JPS6354276B2 - - Google Patents

Description

[Detailed Description of the Invention] The present invention relates to the processing of two-dimensional patterns that are generally called line figures, such as alphanumeric characters, texts containing kanji and kana, graphs, tables, and design drawings. This invention relates to a method for correcting and restoring a two-dimensional pattern mixed with reception noise to the original pattern.

When a two-dimensional pattern called a line figure is transmitted or copied, the quality of the pattern deteriorates. This is due to noise and distortion occurring during transmission or during the copying process. In practice, it is almost impossible to realize a transmission system or copying mechanism that does not contain components other than the original pattern. Therefore, there is a need to establish a technique for correcting and restoring a two-dimensional pattern with superimposed noise or distortion to its original pattern.

By the way, unlike natural images, line figures, which are information sources for texts and drawings, are usually generated using a certain method and according to certain production rules. That is, the quality deterioration of a two-dimensional pattern due to transmission or copying can be considered to be due to the deformation of the two-dimensional pattern into a pattern that does not conform to the above-mentioned generation rules due to noise or distortion. Therefore, it should be possible to restore the original pattern by correcting the components in the two-dimensional pattern that do not conform to the generation rules.

The present invention aims to provide a method for correcting and restoring a two-dimensional pattern based on such a viewpoint.

However, in the present invention, m×n pixels (for example, 3×3 pixels) in a two-dimensional pattern are used as a unit mesh, and a plurality of specific patterns among the patterns appearing in the unit mesh are selected in advance as legal patterns. I'll keep it. Then, observe the pattern of each unit mesh while sequentially moving the position of the unit mesh pixel by pixel on the two-dimensional pattern to be processed, and leave the unit mesh of the legal pattern as it is, and A legal pattern is assigned to the unit mesh (referred to as a legal pattern). When a legal pattern is assigned to a unit mesh of an illegal pattern, other unit meshes (the number of which are modified depending on the size of the unit mesh) that include the pixels in that unit mesh (referred to as the unit mesh of interest) are These patterns are different, but these are called reference unit meshes) and are checked to see if any illegal patterns occur. If no illegal pattern occurs, the modification of the unit of interest mesh is completed. If an illegal pattern occurs, reassign another legal pattern to the target unit mesh, observe the pattern of the reference unit mesh anew to check whether an illegal pattern occurs, and then repeat the reference unit mesh in the same way. The modification of the target unit mesh is repeated until no illegal patterns occur. The legal pattern to be assigned to the unit mesh of the illegal pattern is selected, for example, according to the Hamming distance from the illegal pattern. Of course, the selection rule is not limited to this.

The basic principle of correcting and restoring a two-dimensional pattern by converting an illegal pattern into a legal pattern was proposed by the applicant in a patent application filed in 1983.
A patent application has already been filed as No. 44464 (see Japanese Unexamined Patent Publication No. 57-159362), and the present invention is an extension of this already filed invention. To be more specific, the object of the present invention is to provide a two-dimensional pattern correction method that can easily be implemented in hardware and at high speed, and can perform highly accurate pattern correction.

An embodiment of the two-dimensional pattern correction method according to the present invention will be described below with reference to the drawings.

FIG. 1 is a functional block diagram of an embodiment of the two-dimensional pattern correction method of the present invention.

The image reading unit 101 raster-scans a document, drawing, or the like using a well-known method, samples it pixel by pixel, obtains a two-dimensional pattern to be processed, and stores it in the input buffer 102 . Unit mesh extraction unit 103
Regarding the two-dimensional pattern stored in the input buffer 102, while moving an observation window of the same size as the unit mesh pixel by pixel in the main scanning direction of raster scanning, the area within the observation window at each position (this is the unit mesh) is (equivalent to ) is extracted. That is, the unit mesh extraction section 103 shifts the unit mesh in the main scanning direction pixel by pixel and extracts the unit mesh pattern at each position.

FIG. 2 shows the relationship between the scanning direction of raster scanning and the unit mesh. A 3×3 pixel area surrounded by a thick line 201 in the figure is a unit mesh at a certain point in time. Hereinafter, it is assumed that the size of the unit mesh is 3×3 pixels. Regarding the two-dimensional pattern, such a unit mesh is sequentially shifted pixel by pixel in the main scanning direction. When the unit mesh advances to the final pixel position, shift the two-dimensional pattern by one pixel in the sub-scanning direction (that is, shift by one scanning line), and then move the unit mesh one pixel at a time in the main scanning direction from the first pixel position. to shift sequentially.

The conversion processing unit 104 includes the unit mesh extraction unit 10
3 compares the extracted unit mesh pattern at each position with the legal pattern previously stored in the legal pattern storage unit 105, and
Checks whether the pattern is a legal pattern or an illegal pattern. In Figure 3, the unit mesh size is 3.
This is an example of a legal pattern in the case of ×3 pixels, which is based on the aforementioned Japanese Patent Application No. 56-44464
57-159362)).

When the unit mesh pattern is determined to be a legal pattern as a result of comparison with the legal pattern in the legal pattern storage unit 105, the conversion processing unit 104 stores the unit mesh pattern as it is in the corresponding position of the working memory 106. Then, the process moves on to the unit mesh at the next position. The working memory 106 is used to temporarily hold legal patterns that have already been processed. On the other hand, if the unit mesh pattern is determined to be illegal, correction processing is performed in the following steps.

Step (i): One legal pattern in the legal pattern storage section 105 is allocated to the unit mesh of interest as a processing target (unit mesh of interest), and stored in the corresponding position in the working memory 106. Here, we consider a unit mesh pattern with a size of 3 x 3 pixels as a 9-bit Hamming code, select a legal pattern with the minimum Hamming distance from the pattern of the unit mesh of interest (which is an illegal pattern), and then It shall be assigned to the attention unit mesh. Furthermore, when there are two or more legal patterns with the same Hamming distance, the legal pattern that appears more frequently on the two-dimensional pattern of good image quality is selected and assigned to the mesh unit of interest.

Step (ii): Among the unit meshes in the working memory 106 for which correction processing has been completed in the vicinity of the unit mesh of interest, the pattern of the unit mesh (referred to as reference unit mesh) including the pixels in the unit mesh of interest is re-observed. (i.e., reread) and check whether illegal patterns have occurred. This is because, by replacing the original pattern of the unit mesh of interest with a legal pattern, there is a possibility that the pattern of the unit mesh that includes the changed pixels in the unit mesh of interest becomes an illegal pattern.

Now, the hatched unit mesh 2 in Figure 2
If 02 is the unit mesh of interest, a total of 12 unit meshes within the range surrounded by the dotted line 203 will be the reference unit mesh. However, unit meshes that include only unchanged pixels among the pixels in the unit mesh of interest may be excluded from the target of the check. In this way, when the number of changed pixels is one, the number of reference unit meshes can be reduced to eight.

Step (iii): As a result of the check, if no illegal pattern appears in any of the reference unit meshes, the correction process for the unit mesh of interest is completed and the process moves on to the process for the unit mesh at the next position. If an illegal pattern appears, proceed to the next step (iv).

Step (iv): Among the remaining legal patterns after excluding the legal patterns assigned to the mesh of the unit of interest up to this point, select the legal pattern that has the smallest Hamming distance from the pattern before modification of the mesh of the unit of interest (illegal pattern). , reassign it to the attention unit mesh, and store it in working memory 1.
06, and return to step (ii) again. If two or more legal patterns with the same Hamming distance remain, they are selected in descending order of frequency of appearance on the high-quality two-dimensional pattern, as in step (i). Thereafter, the assignment of composite patterns to the unit mesh of interest is repeated until no illegal patterns occur in the reference unit mesh.

When the legal patterns are collected in the working memory 106 as described above, the conversion processing unit 104 sends the legal pattern series to the output buffer 107. By applying such processing to the entire area of the two-dimensional pattern to be processed, it is possible to obtain a high-quality two-dimensional pattern consisting only of legal patterns from which noise and distortion have been effectively removed.

It goes without saying that the above processing may be executed in real time in synchronization with the reading and scanning of the two-dimensional pattern. In this case, it is clear that the working memory 106 required for processing has a capacity that can store the unit of interest mesh and its reference unit mesh. For example, in the case of a unit mesh of 3 x 3 pixel size, 7 pixels (main scanning direction)
The capacity is sufficient to store two-dimensional pattern information of ×5 pixels (in the sub-scanning direction).

In addition, although the legal pattern to be assigned to the unit mesh of the illegal pattern was selected according to the Hamming distance in the above example, it is also possible to select the legal pattern according to other rules. .

The two-dimensional pattern correction method according to the present invention has been described in detail above, and has the following advantages.

The two-dimensional pattern correction method of the present invention converts an illegal pattern into a legal pattern while shifting a one-dimensional mesh pixel by pixel, and also takes care to prevent illegal patterns from occurring in surrounding unit meshes due to this conversion. to select legal patterns. Therefore, since the validity of connections between pixels and between unit meshes is substantially maintained, sufficiently high precision pattern correction can be expected. In addition, the two-dimensional pattern correction method according to the present invention has a relatively simple processing algorithm as described above, and a small buffer memory is required for processing. It is easy to convert.

[Brief explanation of drawings]

Fig. 1 is a functional block diagram of an embodiment of the two-dimensional pattern correction method according to the present invention, Fig. 2 is a diagram showing the relationship between the raster scanning direction of the two-dimensional pattern and the unit mesh, and Fig. 3 is a specific example of a legal pattern. FIG. 101... Image reading unit, 102... Input buffer, 103... Unit mesh extraction unit, 104...
Conversion processing unit, 105... legal pattern storage unit, 1
06... Working memory, 107... Output buffer.

Claims (1)

[Claims] 1. m×n pixels in a two-dimensional pattern (however, m,
n is an integer of 2 or more) is taken as a unit mesh, a specific plurality of patterns that appear in the unit mesh are selected in advance as legal patterns, and the position of the unit mesh is set to 1 on the two-dimensional pattern to be processed. The pattern of each unit mesh is observed while sequentially moving pixel by pixel, and the unit mesh of legal patterns is left as is, and one legal pattern is assigned to the unit meshes of patterns other than legal patterns (referred to as illegal patterns). At the same time, when a legal pattern is assigned to a unit mesh of an illegal pattern (referred to as a unit mesh of interest), re-observe the patterns of other unit meshes that have been corrected and include pixels in the unit mesh of interest. A two-dimensional pattern correction method characterized by checking whether a legal pattern is generated and repeating the process while changing the legal pattern assigned to the mesh unit of interest as long as an illegal pattern is detected. 2. The two-dimensional pattern correction method according to claim 1, wherein the legal pattern to be assigned to the unit mesh of the illegal pattern is selected according to the Hamming distance from the illegal pattern.