previously acquired by the seventh means 7 is filed. Then, the compatibility with all the format template data is calculated by the eighth means 8 in units of rectangular blocks,
Finally, the applicable format is determined from the high degree of conformity.

On the other hand, the format initial registration unit (FIG. 1B) is
With respect to the blank form, the processes of the first to sixth means are executed to generate a rectangular block, and the obtained rectangular block data is used as the form template data to form a ninth block.
It is saved in the file by means 9 of.

Each of the above processes will be described in detail below.

First, the portions (first to sixth) common to the form type identification unit (FIG. 1A) and the format initial registration unit (FIG. 1B).
Means) will be described.

First, the form is read by a scanner or FAX. At this point, the input image has some inclination due to instability of the paper feed at the time of reading. The method of calculating the angle of inclination from the outer frame of the form, the method of calculating using a histogram, and the like are already known, and since they can be used, they will not be described in detail here. The entire image is rotated based on the calculated angle to correct the tilt. Next, a coarse mesh is applied to the input image to remove the effects of small blurring and cutting. The rougher the mesh, the less the influence of cutting line segments, but this method uses the information of the background area to identify the type of format, so set the roughness of the mesh within the range in which the background area is extracted. Must.

Here, FIG. 2A shows an example in which the mesh is too fine to eliminate the influence of cutting the line segment, and FIG. 2B.
Has an appropriate mesh roughness, eliminating the effects of cutting,
An example in which the background area (corresponding to the entry frame) also appears, Fig. 2
(C) shows an example in which the mesh is too coarse and the background area is crushed. The roughness of the mesh is set empirically. In this way, the effect of small cutting can be eliminated by applying a mesh having an appropriate roughness. The influence of large cutting can be absorbed by the block-based matching described later.

Next, as shown in FIG. 2 (b), all the background areas which are closed areas are extracted from the image (FIG. 3 (a)) in which the mesh having an appropriate roughness is applied (FIG. 3 (b)). The extraction of the background area can be realized by tracking the outer edge of the white pixel. Next, a circumscribed rectangle of each of the extracted background areas is generated (FIG. 3 (c)).

Next, the circumscribing rectangles having a superimposing relationship satisfying a certain condition are integrated among these circumscribing rectangles. FIG. 4 shows how the circumscribed rectangles are integrated. In this circumscribing rectangle integration processing, since the character and the frame line are in contact with each other in the meshed image as shown in FIG. 4A, the background region in one entry frame is divided and extracted. If it happens (Fig. 4
This is to deal with (b)). In such a case, the circumscribed rectangles of the two divided background areas overlap each other (FIG. 4C). Therefore, when the circumscribing rectangles have a superimposing relationship and the respective vertical ranges substantially match, the circumscribing rectangles are integrated (FIG. 4D).

Next, a plurality of circumscribing rectangles having the same size and arranged in the vertical direction (having substantially the same horizontal range) are grouped into one block. FIG. 3D shows a state of blocks obtained as a result of grouping parallel similar rectangles.

The above is a description of the portions (first to sixth means) common to the form type identification unit and the format initial registration unit. Hereinafter, description will be made in the order of (1) format initial registration unit and (2) form type identification unit.

(1) Format Initial Registration Section FIG. 5 shows the result of applying the above processing to an empty format (blank format). 5A is an example of an empty format, and FIG. 5B is an example of extracting a background image.
FIG. 5C is an example of generating a circumscribed rectangle of the background image.
(D) is an example of generation of a rectangular block. Here, the operator confirms the generation result of the rectangular block, assigns the ID (identification number, etc.) of the format, and saves the rectangular block data as format template data.
Create one for each format.

(2) Form Type Identification Unit The form type identification unit matches the rectangular block of the input unknown form document image with the initially registered format template data in units of rectangular blocks. This is shown in FIG. That is, in the form type identification unit, the rectangular block data obtained by extracting the circumscribing rectangle from the input image of the form in the unknown format, and the format template data initially registered by the input image of the form in the empty format The matching degree between the two is calculated in units of rectangular blocks, and the format with the highest matching degree is set as the relevant format. As a method of calculating the goodness of fit,
It can be realized by various methods such as a method of describing the geometrical relative relationship of blocks to perform logical matching, a method of comparing the shapes and sizes of blocks, but here, the area of the overlapping part of blocks is The method using is described.

First, the correspondence between the rectangular block obtained from the input form image of the unknown format and the rectangular block of the format template data is examined. FIG. 7 shows how the rectangular blocks are associated. The correspondence is such that the reference points (upper left coordinates, etc.) are overlapped with each other, and a representative point of a rectangular block in an unknown format (a representative point is one point representing the rectangular block, and the center of gravity or the like may be used. )
It is assumed that the rectangular blocks of the format template data including the are in correspondence. (Here, note that the correspondence between blocks is (unknown format → format template)).

When the correspondence of each block is obtained, the following areas are calculated as shown in FIG.

Su: total sum of block areas in unknown format, St: total sum of block areas in format template data, So: total sum of area of corresponding portions where the corresponding blocks overlap.

The goodness of fit is defined as a function of these areas (goodness of fit F = F (Su, St, So)). As a specific example, the goodness of fit can be calculated using F = (So / Su) * (So / St) or the like.

By the way, since it is possible that the input direction when inputting from a FAX or a scanner is deviated to four directions of 0, 90, 180, and 270 degrees, when calculating the compatibility, each block data is divided into four directions. It suffices to rotate the direction to calculate the fitness and adopt the direction with the highest fitness.

Furthermore, it is conceivable that the reference point (the upper left point in the above description) may become unstable due to the effects of noise superimposition, blurring, and the like. In order to deal with this, the reference points may be matched with respect to the four points of upper left, lower left, upper right and lower right, and the one having the highest degree of matching may be adopted.

The above is the form type identification process.

Finally, the reason why the form type identification method of the present invention is stable against superposition of line noise and cutting of line segments will be described.

First, the reason why the line noise is stable against superposition will be briefly described with reference to FIG. FIG.
9A shows generation of format template data, FIG. 9B shows generation of a rectangular block in the absence of line noise, and FIG.
(C) is generation of a rectangular block when line noise is superimposed. In FIG. 9C, the rectangular blocks are divided by the superposition of line noise, but the correspondence between the blocks is assumed to be the block of the format template data including the representative point of the block of the unknown format. Therefore, the divided blocks are associated with the same block of the format template data. The value of the goodness of fit calculated using the area of the overlapping portion is almost the same in the case where there is no line noise and the case where there is no line noise. This is the reason why the method is stable against the superposition of line noise.

Regarding the cutting of the line segment, the small cutting can be solved by coarsely meshing. Furthermore, in this method, since the structure of a detailed format is not described and rough matching is performed in a large unit called a block, the influence of local line segment breakage is unlikely to spread to the whole. Therefore, it is stable against blurring and cutting of line segments.

[0038]

As is apparent from the above description, according to the present invention, the type of a format such as a form surrounded by a frame line is overlapped with line noise by a simple initial format registration. When the line segment is blurred or cut, it is possible to stably identify the line segment, and it is possible to significantly reduce the work load on the operator at the time of initial registration of the form and the processing of the form.

[Brief description of drawings]

1A and 1B are flow charts of processing showing an embodiment of the present invention.

2A, 2B, and 2C are diagrams showing a state of an image to which a mesh is applied in the above-described embodiment.

3 (a), (b), (c), and (d) are diagrams showing a background area extraction, a circumscribed rectangle generation, and a rectangular block generation in the above-described embodiment.

4 (a), (b), (c), and (d) are diagrams showing a state in which overlapping circumscribed rectangles are integrated in the above-described embodiment.

5 (a), (b), (c), and (d) are diagrams showing a state up to block data generation in an empty format at the time of format initial registration in the above-described embodiment.

FIG. 6 is a diagram showing a state of matching in block units at the time of identifying a form type according to the above-described embodiment.

FIG. 7 is a diagram showing how blocks are associated with each other when the form type is identified in the above embodiment.

FIG. 8 is a diagram showing how the area is calculated when the degree of conformance is calculated when the form type is identified in the above embodiment.

FIG. 9 is a diagram showing a state in which line noise is superimposed at the time of identifying a form type according to the above embodiment.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... 1st means which corrects the inclination of an input image 2 ... 2nd means which applies a coarse mesh 3 ... 3rd means which extracts a background area 4 ... 4th means which produces | generates a circumscribed rectangle 5 ... Superimposed circumscribing Fifth means for integrating rectangles 6 ... Sixth means for grouping circumscribed rectangles to generate blocks 7 ... Seventh means for reading a form feature data file 8 ... Eighth means for calculating conformance 9 ... Ninth means for saving rectangular block data in a file as format template data

Claims (3)

[Claims] 1. A method of identifying the type of format of the input form from the form image of the unknown format input by the image input means, wherein the input form image of the unknown format is roughly meshed. All background areas that are closed areas are extracted from the image on which the mesh is applied, and a circumscribed rectangle of each of the extracted background areas is generated. One of the size of the generated circumscribed rectangle and a geometrical relative relationship is generated. Alternatively, a rectangular block generation step of unknown format in which both are examined to group similar circumscribed rectangles to generate a rectangular block, and the format template data generated and saved in advance with the generated rectangular block data as a unit A form type identifying method comprising: a format determination step of calculating a goodness of fit and determining that the format is a relevant format when the goodness of fit is high. 2. A step of inputting a form image in a blank form by an image inputting means, and a step of coarsely meshing the input form image of the blank form, and forming all closed areas from the meshed image. A background area is extracted, a circumscribed rectangle of each of the extracted background areas is generated, and one or both of the size of the generated circumscribed rectangle and the geometrical relative relationship is examined, and similar circumscribed rectangles are grouped into rectangles. And a step of generating a rectangular block of an unfilled format for generating a block, and a step of initial registration of a format for generating and storing format template data describing the characteristics of the format in units of the generated rectangular blocks. The form type identification method according to claim 1. 3. The format determining step spatially superimposes rectangular block data of unknown format and format template data, and a representative point of the rectangular block having the unknown format is a range of a rectangular block in the format template data. If the rectangular blocks correspond to each other, the rectangular block data of the unknown format is obtained by calculating the overlapping area of all the rectangular blocks having the corresponding relationship and using the value of the calculated area. 3. The form type identification method according to claim 1 or 2, further comprising a matching degree calculating process for calculating a matching degree between the format template data and the format template data.