JP3237906B2 - Inspection equipment for printed matter - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a printed matter inspection apparatus for checking the effectiveness of a printed matter having, for example, two identical character string images.

[0002]

2. Description of the Related Art Conventionally, as to an apparatus for inspecting the printing state of a character in a printed matter in which a character is printed on a pattern, for example, as disclosed in Japanese Patent Application Laid-Open No. 2-56888, a shaded image of a printed character is disclosed. A density histogram is created from the data, an appropriate threshold is determined from the shape,
After performing the value conversion process, character extraction was performed.

[0003]

However, in the above-described conventional method, when the reflection brightness of a character and a pattern are close, the respective density distributions in the density histogram are not separated, and the character is sufficiently extracted. It is difficult to determine a possible threshold value for binarization, and there is a problem that high-precision inspection cannot be performed.

Therefore, according to the present invention, even when the density difference between a character portion and a background portion is small, the threshold value for binarization for extracting a character string can be set to an optimum value, and only a character string can be obtained with high precision. It is an object of the present invention to provide a printed matter inspection apparatus which can perform a high-accuracy inspection by extracting the printed matter.

[0005]

Inspection apparatus printed material of the present invention According to an aspect of the possess two or more the same string, and these characters
At least one of the strings is composed of a character portion and a background portion.
The density difference is large, and the other character strings are
An image input means for collecting image data density difference from a small printed material with the parts, by binarizing the more collected image data to the image input unit, the character portion and the background portion
The first character string extraction unit, as the concentration characteristic of more extracted character string into the first character string extraction unit, a character portion number of pixels for each character to extract character strings is large density difference between the
Density feature calculating means for calculating the total number of pixels of the character part and the character part and the background part in the first character string extracting means
The binarization result of a character string having a large density difference with
Calculated as a density feature by the density feature calculation means.
The number of characters in the character part and the total number of characters in the character part
Based on image data collected by the image input means.
Is binarized to obtain the density between the character portion and the background portion.
A second character string extraction means for extracting the difference is less string,
Said first character string extraction unit and the second character string extraction means respectively recognizing means more extracted the two or more strings to <br/>, said recognized by the recognition unit 2
Matches two or more strings , and each string matches
And determining means for determining that the printed matter is valid .

[0006]

[Function] If the same character string is printed in multiple places,
At least one of the character strings is
The other character strings have a large difference and the text part and the background part
In printed matter with a small density difference between
When binarizing and extracting a character string with a small density difference
Result of binarization of a character string with a large density difference between the
And the number of pixels in the character part and the total number of pixels in the character part for each character
, The threshold value of the binarization for extracting the character string is set to the optimum value even when the density difference between the character part and the background part is small, and only the character string is extracted. Can be extracted with high accuracy. Therefore, it is possible to perform the validity inspection of the printed matter with high accuracy.

[0007]

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows a schematic configuration of a printed matter inspection apparatus according to this embodiment. In the figure, reference numeral 11 denotes a CPU (Central Processing Unit) for controlling the entire apparatus and performing various image processing.
U11 includes an image scanner 12, a disk device 13, and a display device 1 for collecting image data from printed matter.
4. Keyboard 15 for inputting operation instructions and the like, image memories 16, 17, and 18, memory 19 for storing a feature amount of each character in a reference character string, and memory for storing initial values, reference values, and the like. 20 are connected.

The image memory 16 stores the image scanner 12
And stores the original image data collected by
The image memory 17 stores data that has been subjected to high-frequency noise suppression processing of an image input system and filtering processing for enhancing edge contrast. The image memory 18 stores binary image data used for character extraction. The memory 19 stores a feature amount (details will be described later) relating to a reference character string used in binarization.

FIG. 2 shows an example of a printed material to be inspected. In the printed matter 31, for example, the same character string is printed at two places, and one character string 32a printed at the upper left has a large difference in the reflection brightness between the character and the pattern (the character part and the background part have a large difference). The character string 32b printed at the lower right has a small difference in the reflection brightness between the character and the pattern (the character part and the background part have a small difference in density). It is assumed that the character string is generally difficult to extract characters.

Next, taking the printed matter 31 of FIG. 2 as an example, the inspection method of the present invention will be described with reference to the flowchart of FIG. First, the image data of the printed matter 31 is read by the image scanner 12,
The collected density image data (hereinafter, low density means low lightness) is stored in the image memory 16 (S1).

When the collection of the image data is completed, first, the character string 32a at the upper left of the printed matter 31 is extracted and its characteristic amount is calculated (S2). The calculated special amount is stored in the memory 19 and becomes reference data for character extraction of one character string. Here, a method of extracting a reference character string and calculating a very small amount in step S2 will be described.

FIG. 4 is a flowchart showing an outline of a processing procedure for extracting the reference character string and calculating the extraordinary amount. First, a rough area (for example, i × j pixels) including a character string to be extracted is stored in the keyboard 15.
(S8). Then, when a density histogram is created for the grayscale image data in the image memory 16 with respect to the image data in the designated area, two peaks indicating a character portion and a pattern portion generally appear as shown in FIG. Here, for example, by analyzing a density histogram, a density value tm that gives a minimum value of the histogram between two peaks (maximum values) is obtained (S9), and threshold processing is performed on the grayscale image data based on the tm. Extracts the upper left character string 32a as a binary image (S1).
0).

Next, a marginal distribution as shown in FIG. 6 is obtained by accumulating the extracted binarized data in the horizontal direction and the vertical direction with the extracted pixels, respectively (S11).
From the peripheral distribution accumulated in the horizontal direction, for example, the range of the “non-zero” distribution (that is, the range in which even one pixel related to the character image exists) is set as the vertical direction of the character area, and similarly, the area is accumulated in the vertical direction. From the distribution at this time, for example, a range of a plurality of "non-zero" distributions is set as a horizontal direction of each character, and a region of each character is detected as shown in FIG. 7 to perform character extraction (S1).
2).

When the extraction of characters is completed, the number of pixels in the character portion of each character and the total number of pixels in the character string, which is the sum of the characters, are counted as special amounts of characters used in the subsequent binarization processing. This is stored in the memory 19 as a character feature amount (S13).

After extracting the reference character string 32a and calculating the extraordinary amount (the number of pixels) of the character in this manner, the dictionary of the extracted character string 32a is then extracted from the dictionary. Characters are recognized by performing comparison processing with the pattern, and the recognition result is stored in the memory 20 (S3).

When the recognition of the reference character string 32a is completed,
Next, for the character string 32b at the lower right where it is difficult to extract characters due to the closeness of reflection between the character and the pattern, character extraction is performed using the special amount of the reference character string 32a at the upper left stored in the memory 19. Perform (S4). Here, a method of extracting a character string in step S4 will be described.

FIG. 8 is a flowchart showing an outline of a processing procedure for extracting the character string. First, a rough area (for example, i × j pixels) including a character string to be extracted is specified from the grayscale image data stored in the image memory 16 using the keyboard 15 as in step S8 (S14). ).

Next, a filtering process is performed on the image data in the designated area, and the result is stored in the image memory 17 (S15). An example of this filtering processing is, for example, smoothing using a spatial filter of a 3 × 3 mask as shown in FIG. 12, and as shown in FIG. 9, an oblique line containing a high-frequency component existing in the background portion of the character string. This is effective for removing the repetitive pattern P.

When the filtering process is completed, a density histogram is created for the image data in the designated area among the grayscale image data in the image memory 16, and a threshold value for cutting out the character area is determined. However, FIG.
As shown in FIG. 1, in the character string 32b at the lower right, since the reflection lightness between the character and the pattern is close, the density histogram becomes one peak. Therefore, the character strings 32a and 32b
Of the character string 32a previously stored in the memory 19, based on the assumption that the total number of pixels in the character part substantially matches and the density of the character part is lower than that of the background part. Character string 3 for the total number of pixels (i × j) specified in S4
The ratio r% of the total number of pixels in the character portion of 2a is calculated, and a density value which becomes r% from the lower cumulative value of the density histogram is obtained (S16).

The character string 32b is binarized using the density value thus obtained as a threshold value, and the result is stored in the image memory 1.
8 (S17). Thereafter, a marginal distribution as shown in FIG. 10 is obtained from the binary image stored in the image memory 18 in the same procedure as in steps S11 and S12 in which the marginal distribution and each character area are obtained from the binary image of the character string 32a. (S18), each character area is detected (S19).

At this time, the binary image obtained by the uniform threshold value contains a faint character such as the character "M" as shown in FIG.
Therefore, if the threshold level is increased to eliminate blurring of the character, the noise component N overlaps the background of the character “682” as shown in FIG. That is, by binarizing the entire area with a uniform threshold as in step S17, sufficient character extraction cannot be performed. Therefore,
With respect to the grayscale image data after filtering stored in step S15, character extraction is performed by resetting a threshold value for a finer region, for example, for each character.

That is, as in step S16, using the number of pixels of each character of the character string 32a stored in the memory 19, the ratios r1, r2,. rn% (n = the number of characters) is calculated, and the respective density values at which the cumulative value of the density histogram in each character area becomes r1, r2,.
20). Using these density values as threshold values for each character, a binarization process is performed for each character area, the result is stored in the image memory 18, and character extraction ends (S21).

When the character string 32b is extracted in this way, next, similar to the character string 32a, character recognition is performed for each character in the character string 32b, and the result is stored in the memory 2.
0 (S5).

As described above, the character string 32a in which the reflection brightness between the characters and the pattern printed on the printed matter 31 is different and the character string 32b in which the reflection brightness between the characters and the pattern is close and the recognition process are performed. After doing, finally two identical strings 3
The recognition results for 2a and 32b are compared with each other (S
6) For example, the validity of the printed matter 31 is determined such that the printed matter 31 is valid only when all the character recognition results of the two identical character strings 32a and 32b match, for example (S7).

As described above, for example, the same character string is printed in two places, one character string has a large difference in the reflection brightness between the character and the pattern, and the other character string has the reflection brightness between the character and the pattern. When a character string with a small difference in reflection brightness between a character and a pattern is binarized and extracted in a printed matter having a small difference between the two, a binarization result of a character string with a large difference in reflection brightness between a character and a pattern is used. By performing binarization with reference to the number of characters in the character portion and the total number of characters in the character portion for each character, even if the difference in the reflection brightness between the character and the pattern is small, the binarization for character string extraction is performed. By setting the threshold value to an optimal value, only the character string can be extracted with high accuracy. Therefore, it is possible to perform a high-precision inspection of the effectiveness of the printed matter.

In the above-described embodiment, the same character strings are printed one by one in each of a character string having a large difference in reflection brightness between a character and a pattern and a character string having a small difference in reflection brightness between a character and a pattern. However, in addition to the above case, at least one character string is printed even when three or more identical character strings are printed, such as when there are a plurality of character strings with a small difference in reflection brightness between the character and the pattern. The present invention can be applied if a character string having a large difference in reflection brightness between a character and a pattern is printed at a location.

In the above embodiment, the total number of pixels (i ×
j), the ratio r% of the total number of pixels of the character portion of the character string 32a, and the ratio r1, r2,... rn% of the total number of pixels of each character portion to the total number of pixels of each character region (n = number of characters)
Are used, but these ratios r% and r1, r2,
... Rn% can be varied depending on conditions such as the binarization level of the input image data and the reference character string.

Further, in the above-described embodiment, as the characteristic information of each character constituting a certain character string to be referred to, it is determined that the number of pixels (area) of the character part and the density of the character part are higher than that of the pattern part. However, the present invention is not limited to this. For example, geometric features such as the circumference of a character, a character width, a center of gravity, and the number of connected components, and density features such as variance of gray image data of a character portion are used. It can be employed and combined to determine the binarization threshold.

Further, in the above-described embodiment, the case where the grayscale image data is used has been described. However, even when the color image data is input and the color is formed, for example, the R, G, B image data is used. The results of various conversions can be treated as grayscale image data, and the above-listed feature information can be used to determine a binarization threshold.

[0031]

As described above in detail, according to the present invention, even when the density difference between the character part and the background part is small, the binarization threshold value for character string extraction is set to the optimum value. Can,
It is possible to provide a printed matter inspection apparatus capable of extracting only a character string with high accuracy and performing high-accuracy inspection.

[Brief description of the drawings]

FIG. 1 is a block diagram schematically showing a configuration of a printed matter inspection apparatus according to one embodiment of the present invention.

FIG. 2 is a plan view showing an example of a printed material to be inspected.

FIG. 3 is a flowchart illustrating a processing procedure of printed matter inspection.

FIG. 4 is a flowchart showing a processing procedure for extracting a reference character string and calculating a feature amount;

FIG. 5 is a diagram showing a density histogram of a reference character string.

FIG. 6 is a view for explaining calculation of a marginal distribution.

FIG. 7 is a view for explaining extraction of a character area.

FIG. 8 is a flowchart showing a processing procedure of character string extraction.

FIG. 9 is a view for explaining the effect of the filtering process.

FIG. 10 is a view for explaining calculation of a marginal distribution.

FIG. 11 is a diagram illustrating a relationship between a density histogram of a character portion and a threshold.

FIG. 12 is a diagram showing an example of a 3 × 3 mask used for filtering processing.

FIG. 13 is a diagram showing an example of binarization of a character.

[Explanation of symbols]

11: CPU, 12: Image scanner, 16, 17,
18 image memory, 19, 20 memory, 31 printed matter, 32a, 32b character string.

──────────────────────────────────────────────────続 き Continued on the front page (58) Fields investigated (Int. Cl. ⁷ , DB name) G06K 9/00-9/82 G06T 1/00 G07D 7/12 G07D 7/20

Claims (1)

(57) [Claims] 1. A possess two or more the same string, and this
At least one of these strings is a character part and a back
The density difference from the scenery is large, and other character strings
Binarization parts and an image input means for collecting image data density difference from a small printed material with the background portion, the image data more collected in the image input means
By doing so, the density difference between the character part and the background part is large.
A first character string extraction means for extracting a character string have, the first as a concentration feature amount of a character string that is more extracted character string extraction unit, a character portion number of pixels and statements for each character
A density feature calculating means for calculating the total number of pixels of the character part, and a character part and a background part in the first character string extracting means.
The binarization result of a character string having a large density difference, and the density
Calculated as the density feature by the statistical feature calculation means
Based on the number of characters in the character part of each character and the total number of pixels in the character part
The image data collected by the image input means
By converting into a value, the density difference between the character portion and the background portion is reduced.
Recognition and second character string extracting means for extracting a small string, the first character string extraction unit and the second character string extraction means <br/> more extracted the two or more strings, respectively a recognition means for, the two or more character string recognized by the recognition means
Collate, and if the character strings match, the printed
Inspection apparatus printed matter, characterized by comprising determination means that the effective, the.