JPS5818666B2 - mask - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention provides a mask matching character recognition system, particularly a character recognition system that matches an observed pattern with a standard mask pattern. A plurality of line width-changed mask patterns obtained by changing the character line width of the standard mask pattern are matched with the observed pattern, and matching is performed using a standard mask pattern with so-called weight variation. The present invention relates to a mask matching character recognition method that achieves similar results while reducing the storage capacity of a storage device.

In reading printed characters, a so-called pattern matching method is generally widely adopted.

That is, a method is adopted in which the observed pattern of the character to be recognized is matched with a standard mask pattern prepared in advance, and the degree of mismatch (or M fJJ degree) between the two is determined.

However, if the observed pattern is obtained from characters with low printing quality, the detection of the degree of mismatch may be greatly affected by partial overlap, blurring, blurring, etc. when matching with the standard mask pattern. be.

For this reason, the standard mask pattern is sometimes weighted.

That is, the observation pattern 1 as shown in FIG. 1 is matched with the standard mask pattern 2 as shown in FIG.
Detect the degree of mismatch between the two.

In this case, in order to carry out the weighting as described above, the standard mask pattern 2 is given weights for the black character area and the white background area, respectively.
1, 2, and 3 are given so that they become larger as the distance from the boundary between the black area and the white area increases.

The area near the boundary is easily affected by the above-mentioned blurring, blurring, blurring, and even slight positional deviation, so the weight is
0'' is given so that even if a mismatch occurs at this location, it will not be counted as a mismatch.

However, although the weighted standard mask pattern as shown in FIG. 2 above is to be stored in a storage device, when storing the information of one mesh of the standard mask pattern, as shown in FIG. A total of three bits are required: one bit to indicate whether the mesh is a black mesh or a white mesh, and two bits to indicate the weight.

That is, for example, 3 bits are required to store one mesh of information.

When recognizing only characters such as alphabets, numbers, and katakana as shown in Figure 2, the character strokes are simple, the number of meshes expressing one character is small, and the number of character cylinders is small, so the above Even if it was stored in a storage device in the form of 1 mesh of 3 bits, it did not pose much of a problem.

However, when recognizing complex characters such as kanji, it requires 4 to 9 times as many meshes as the numbers mentioned above to represent one character, and the number of character cylinders is also 2. 000 to more than 10,000.

Therefore, if an attempt is made to store the data in the form of 1 mesh of 3 bits, the required storage capacity will be enormous, and the storage capacity will be three times that of storing only the standard mask pattern without weighting.

The present invention aims to solve the above points,
By storing only standard mask patterns with no weight on the storage device, the storage capacity is prevented from increasing.
On the other hand, it is an object of the present invention to adopt a method of changing the character line width of a standard mask pattern to obtain an effect substantially similar to that of pattern matching in which weights are shifted.

Therefore, the mask matching character recognition method of the present invention stores the standard mask pattern in the mask matching character recognition method that matches the observed pattern of a given character to be recognized with a standard mask pattern prepared in advance. and a line width changing processing means for changing the character line width of the standard mask pattern, and a plurality of line width changing mask patterns in which the observed pattern and the line width obtained by the line width changing processing means are changed. This feature is characterized in that the matching results are accumulated to obtain the overall matching.

This will be explained below with reference to FIG. 4 and subsequent figures.

FIG. 4 is an explanatory diagram for explaining the concept of a character recognition method according to the present invention, FIG. 5 is an explanatory diagram for explaining the configuration of one embodiment of the present invention, FIG. FIG. 7 shows the configuration of an embodiment of the system shown in FIG.

In the case of the present invention, one standard mask pattern, ie, standard mask pattern 3 expressed in black and white binary values without weight reduction, is stored in the storage device.

In the present invention, in order to match the observed pattern,
Read out the standard mask pattern 3 above, and
Processing to thicken and/or reduce the character line width of pattern 3 is performed to obtain line width change mask patterns 4 to 7.

Then, each of these line width changing mask patterns A to 7 is matched with the observed pattern.

That is, assuming that the currently observed pattern and the standard mask pattern are patterns with character partial points, (1) The character line width of standard mask pattern 3 is thickened, and its black and white are reversed. ] Create mask pattern 4 with thicker line width.

The black area of pattern 4 (the area marked with 9 points in the figure) is
In other words, it corresponds to an area with a weight of "1" or more in the white area (area other than characters) shown in FIG.

In the case of the present invention, as will be described later, when matching the first thick mask pattern 4 with the observed pattern, meshes that do not match the observed pattern within the black area of the mask pattern 4 are Count the number (count value N1
).

In this way, it is observed to what extent the observed patterns are inconsistent within the white area shown in FIG. 2, which has a weight of "1" or more.

That is, the number of mismatches in the area with weight "1" in the white area shown in FIG. 2, the number of mismatches in the area with weight "2", the number of mismatches in the area with weight "3", etc. ... will be counted.

(2) The character line width of the standard mask pattern 3 is increased twice to create a second line width mask pattern 5 with its black and white inverted.

The black area of pattern 5 (the area marked with 9 points in the figure) is
In other words, it corresponds to an area with a weight of "2" or more in the white area (area other than characters) shown in FIG.

In the case of the present invention, as will be described later, when matching the second thicker mask pattern 5 with the observed pattern, there is a mismatch with the observed pattern within the black area of the mask pattern 5. Count the number of meshes (count value is N
2).

In this way, it is observed to what extent the observed patterns are inconsistent within the white area shown in FIG. 2 with a weight of [-2'' or more.

That is, the number of mismatches within the area of weight "2" in the white area shown in FIG. 2, the number of mismatches within the area of weight "3",
... will be counted.

When all the above count values N1 and N2 are summed, the count value sum ΣNi represents the following.

That is, the observed pattern is matched with the white area of the weighted standard mask pattern as shown in FIG.
If a mismatch occurred in an area given a weight of ``2'', the number of meshes was counted once, and if a mismatch occurred in an area given a weight of ``2'', the number of meshes was counted twice. It turns out.

If a third mask pattern with a thicker line width is prepared, the number of mismatched meshes in the area given the weight "3" in the white area of the weighted standard mask pattern as shown in FIG. However, it goes without saying that this means that it has been counted three times.

(3) Perform processing to narrow the character line width of the standard mask pattern 3 to create a first narrow line width mask pattern 6.

The black area of pattern 6 (the dot in the figure; the dotted area) is
In other words, it corresponds to an area with a weight of "1" or more in the black area (character area) shown in FIG.

In the case of the present invention, as will be described later, when matching the first narrow line width mask pattern 6 with the observed pattern, meshes that do not match the observed pattern within the black area of the mask pattern 6 are Count the number (count value Ml
).

In this way, the weight in the black area shown in FIG. 2 is greater than or equal to "1".

The extent to which the observed patterns are inconsistent within the region is observed.

In this case, the weights "2" and "3" in the black area shown in FIG.
It goes without saying that the number of mismatches within the area of `` is also counted as 1-.

(4) The character line width of the first line width narrowing mask pattern 6 is further narrowed to create a second line width narrowing mask pattern 1.

The black area of the pattern 7 (the area marked with 9 dots in the figure) is, so to speak, in the black area (character area) shown in FIG.

It corresponds to a region with a weight of [2] or more.

In the case of the present invention, as will be described later, when matching the second narrow line width mask pattern 7 with the observed pattern, there is a mismatch with the observed pattern within the black area of the mask pattern 7. Count the number of meshes (count value M
2).

In this way, it is observed to what extent the observed patterns are inconsistent within the black area shown in FIG. 2, which has a weight of "2" or more.

In this case as well, it goes without saying that the number of mismatches in the area of weight "3" (if any) in the standard mask pattern as shown in FIG. 2 is also counted.

When the above count values M1 and M2 are summed, the count value sum ΣMi is matched with the black area of the weighted standard mask as shown in FIG. Therefore, one time was counted, two times were counted corresponding to the weight "2", and so on.

From the above, in the case of the present invention, by matching the observed pattern with each of the above-mentioned mask patterns 4 to 7, and adding the mismatch count values (ΣNi + ΣM1) to achieve overall matching, it is possible to substantially match the second This is equivalent to matching the weighted standard mask pattern 2 as shown in the figure.

And despite this, the storage capacity for storing standard mask patterns can be compromised.

FIG. 5 shows one implementation f1M configuration that performs the process shown in FIG.

In the figure, 1 represents an observation pattern, and 3 represents a standard pattern in which no weight N-subtraction is performed.

8-1 and 8-2 are respectively thick processing units which scan the original pattern using an observation window consisting of, for example, a 3x3 mesh, and the central mesh of the observation window is a white mesh point, and the surrounding 8 meshes are For example, if there is a one-mesh black mesh point, it can be considered that a known process such as changing the center mesh to a black mesh point is performed.

Further, 9-1 and 9-2 are narrow processing units, respectively, which use observation windows similar to those in the thick processing unit above, and whose center mesh is a black mesh point, and on the 8 meshes around it, for example, 1 mesh or more. If a white mesh point exists, it may be considered that a known process such as changing the center mesh to a white mesh point is performed.

10-1 to 10-4 are mismatch counters that count the number of mismatched meshes within the black area in each of the mask patterns A to 7 shown in FIG.

Reference numeral 11 is an adder for each of the above-mentioned mismatch counters 10-1 to 1.
Add the contents of 0-4.

Further, 12-1 to 12-4 each represent a pattern reversal processing section.

In the case shown in FIG. 5, the output of the pattern reversal processing section 12-1 is a first thick mask pattern 4 as shown in FIG.
is obtained, the second thicker mask pattern 5 as shown in FIG. 4 is obtained as the output of the pattern inversion processing section 12-2, and the second thicker mask pattern 5 as shown in FIG. 4 is obtained as the output of the narrowing processing section 9-1. 1st
A line width thinning mask pattern 6 is obtained, and the thinning processing section 9
-2 output with a second line width narrowing mask as shown in Figure 4.
Pattern 1 is obtained.

Then, each of these mask patterns 4 to 7 is matched with observed pattern 1, and the number of mismatched meshes is counted by each of mismatch counters 10-1 to 10-4, and added up by an adder 11.

In Figures 4 and 5 above, it is shown that the thinning process is performed by the thinning processing section, but thickening processing is generally easier than narrowing processing, and two types of circuits are required for thickening processing and thinning processing. However, since this is done using only one type of circuit, a method can be adopted in which the character line width is changed only by thickening processing without performing thinning processing.

FIGS. 6 and 7 show an embodiment of the present invention using only the thickening process.

In the case of this embodiment, a mask pattern 3 with a standard line width without weight deviation as shown in the top position of FIG. 4 is used as shown in FIG.
Narrow mask pattern 13 with narrow character line width as shown
is stored in the storage device 1 as a new standard pattern.

Note that the mask pattern is created in advance so that the narrow mask pattern 13 as shown in FIG. I'll keep it.

That is, it is not necessary to store the standard mask pattern 3 just for performing mask matching character recognition according to the present invention.

In the case of this embodiment, the character line width is increased twice from the thin mask pattern 13 as shown in FIG.
(If the pattern 13 does not have a chip like the mask pattern 6 shown in FIG. 4), the character line width is increased and the black and white are reversed.

Create a thick mask pattern 14.

Furthermore, the second thicker mask pattern 15 is created by making it thicker once more and inverting the black and white.

Then, the observed pattern and each of the mask patterns 13,
14°15.

In this case, since there is only a narrow mask pattern represented by pattern 13, the area in the black area shown in FIG. 2 with a weight of "2" is also considered to have a weight of "1".

In Figure 7, the symbol 8-1.8-2゜10-1 in the figure
．． 10-2.10-3.11.12-1°12-2 corresponds to FIG. 5, and 16.17 represents a time delay circuit.

In the case shown in Figure 7, the "standard pattern" is shown in Figure 6.
The output of the pattern reversal processing section 12-1 corresponds to the narrow mask pattern 13 as shown in FIG.
The output of the pattern inversion processing section 12-2 corresponds to the second thick mask pattern 15 shown in FIG. 6C.

Then, each of these mask patterns 13, 14, 15 is matched with the observed and measured patterns, and each mismatch state is counted and summed by an adder 11.

The larger the total value, the greater the mismatch degree.

As explained above, according to the present invention, in the storage device,
Standard mask pattern without weight reduction only.
(2) Alternatively, it is sufficient to store a standard mask pattern and a narrow mask pattern that are not weighted.

Then, it is possible to obtain an effect substantially equivalent to that of pattern matching in which weighting is performed.

[Brief explanation of the drawing]

Figures 1 to 3 are explanatory diagrams explaining the storage state of standard mask patterns with weights used in conventionally known character recognition methods, and Figure 4 explains the concept of the character recognition method according to the present invention. 5 is an explanatory diagram illustrating the configuration of one embodiment of the present invention, FIG. 6 is an explanatory diagram illustrating the concept of another embodiment according to the present invention, and FIG. 7 is an embodiment of the method shown in FIG. 6. Show the configuration. In the figure, 3 is a standard mask pattern without weight shift, 4 to 7 and 13 to 15 are mask patterns for changing the line width, and 81. B-2.9-1.9-2 respectively represent line width changing processing means.

Claims (1)

[Claims] 1 In a mask matching character recognition method that matches the observed pattern of a given character to be recognized with a standard mask pattern prepared in advance, the standard mask pattern is memorized, and the character line width of the standard mask pattern is and aligning the observed pattern with a plurality of line width changing mask patterns whose line widths have been changed obtained by the line width changing processing means, and using the matching results. A mask matching character recognition method characterized by accumulating and obtaining overall matching.