JPS6349269B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention relates to a character transcription device, and more particularly to a character transcription device that converts character information written on an input form into visually superior information and outputs the converted information. Conventionally, there are the following devices for processing character information. The first is a typewriter device,
This is a method in which a person visually checks character information written in a manuscript and presses a key with his/her hand, thereby printing a font in a certain shape using corresponding typefaces.
The second is an optical character reader (OCR), which mechanically recognizes characters written in predetermined positions on a document and inputs the results into, for example, a computer. The third type is facsimile, which separates the input document into pixels, converts them into electrical signals, and transmits them to a remote location.At the receiving location, the electrical signals are reversely converted into visible information to obtain a faithful copy. The above devices are auxiliary devices used in offices and other places where a large amount of documents are handled, and they have the characteristics shown in the table below.

[Table] As is clear from the table above, typewriters print beautiful characters, but they have the drawback of always requiring human operation. Although OCR is originally intended as a recognition device, it is relatively easy to convert recognition results into visible information using an additional device. However, current recognition technology has the disadvantage that it is difficult to completely recognize handwritten characters including Chinese characters, and it is expensive. Also,
Although facsimile is superior in terms of faithful reproduction, it has the disadvantage that it does not include the quality of copying. Therefore, the main object of the present invention is to create a completely new type of character that has a character information input/output mechanism like a facsimile, but can automatically convert input character information into visually superior output character information. The purpose of the present invention is to provide a fair copying device. To summarize, this invention stores in advance a plurality of types of geometric information that should serve as standards for extracting the geometric features of characters, and the shape information extracted from the character information is based on any of the standards. Determine and select the most similar to the geometric information, and derive parameters for correcting the text information based on the selected reference geometric information and the shape information extracted from the text information. However, the character information is corrected based on the derived parameters. The above objects and other objects and features of the present invention will become more apparent from the following detailed description with reference to the drawings. Note that the character information used in this specification means not only kanji, alphanumeric characters, and numbers, but also simple symbols. FIG. 1 is a schematic block diagram showing one embodiment of the present invention. In the configuration, the photoelectric converter 1 is
The input document as shown in the figure is broken down into pixels and converted into electrical signals. Note that in FIG. 2, dotted lines indicate squares. The output of the photoelectric converter 1 is given to a memory 2 and stored therein. This memory 2 has a capacity capable of storing the entire screen of an input document as shown in FIG. Although not shown, the feature extractor 3 is composed of, for example, a microcomputer or a ROM in which geometric data is stored in advance, and includes shape information extraction means, similarity judgment means, geometric information storage means, and correction parameters. and deriving means. This feature extractor 3 reads character information stored in the memory 2 character by character and extracts its characteristics. For example, in FIG. 2, character information for each square is read out to the feature extractor 3, and its geometric features are extracted. Geometric feature data extracted by the feature extractor 3 is provided to a normalizer 4. Although not shown, the normalizer 4 includes, for example, a microcomputer, and reads the corresponding character information from the memory 2 based on the geometric feature data given from the feature extractor 3, and determines its position, font direction, and size. Corrects the sharpness, etc. The character information corrected by the normalizer 4 is given to the memory 5 and stored therein. The character information stored in this memory 5 is read out to a recorder 6. Although not shown, the recorder 6 includes, for example, a printer and converts the read character information into visible information. In operation, the photoelectric converter 1 supplies read character information, for example, as a black and white binary signal, to the memory 2 for storage. The shape information extracting means included in the feature extractor 3 reads character information stored in the memory 2 character by character and determines the convex hull of the read character. FIG. 3 is a diagram showing an example of this convex hull, and the convex hull of the character ``husband'' is shown with a dotted line. As is clear from FIG. 3, a convex hull is a line that surrounds a character in such a way that all points on a straight line connecting any two points forming the character are inside it. That is, the points forming the character are contained within all convex hulls. Next, the similarity judgment means included in the feature extractor 3 (comprised of a microcomputer, etc.)
determines which of a plurality of pre-stored geometric patterns the extracted convex hull is most similar to. These plural types of geometric patterns are stored in a geometric information storage means (for example, constituted by a ROM) included in the feature extractor 3. FIG. 4 is a diagram showing an example of geometric patterns stored in advance, and seven types of geometric patterns are shown here. FIG. 5 is a diagram showing the relationship between the convex hull in FIG. 3 and the geometric patterns determined to be similar by the feature extractor 3. That is, here it is determined that the triangles in FIG. 4 are similar. Note that in FIG. 5, circles indicate points where the convex hull and characters overlap. The similarity judgment operation of the feature extractor 3 is performed, for example, as follows. First, the feature extractor 3 extracts the overlapping area A between the convex hull and a pre-stored geometric pattern.
The ratio A/A+B is determined from the non-overlapping area B. This ratio is calculated for each geometric pattern, and it is determined that the convex hull is similar to the geometric pattern with the largest ratio. Furthermore, the correction parameter deriving means included in the feature extractor 3 calculates the degree of skewness at each position of the convex hull for the selected geometric pattern. The obtained skewness is then used in the normalizer 4 along with data such as the characteristics of the selected geometric pattern, such as the center of gravity, center, maximum side length, and inclination with respect to the squares of the input document.
is given as a parameter for correction. The normalizer 4 reads character information corresponding to various data provided from the feature extractor 3 from the memory 2. Then, the normalizer 4 corrects the read character information based on the various data provided. For example, the center of gravity or the center of the character information read from memory 2 is translated in parallel so that it coincides with the center of the square of the input manuscript (see Figure 2), and the length of the maximum side and the length of the square of the input manuscript are The length of the maximum side is expanded or reduced so that the ratio of the character information is kept constant, and if the read character information is tilted with respect to the squares of the input document, the tilt is corrected by rotating. Further, the normalizer 4 partially transforms the convex hull of the read character information based on the degree of skewness between the selected geometric pattern and the convex hull so that it approaches the selected geometric pattern. The character information corrected as described above is stored in the memory 5 and converted into visible information by the recorder 6. FIG. 6 is a diagram showing an example of a case where the characters shown in FIG. 3 have been corrected. As is clear from this figure, in the above embodiment, the convex hull of the character to be corrected is further approximated to the selected geometric pattern, and the size, position, inclination, etc. of the character are made regular. Therefore, input characters can be converted into more visually appealing characters and output. In addition, in the above-mentioned embodiment, the case where seven types of geometric patterns are stored in advance has been explained, but these may be further classified into more detailed classifications, or conversely, they may be roughly classified. Good too. Furthermore, in the above-described embodiment, a plurality of geometric patterns are stored as pre-stored geometric data in order to classify the external features of the characters. Any other data may be used as long as it is data from which features can be extracted.
For example, data serving as a reference for the inclination of each stroke of a character may be stored. In this case, the feature extractor 3 decomposes the character to be corrected into each stroke, and determines which of the pre-stored reference data of stroke inclinations the inclination of the stroke belongs to. And normalizer 4
The inclination of each stroke of the character information corresponding to the character information is corrected to match the reference data. In addition, data on the thickness of lines that form characters, curve function data that should be used as a reference for items that include many curved parts, such as numbers, etc., may also be stored. FIG. 7 is a schematic block diagram showing another embodiment of the invention. The feature of this circuit in FIG. 7 is that, unlike the circuit in FIG. 1, it has a transmission function. That is, the output of the photoelectric converter 1 is modulated by a modulator 7, transmitted to a remote location via a communication path 9, demodulated by a demodulator 8, and stored in the memory 2. For example, this modulation/demodulation may be performed in a manner similar to so-called facsimile. Note that this modulator 7
A demodulator 8 may also be inserted between the memory 5 and the recorder 6 as shown in FIG. Note that in FIGS. 7 and 8, parts that can be constructed in the same way as in FIG. 1 are given the same reference numerals. As described above, according to the present invention, shape information extracted from character information is selected by determining which of multiple types of reference geometric information stored in advance is most similar, and the selected shape information is selected. Parameters for correction are derived based on the reference geometric information extracted from the character information and shape information extracted from the character information, and the character information is corrected based on the derived parameters. Fine-grained correction can be performed according to the geometric characteristics of the characters, and handwritten characters can be converted into highly legible and beautiful characters. Furthermore, it is possible to obtain a character transcription device that is relatively inexpensive compared to conventional optical character reading devices.

[Brief explanation of drawings]

FIG. 1 is a schematic block diagram showing one embodiment of the present invention. FIG. 2 is a diagram showing an example of an input document. FIG. 3 is a diagram showing an example of the relationship between characters and convex hulls. FIG. 4 is a diagram showing an example of a geometric pattern stored in advance in the feature extractor 3. FIG. 5 is a diagram showing the relationship between the convex hull in FIG. 3 and the geometric pattern to which this convex hull is determined to be most similar. FIG. 6 is a diagram showing an example of a case where the characters in FIG. 3 have been corrected. FIG. 7 is a schematic block diagram showing another embodiment of the invention. FIG. 8 is a schematic block diagram showing still another embodiment of the present invention. In the figure, 1 is a photoelectric converter, 2 and 5 are memories, 3 is a feature extractor, 4 is a normalizer, 6 is a recorder, 7 is a modulator, and 8 is a demodulator.

Claims (1)

[Claims] 1. Character information reading means for reading character information written on an input manuscript and deriving it as an electrical signal; Geometric information storage means that stores geometric information in advance; Shape information extraction means for extracting information regarding the shape of character information read by the character information reading means; and Shape information extraction means. similarity determining means for determining and selecting which of the plurality of types of reference geometric information the shape information extracted by the means is most similar to, and means for deriving parameters for correcting the character information based on the shape information extracted by the method and reference geometric information selected by the similarity determining means; and based on the derived parameters. a correction means for correcting the character information so as to approximate the selected reference geometric information; and an output means for converting the output of the correction means into visual information and outputting the same. A character transcription device equipped with. 2. The shape information extraction means includes means for extracting a convex hull pattern of the character information, and the geometric information storage means stores a plurality of types of convex hull patterns to serve as a reference. The character transcription device according to paragraph 1. 3. The similarity determination means performs the similarity determination based on the ratio of the overlapping area and non-overlapping area of the convex hull pattern of the character information and the plurality of types of convex hull patterns to be the reference. , a character transcription device according to claim 2. 4. The character transcription device according to any one of claims 1 to 3, which is configured to be incorporated into a facsimile machine.