JPS581784B2 - Character processing method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a character processing device that expresses characters, etc. in a matrix.
The present invention relates to a character processing method in which unknown points (hereinafter referred to as predicted points) are determined based on surrounding conditions at the stage of reproducing such a pattern, and reproduction close to the original pattern is performed.

In character processing devices, especially in kanji processing devices, each character shape of a kanji is decomposed into a dot matrix of 16 x 16, 22 x 24, 32 x 32, etc., and each element is “white” or “0” without forming a single dot. Generally, it is stored in two values: 1 and "black"("1"), which forms one dot.

However, the characteristics of kanji are not only that the patterns themselves are very complex compared to numbers and alphabets, but also that there are many types of characters, ranging from 3,000 to 10,000.

Furthermore, if a certain degree of faithful reproducibility is required for the shape of Chinese characters, the number of dots will be increased, that is, a dot matrix of approximately 32×32 will be required.

Therefore, the number of dots per Kanji character is 1024, and the storage capacity is 1024 bits per character.

Therefore, in order to express all the kanji characters, a large capacity memory of 3M to 10 IOM bits is required.

The present invention aims to provide a character processing method that can express all Chinese characters and other characters by compressing the memory capacity as much as possible.

Traditionally, pattern memory compression is run-length (RUN-
The mainstream is the LENGTH) method, which stores the number of consecutive white or black dots in the scanning direction, and the number of black or white dots per character changes depending on the complexity of the character. The degree of compression changes (indefinite length compression).

Therefore, since there is no fixed relationship between the code assigned to each character and the address of the memory that stores the glyph information, there is a separate Memory is required.

Considering the total number of bits including this memory, the compression ratio of the memory is only about 80 bits.

Now, if we use a 32 x 32 dot matrix to represent kanji, the number of dots per character is 1 as mentioned above.
024 pieces of information, and if you try to remember all of this information, it will take 1
A storage capacity of 0.024 bits is required.

However, this information contains a lot of redundancy.

Of these 1024 bits, one pixel (
By storing data alternately (dots) and in each row in a checkerboard pattern, the number of bits becomes 512, and the compression ratio of the memory becomes constant at a factor of 50 (constant length compression).

Therefore, the information density stored in memory doubles at once.

However, if information is extracted from memory and then expressed as is, it will only be expressed in a checkered pattern as shown in FIG. 1, and the pattern will not be the same as the original information, leaving a problem.

When stored in a checkered pattern, it is necessary to reproduce the information of the portion removed from the original pattern (the portion marked with an x in FIG. 1) by some method to obtain 1024 bits of information as a whole.

However, if the information of the removed part is not obtained from the original pattern, the method of reproducing the information of the removed part may cause a prediction error in the reproduction unit, that is, an error if the information is reproduced differently from the information of the original pattern. The rate is determined.

Therefore, the present invention provides a simple reproduction method to reduce the above error rate.

For example, a portion removed from the original pattern, hereinafter referred to as a predicted point, is reproduced by inferring and determining the information (white or black) of the predicted point depending on the surrounding situation of the predicted point.

In this case, characters are classified into several groups, and prediction point information is inferred and determined for each common character group.

Conventionally, in a facsimile, information is compressed and transmitted as shown in Figure 1, and at the stage of reproduction, the predicted point marked with an x is selected according to the situation of the surrounding four sampling points (known pixels) on the upper, lower, left, and right sides. There is a method that infers and determines whether the pixel is a black pixel "1" or a white pixel "0".

However, when the pattern to be handled is kanji, it is complicated, so
It is not practical to use the above method to determine and reproduce each predicted point marked with an x.

Therefore, in the present invention, about 3000 kanji, about 1
As a result of a statistical investigation of 200,000 cases, the second
As shown in the figure, for example, the surrounding 12-bit sampling point (
It was found that the method of inferring the central prediction point (marked with ■ in the figure) using the ● mark in the figure is the most realistic and yields good results.

FIG. 3 is a block diagram showing the configuration of a Kanji printer according to the present invention, and the basic system of the present invention will be first explained with reference to this figure.

Also, let us assume that the surrounding 12 bits are taken for prediction as shown in FIG. 2C.

In the figure, 1 is an LS consisting of a microcomputer.
This is an arithmetic processing unit for I elements, and will be referred to as a CPU hereinafter.

Reference numeral 2 denotes a program memory in which various instructions are written to operate this device, and a program is set up using the various instructions mentioned above.

That is, the CPU 1 sequentially reads instructions from the program memory 2, executes processing according to the commands of the instructions, and executes processing such as pattern reproduction by exchanging signals with the outside.

Further, numeral 5 in the figure is an external interface to which information such as printed characters is input, and the input character code data is taken in by the processing of the CPU 1, and pattern data corresponding to this character code is read out from the pattern memory 7. The CPU 1 extracts an instruction from the program memory 2 and generates the address signal.

Here, as explained in FIG. 1, the pattern memory 7 stores one pattern in 512 bits in a checkered pattern, and is not stored in every other bit as shown in FIG. All patterns can be memorized.

Further, reference numeral 3 denotes a working memory for storing compressed pattern data read out from the pattern memory 7.

In this case, compressed pattern data is stored in the working memory 3 at every other address according to the processing of the CPUI, and in detail, it is stored in a checkered pattern as shown in FIG.

Additionally, this working memory 3 is a random access memory (1 bit corresponds to 10 bit address input).
RAM).

Next, the information currently stored in the working memory 3 is 512 bits of confirmed information in a checkerboard pattern, and the remaining 5
A playback pattern for one character is created by estimating the 12-bit prediction point and storing the result.

The file memory 4 creates information on this predicted point.

As shown in Fig. 2C, this file memory 4 outputs "white" or "black" information depending on the situation of the 12-bit sampling points (● marks) surrounding the prediction point marked ■. This is a read-only memory (ROM) that stores accumulated data for each pixel.

For example, if most of the surrounding 12 bits are black pixels, it can be determined that the center predicted point is also "black", so by inputting the 12 bits of data from the sampling point above, 1 bit of predicted data can be obtained. Output.

The prediction data output from this prediction file memory 4 is stored at the addressed prediction point of the working memory 3 mentioned above.

Furthermore, 6 is a printing section, and when all the predictions are completed, that is, when all the data are stored at the prediction points of the working memory 3, the printing section 6 prints the data according to the pattern information stored in the working memory 3 according to the instructions of the CPU 1. form a dot pattern.

In the printer configured as described above, the operation will be explained in order to understand the relationship in more detail.

First, the CPU 1 receives input through the external interface 3.

Print character code data is taken in, and a pattern (kanji) corresponding to the data is read out from the pattern memory 7.

Then, the read pattern data is transferred to the CPU 1
According to the address signal processed by the address signal, the bits are sequentially stored in the working memory 3 every other bit in the column or row direction.

Therefore, in FIG. 4, the addresses of sample points 1 to 12 to be referred to in order to estimate and determine the value of the predicted point marked ■ at the address that is now n are as shown in FIG.

The address shown in FIG. 5 is created by the CPU 1,
This address is given to the working memory 3, and a total of 12 bits of known data for each of addresses 1 to 12 are given to the file memory 4, to obtain the estimation result of the predicted point marked ■ at address n.

Obtained from this file memory 4: black "1" or white "0"
data is stored in the n address of the working memory 3.

This process is performed by the CPU 1 sequentially reading out the program memory 2.
It is executed according to the information.

Now, returning to the original explanation, the data output from the file memory to the predicted pixel at address n (white "0" or black "0"
1'' signal) is stored in the working memory 3, the process moves to the predicted pixel at the next n+2 address and repeats the same operation to sequentially store the data output from the file memory 4 to the predicted pixel. .

This operation is performed sequentially for each prediction point (n is 1, 3,
5...1023 or 2,4,6...1
024) If a total of 512 prediction operations are performed, 1024 bits of pattern data for one character will be stored in the working memory 3.
can be stored in

Then, the pattern data stored in the working memory 3 is sent to the printing section 6 to form a dot pattern.

When prediction is performed using the above method, the error rate for prediction for approximately 3,000 characters, that is, the rate at which predictions are made that are different from the actual pixels, is experimentally determined to be around 4 to 7%, which is in the practical range. It's in.

The important point of the present invention is that it is further developed based on the above-mentioned system.

In the above basic method, characters are reproduced using the file memory taken for all characters, so even though a certain character is "black" in a certain 12-pixel situation, the overall average file memory is used. There is a state in which it is determined that the color is "white".

In order to improve this, it is necessary to group together groups of characters that have common errors in the overall average file memory, and have separate file memories for them.

In this way, a file memory is provided for each common character group, and unknown pixels (predicted points) that are inferred and determined can be reproduced more accurately, reducing the error rate.

An example will be described below, in which a case where character groups are classified into four categories will be exemplified.

Further, as an example, it is assumed that 12 surrounding sampling points are selected in order to determine a prediction point as shown in FIG. 2C.

First, one method is to divide the character group according to the complexity of the characters.

The complexity of a character can be considered to be the number of strokes, that is, the number of compressed memory information "1" black pixels.

Therefore, it is easy to divide a character group into four parts depending on the number of black pixels.

The method of dividing this group of characters changes depending on the font, or the optimal division can be performed depending on the contents of the memory of each device.

At this time, alphabets, hiragana, numbers, katakana, and symbols have clearly different fonts than kanji in memory and should be classified as a separate category.

Therefore, in the four classifications, it is most appropriate to treat characters other than Kanji as one block, and classify each Kanji into three blocks according to the number of black pixels.

In this way, it is possible to identify whether the target character is a kanji character or not, based on the code signal of the target character, and to identify which block of kanji characters are classified based on the black pixels from the compressed data read from the pattern memory 4. Playback is possible by selecting a file and using this file memory.

Furthermore, by simultaneously adding data indicating which file memory to select for each Chinese character or other character to the pattern memory 4, each file memory can be selected without having to count black pixels.

FIG. 6 is a block diagram for implementing the present invention described above, and although not shown in the figure, it is controlled by the CPU 1 and program memory 2 described in FIG. 3, and processes such as data transmission are executed.

In the figure, 4 is the pattern memory explained in FIG. 3, and 8 is a separation circuit that separates the pattern data sent from the pattern memory into a signal for selecting the file memory group 12 and checkered compressed data.

That is, the number of black pixel "1" signals in the compressed data is counted by a counter or the like, the count contents are sent to the decoder 9, and the decoder 9 outputs a selection signal for the file memory 12.

This signal selects any file memory in the file memory group 12.

On the other hand, compressed data is temporarily stored in the stack memory 10.

The above file memory group consists of four file memories, and as explained earlier, in each file memory, items with the same or similar number of black pixels are classified and filed, such as numbers, alphabets, etc. The character group is stored in one file.

Further, 11 is a circuit for obtaining data of sampling points 1 to 12 from the stack memory 10 as shown in FIG. 4, and inputting this data as an address to the currently selected file memory 12.

Reference numeral 13 denotes a prediction point memory that sequentially stores the data output from the file memory 12. When the prediction data of all prediction points is stored in the memory 13, the data of the stack memory 10 storing this data and the compressed data is stored. are alternately output and stored in the pattern reproduction memory 14.

Next, to briefly explain the data flow in the block diagram shown in FIG. 6, first, as explained in FIG. 3, character data is taken in by the CPU 1 via the external interface.
A pattern corresponding to this data is read out from the pattern memory 4.

This read pattern data is sent to the separation circuit 8, where the number of black pixels is counted and the stack memory 1
Stored as 0.

Here, the count data of the black pixels is sent to the decoder 9, and one of the memories in the file memory group 12 is selected according to the count number.

That is, the file memory corresponding to the pattern data read out from the pattern memory 4 is selected.

On the other hand, the stack memory 10 sequentially sends data at 12 sampling points corresponding to FIG. 2C to the circuit 11, and inputs this 12-bit data as an address into the selected file memory.

Therefore, data on which predicted points are estimated is output from the file memory, and this data is sequentially stored in the memory 13.

And if the prediction data of 512 prediction points is obtained,
Data from the stack memory 10 and data from the memory 13 are alternately stored in the pattern reproduction memory, and the pattern data of the input character is reproduced.

Further, the pattern data stored in the pattern reproduction memory 14 is sent to the printing section through processing by the CPU 1, and a dot pattern is formed.

By having a file memory for each character group as described above, the error rate is reduced and a reproduced pattern close to the original pattern can be obtained.

FIG. 7 is a block diagram for implementing another separation of character groups.

This feature, that is, the classification of character groups, is based on 12 of the sampling points.
Classified by how pixels are taken.

Here, in a certain state, there are many black pixels, so in that case, the predicted point is determined to be black, but the number of white pixels cut off in that case becomes the number of errors when reproduced.

Therefore, even for the same character, there is a difference between the number of errors when the character is written as a in FIG. 2 and the number of errors when the character is written as C.

This difference in errors is a characteristic of the characters.

That is, depending on how the 12 pixels a to d shown in FIG. 2 are arranged, they are classified into character groups suitable for the arrangement.

Next, the means for dividing into character groups will be described below.

File memories for all characters are created in Figure 2 a, b, c, and d.

Here, in Figure 2, a should be connected to horizontal pixels, b should be connected vertically, C should be connected equally to the surroundings, and d should be connected diagonally to the right. I can do it.

Next, each file memory is used to calculate the number of errors for each character, and the ones with the fewest errors are collected and classified.

For each character group classified in this way, a dedicated file memory is created for the corresponding pattern in FIG. 2 a = d only for the characters belonging to that group.

This is the file memory for each character group.

Character groups can be classified in this way.

Which of the file memories created as described above is to be used can be selected by adding two bits as select bits to the pattern memory.

In other words, the pattern memory is 51 for each character.
By preparing a 212×4 bit memory as a file memory, a pattern similar to the original pattern can be reproduced.

An example of this will be explained with reference to the block diagram of FIG.

This figure 7 is almost the same as that of figure 6, and in the figure 8'
is a separation circuit that separates the signal for selecting the file memory group 12' from the pattern data sent from the pattern memory 4 and the checkered compressed data.

This separation circuit 8' transfers the pattern data of the pattern data read from the pattern memory 4 to the stack memory 1.
0, a select signal for selecting a file memory group is sent to the decoder 9'.

The select signal sent to the decoder 9' is decoded, selects one of the file memory groups 12', and is input to the 12 pixel selection circuit 11'.

That is, the signal from the decoder 9' is sent to the file memory group 12.
', select the file memory corresponding to the above pattern data, and use the file memory,
This is for selecting any situation among the surrounding 12 pixels a = d in Figure 2.

Therefore, the data from the stack memory 10 is inputted to the 12 pixel selection circuit 11', and in this block 11', the data shown in FIG.
For example, 12-bit data at the sampling point shown in a is input as an address to the memory being selected in the file memory group 12'.

As a result, predicted data of the predicted point is output.

This output data is stored in a memory 13 that sequentially stores prediction data.
is memorized.

Therefore, when the prediction data of all 512 prediction points are stored in the memory 13, this data and the stack memory 10
The data in the characters are sequentially stored alternately in the pattern reproduction memory 14, and one character is reproduced.

This reproduced data in the pattern reproduction memory 14 is C
According to the PUI processing, the image is sent to a printing unit or the like and a dot pattern is formed.

The pattern removed in a checkered pattern as described above is reproduced, and since a file memory suitable for the character is used depending on the character, the error rate is reduced and the reproduction pattern is almost the same.

Note that the classification of characters in the present invention is merely an example, and classification suitable for each font may be performed.

Further, although four examples of how to take 12 pixels have been described as shown in FIG. 2 a, b, c, and d, this method is not limited to this.

For example, even if this method is best for 32x32 Mincho typefaces, it probably does not apply to Gothic typefaces, and even if there is a Mincho typeface, there are many types of fonts, and each type has its own unique pattern. Because there is.

Furthermore, the number of file groups to be divided and the number of surrounding pixels for determining predicted pixels are 12 pixels and 4 groups in this example, but are not limited to this.

As mentioned above, since the present invention separates characters and provides a unique prediction file memory for each character group, the error at each prediction point is reduced and the pattern when reproduced is similar to the original pattern. be the same or nearly the same.

Moreover, the biggest advantage is that the pattern memory capacity is only half that of the conventional one, making it very economical.

Furthermore, although the present invention has been described with respect to a printer, the present invention is not limited to this and can be applied to character processing devices such as word processors and displays.

[Brief explanation of the drawing]

Figure 1 is a diagram provided to explain the present invention, Figure 2 a,
b, c, and d are diagrams showing several examples of how to take sampling points surrounding the predicted point according to the present invention, and Fig. 3 is a block diagram showing an example of the case where the basic method of the present invention is implemented in a kanji printer. , FIG. 4 is a diagram for explaining the surrounding situation of the predicted address at n, and FIG. 5 is a diagram showing the addresses at each sampling point in FIG. 4.
FIG. 6 is a block diagram for explaining the method according to the present invention, and FIG. 7 is a block diagram for explaining another example of the method according to the present invention. 4: pattern memory, 12, 12': file memory group.

Claims (1)

[Claims] 1. Expressing a character pattern with a dot matrix, storing known data obtained by sampling each character pattern in a checkered pattern, and reproducing the character pattern by interpolating predicted points that were not sampled from the known data. In the character processing method, there is a pattern memory that stores data sampled in a checkered pattern for various character patterns corresponding to various character data, and a pattern memory that stores data sampled in a checkered pattern for each character pattern corresponding to various character data, and a pattern memory that stores predicted points that are not sampled from the sampling data taken out from the pattern memory. a file memory having multiple types of reference files for determining data; and a file memory having a plurality of types of reference files for determining data; means for selecting a reference file associated with one of the files in accordance with the reproduced character pattern; and means for retrieving sampling data corresponding to given character data from the pattern memory and corresponding to the character data. However, the reference file in the file memory is selected by the selection means to determine the data of the prediction point that was not sampled, and the data of the determined prediction point is interpolated to the sampling data to reproduce the character pattern. A character processing method characterized by the following.