JPS6365149B2 - - Google Patents

Description

[Detailed description of the invention]

[Technical Field] The present invention relates to a technique for compressing a character pattern represented by a dot matrix. [Background Art] Dot-type printers or display devices are often used as terminal devices for computer systems, but in order to clearly express characters, especially kanji, dots of at least 16 x 16 dots, preferably 24 x
Must be displayed in 24 dots. 24×24 for one character
When expressed as a dot matrix, 8000 characters are 576
It will be a kilobyte and will occupy a large amount of memory capacity. personal computer, word
Small computers such as processors not only have small internal memory, but also use relatively small-capacity external memory such as floppy disks, so character patterns are compressed and encoded and stored in memory, and each character is It is proposed to restore the original dot matrix. If you divide the kanji pattern into several pixels,
The state of each pixel is not independent of the states of surrounding pixels, and there is an extremely strong correlation between the pixels on the upper, lower, left, and right sides. Several compression methods have already been proposed that take advantage of the properties of kanji patterns. Japanese Patent Laid-Open No. 55-39956 discloses, for example, that a 24 x 24 dot matrix is divided into 2 x 2 sub-matrices, and each sub-matrix is calculated based on the frequency of appearance based on its neighboring sub-matrix. A high compression rate can be obtained if Huffman encoding is performed using Huffman encoding, but in Huffman encoding, one set of codes consisting of 16 code words is obtained for one reference subpattern, so there are 16 ways in total. It has been pointed out that different codes are obtained, and the code given to each event is a code with no regularity in length, making it difficult to restore the character pattern and requiring a very complicated restoration circuit. In order to maintain regularity in the length of the code, the one in the same publication assigns variable-length code words with lengths equal to the ranks according to the 16 occurrence frequencies to the 16 sub-patterns that serve as conditions. However, the disadvantage is that the compression ratio is relatively low. [Summary of the Invention] An object of the present invention is to provide a character pattern compression method with a high compression rate and easy restoration. According to the present invention, a character pattern is compressed and encoded using a set of codes and restored by referring to a comparison table, so that the restoration is simple and quick and does not require a complicated restoration circuit. According to the present invention, a character pattern expressed by a dot matrix of M rows and N columns is divided into p dot subpatterns, and one or more left and right or top and bottom subpatterns (thus The frequency of occurrence of each sub-pattern is determined for a considerable number of characters, subject to the conditions of each of the determined patterns (hereinafter referred to as reference patterns), and the frequency of occurrence of the same rank across all reference patterns is aggregated and its frequency of occurrence is calculated. A variable-length code is assigned to each rank based on the distribution, and at the same time or before and after this, a comparison table is created between sub-patterns and the ranks that are conditioned on each of the reference patterns, and each sub-pattern of the character pattern is The character pattern is compressed by representing it with a variable length code assigned to the rankings related by the comparison table using the reference pattern as a condition. That is, the reference pattern is a(0), a(1), ...a(2 ^q -1), where q is the number of bits of the reference pattern.
Expressing ^this as The frequencies are totaled to obtain a histogram S(k) of the total frequency of appearance. At this time, since the order of appearance frequency of subpatterns differs for each reference pattern, a comparison table g=T(a,k) is created between subpattern g and appearance frequency order k, with reference pattern a as a condition.
Then, the 2 ^p ranks k are variable-length coded so that shorter code words are assigned to the more frequently appearing words.
For each subpattern of each character pattern to be compressed, a rank k of appearance frequency is found from the comparison table using the reference pattern as a condition, and the subpattern is replaced with the variable length code assigned to k. In this way, each character pattern is converted into compressed data. In this way, according to the present invention, a character pattern is compressed using only one set of variable length codes consisting of ^2p code words (p is the number of bits of a subpattern), so decompression is easy and does not require a complicated decompression circuit. do not. Furthermore, according to the present invention, compression is performed using the order of appearance frequency of subpatterns and the appearance frequency distribution.
High compression ratios are achieved. According to the Huffman encoding method, the steeper the frequency distribution of subpatterns, the higher the compression rate can be obtained. If Huffman encoding is applied to each standard pattern as in the past, there will be no regularity in the length of the code as a whole, and the compression rate will be low for standard patterns with a gentle appearance frequency distribution, which will limit the overall compression rate. However, according to the present invention, these problems are solved. [BEST MODE FOR CARRYING OUT THE INVENTION] A character pattern F(m, n) expressed by a dot pattern of M rows and N columns, m=0, 1, 2, . . .
When storing M, n=0, 1, 2, . . . N in a computer memory, eight pixels are usually treated as one byte. In order to shorten the processing time for compression and decompression, it is desirable to process several bits as one set, but since information is stored in bytes, it is better to select half of that, 4 bits, as one unit. In this case, the character pattern is a matrix G(m, j) with 4 bits as one subpattern, j=0, 1, 2,
...It can be considered as N/4-1. For example, the Kanji pattern shown in FIG. 1 consisting of 576 bits in 24 rows and 24 columns is treated as 144 sub-patterns.
Let G(m,j) be the sub-pattern of interest,
The subpattern above it is G(m-1,j),
Furthermore, the subpattern above it is G(m-2,j)
It is. G(m-2,
It has been found that a good compression ratio can be obtained by using 8 bits consisting of consecutive G(m-1, j).
There are 256 types of bit patterns that A(m,j) can take, and these are expressed as a(0), a(1), . . . a(255). A sub-pattern g that has each of the reference pattern a as a condition for a large number of characters to be processed.
A histogram H(a, g) is created by finding the frequency of appearance of (0), g(1), . . . g(15). Reference pattern a
If the appearance frequency of subpattern g(k) subject to condition (i) is expressed by αik, the histogram H(a, g) is expressed as shown in Table 1 below. a(256) is a reference pattern used when processing the first line of the character pattern, and is an 8-bit pattern consisting of all 1s.

[Table] There is no standard pattern that should be used as a condition when processing the first line of a character pattern. Suppose all 0
If a byte consisting of bits is treated as a reference pattern, the frequency of occurrence of the subpattern in the first row is actually added to the frequency of occurrence of the subpattern conditional on the reference pattern of all 0 bits, and the histogram H This increases the entropy of (0, g) and reduces the compression ratio. Therefore, the first row will be processed using 8 bits consisting of all 1's as a reference pattern. This is represented by histogram H (256, g) in Table 1. When processing the second line, it is assumed that there is a line consisting of all 1's above the first line. 257 sets of histograms obtained for reference patterns a(0) to a(256) are aggregated for each appearance frequency of the same rank to create one set of total histograms,
Based on this, a set of variable length codes is generated. about
Calculate the appearance frequency histogram H(a, g) of Table 1 for 4000 kanji and calculate the order of appearance frequency, k=0,
1,...15, a histogram h(a,k) as shown in Table 2 is obtained. Comparison table g between the position of g in the original histogram H(a, g) and the rank k in the size-ordered histogram h(a, k)
=T(a,k) is as shown in Table 3. In Table 3, the value of g is expressed as a decimal bit pattern.

【table】

【table】

[Table] After arranging each of the 257 histograms in order of size, a(0) to a(256)
One histogram S(k) is created by adding up the frequencies of occurrence of the same rank over the period. In S(k), it is preferable to convert the appearance frequency into a probability and use it. That is,
S(k) is obtained by converting each column of Table 2 into a summation probability. Based on the probability distribution in S(k), variable length codes are assigned to k=0, 1, . . . , 15, respectively. Variable-length coding methods include the Huffman coding method and its improvements or modifications, and when a variable-length code is assigned to S(k) using the Huffman coding method, the results shown in Table 4 are obtained.

【table】

[Table] After the comparison table g=T(a,k) and the variable length code of Table 4 are prepared in this way, each character pattern finds its reference pattern a with respect to the sub-pattern of interest, and the comparison table g=T Find k from (a, k) or its inversion table k=t(a, g), Table 4
The code assigned to k is found from the variable length code table of , and the subpattern is expressed by this code, thereby converting it into compressed data. To restore the original character pattern from compressed data, decode the variable length code to find k, find the reference pattern a from the top two subpatterns of the subpattern of interest, and create a comparison table g=T( a, k) to subpattern g
All you have to do is get . In the case of the character pattern shown in FIG. 1, all six subpatterns g in the first row are 0, so k=0 is obtained for each of the six subpatterns from a(256) in the comparison table. According to the code table of Table 4, the variable length code for k=0 is 0, so the 6 subpatterns in the first row are represented by 6 bits of "000000". Similarly, the subpatterns in the next row are sequentially variable-length coded. So, if we take a subpattern of 17 rows and 9 columns to 12 columns as an example of one subpattern, the reference pattern is 11001100, that is, in decimal notation a=
204, and the subpattern 1100 is g in decimal notation.
=12. Since k=0 is obtained from row a (204) of the comparison table and its variable length code is 0, it is represented by this subpattern 0. In this way, the character pattern is converted into compressed data. This method yields a compression ratio of about 50% for 8000 characters. Although the best embodiment has been described above, if the character pattern is compressed by the method of the present invention and then compressed in two stages by applying a well-known run-length compression method to this compressed data, an even higher compression rate can be obtained. is obtained. The run-length compression method is described, for example, in Japanese Patent Laid-Open No. 55-69842. FIG. 2 is a block diagram of a restoring device for restoring compressed character patterns according to the present invention. The compressed data is stored on an external storage device such as a floppy disk, and when used it is stored on a personal computer.
read into the internal memory of a computer or word processor. In addition, compressed data is
It may be stored in read-only memory within a computer or word processor. Memory 1 in FIG. 2 is an internal memory or read-only memory for storing such compressed data. Kanji characters are generally represented by 2-byte character codes. Compressed data of the character identified by the character code is read from the memory 1. The compressed data is sequentially sent bit by bit to the Huffman code decoder 3 and decoded, so that the value of k is obtained at the output of the decoder 3. Since k has a value from 0 to 15, it is represented by 4 bits. k is sent to the index register 7 of the look-up table 5 and is used to obtain the corresponding sub-pattern g from the look-up table 5 subject to the reference pattern a.
g is sent through a 4-bit output register 9 to a 12-stage parallel 4-bit shift register 11, and is also sent to a character pattern memory where it becomes part of the restored character pattern. The 6th stage of the shift register 11 includes a subpattern immediately above the subpattern of interest, and the 12th stage further includes a subpattern above it. Therefore, the values of the 6th stage and the 12th stage of the shift register 11 are used as the reference pattern a in the register 1.
3 and input to index register 7. In this way, a subpattern g is obtained from the reference table g=T(a,k) based on the values of k and a. When decoding the first part of compressed data, by setting all stages of the shift register 11 to 1, a reference pattern a consisting of all 1s is created.
(256) may be applied to the index register 7. As decoding progresses, subpattern g is sent to shift register 11, and reference pattern a is formed in register 13.

[Brief explanation of drawings]

FIG. 1 is a diagram showing an example of a character pattern to be compressed, and FIG. 2 is a block diagram of a character pattern restoring device.

Claims (1)

[Claims] 1. A method for compressing a character pattern represented by a dot matrix of M rows and N columns, which divides the character pattern into sub-patterns of several dots for a large number of target characters, For each sub-pattern, find the frequency of appearance of the sub-pattern subject to a reference pattern consisting of one or more sub-patterns adjacent to the sub-pattern, and calculate the frequency of appearance of the same rank across all reference patterns. A variable-length code is assigned to each rank based on the appearance frequency distribution, and a comparison table is created between the sub-patterns and the ranks, each of which is conditioned on each of the reference patterns, and each sub-pattern of the character pattern is determined with each of the reference patterns as a condition. A character pattern compression method characterized in that a character pattern is represented by a variable length code assigned to the rankings related by the comparison table. 2. The method of claim 1, wherein the variable length code is a Huffman code. 3. The method according to claim 2, wherein the sub-pattern is four dots that are continuous horizontally or vertically. 4. The method according to claim 3, wherein the reference pattern is composed of a subpattern directly above the subpattern of interest and a subpattern further above the subpattern.