JPH08297718A - Character segmentation device and character recognition device - Google Patents

Abstract

PURPOSE: To provide a character segmentation device which can correctly extract the separate characters such as 'i', etc., and also the pseudo separate characters generated by blurs by integrating the combinations of connected components having a specific relation among them into a single extracting object. CONSTITUTION: A connection component extraction means 13 extracts the black pixel connection components consisting of dots and lines which construct characters. A connection component integration means 15 integrates the combinations which have their tilts includes in a fixed range to the character strings put on a straight line connecting between the center of a prescribed graphic including a concatenation component and the center of a prescribed graphic including other connected components into a single extracting object among those combinations of black pixel connection components which are extracted by the means 13 and not deleted by a connection component deletion means 14 respectively. A character extraction means 16 extracts the image information of every character out of the black pixel connection components excluding those that are invalidated by both means 14 and 15.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device,
In particular, the present invention relates to character cutting out in the preprocessing and character recognition using the character cutting.

[0002]

2. Description of the Related Art Conventionally, as a technique for cutting out a character read by L / E conversion or the like, a technique mainly using a peripheral distribution and a technique using a method of extracting a black pixel connected component have been proposed. ing. Hereinafter, these two methods will be mainly described. A first character cutting device using the marginal distribution method includes a character string extracting means, a marginal distribution extracting means, and a character extracting means. Then, the character string extracting means extracts information about the position of the character from the image in which the character is described. Next, the marginal distribution extraction means scans the image in the direction orthogonal to the direction of the character string, with respect to the area of the image indicated by the information about the position of the character string extracted by the character string extraction means, The number of black pixels is counted for each scan, and this process is performed in turn in the direction of the character string, thereby extracting the distribution of black pixels existing around the area where the character string exists. Next, the character extracting means extracts the image information of each character in consideration of the estimated value of the average character width based on the marginal distribution extracted by the marginal distribution extracting means.

A second character extracting device using the method of extracting a black pixel connected component has a character string extracting means, a connected component extracting means, and a character extracting means. Then, the character string extracting means extracts information about the position of the character string from the image in which the characters are described. Next, the connected component extracting means labels the connected component (consisting of black pixels) of consecutive black pixels to the area of the image obtained from the information about the position of the character string extracted by the character string extracting means. Perform to extract each connected component. Specifically, all the pixels in the image area where the character string exists are scanned, and the black pixels connected to each other in four directions including up, down, left and right, or in eight directions including diagonal directions are extracted as connected components. Then, the same label is assigned to all pixels belonging to the same connected component. At this time, different labels are assigned to different connected components. (For example, described on page 75 of "Introduction to Computer Image Processing" (supervised by Soken Publishing, Hideyuki Tamura)). Next, the character extraction means, for each label created by the connected component extraction means,
Extract information about individual characters.

Next, in any of the above techniques, the data of the extracted individual character image is sent to the main body of the character recognition device for character recognition. Well-known techniques for character recognition include a template matching method and a decision tree method. Furthermore, in any of the methods, in order to improve the recognition accuracy, the method of using the correlation with the character of the previous word, and when there is a white space between words such as a Western text, this white space is used. There is a method of recognizing a word and further comparing it with a word dictionary. However, since these are not directly related to the gist of the present invention and are well-known and obvious techniques, description thereof will be omitted. This also applies to the later embodiments.

[0005]

However, in the case of using the marginal distribution, especially in the case of English, etc., when the letters are mixed vertically or horizontally and the character width is not constant, the characters are correctly Cannot be extracted. Figure 1
6 is shown. In the figure, there is an entry into the "T" and "a" of Take in the first sentence (a). Also, the second
“F” and “y” in sentence (b) are close to each other. Furthermore, the lowercase letters "i" and "l" have a narrower character width as compared to other lowercase letters "a", as well as uppercase letters.
A part of the letter is projected at the bottom of the line. In the third sentence (c), the English sentence in the German sentence is a cursive script,
Overlap has occurred in the vertical direction. Also, the fourth sentence (d)
Is a German letter, and separation occurs.

Further, the method of extracting the black pixel connected component was originally made in view of the drawback of the marginal distribution method, but again, for example, in the alphabet "i" or "j" or in the German Separated characters such as umlauts, or characters that are originally concatenated but are missing due to image fading cannot be extracted correctly. In addition, when a part of the characters included in another character string is protruding in the rectangular area of the extracted character string due to the inclination of the image or the narrow spacing between the character strings (rows), A part of the protruding character may be mistakenly extracted as a part of the character to be recognized.

[0007] Similarities also occur when there are ruby in Japanese documents, sending kana in Chinese sentences, and other upper and lower subscripts in both Japanese and Western texts. Next, if the character extraction cannot be performed correctly, the character recognition itself cannot be performed correctly. In particular, photographs and the like of old documents and valuable documents inherently have many distortions and blurring of characters and documents, and also blurring occurs when reading by L / E conversion.

Therefore, it is possible to correctly extract a separated character or a pseudo-separated character caused by a blur, and also when a part of the character in another string is projected into the character string to be recognized. It is desired to realize a character segmentation device that can be correctly extracted. An object of the present invention is to provide a useful character slicing device capable of satisfying such a demand as much as possible.

[0009]

In order to achieve the above-mentioned object, in the invention of claim 1, from the read data of the image in which the character (including the symbol) is described, the X on the image data surface of the character string is read. In a region indicated by a character string extraction unit that extracts information about a position including coordinate values, Y coordinate values, column heights, and the like, and information about the position of the character string extracted by the character string extraction unit, Originally, a connected component extracting means for extracting a connected component that is a character constituent component formed by connecting (consecutive) black pixels that represent points and lines that form a character, and each connected component extracted by the connected component extracting means. A straight line connecting the center of a predetermined graphic including one connected component and the center of a predetermined graphic including another connected component, which occurs in a separated character such as "i" in a combination of components. Above character string (vertical writing Then, a connected component integrating means for integrating a combination whose inclination with respect to a line) falls within a certain range as one character to be extracted, and a final operation under the action of the connected component extracting means and the connected component integrating means. The character recognition information extracting means for extracting information about the character to be recognized from the connected component extracted as the recognition target, for example, as it is, or for extracting a feature point when collating with the template is provided. There is.

According to the second aspect of the present invention, for example, a character string extracting means for extracting information about the position of the character string from the read data of the image in which characters such as the alphabet of the typeface are described, and the character string extracting means. In the area indicated by the information about the position of the character string extracted by
Connected component consisting of connected black pixels (including single black pixel)
Of the connected components extracted by the connected component extracting means and the connected component extracting means, and "..." indicating the kanji "kawa" or "two" or the German character umlaut.
Or the letter “M” in the alphabet, the shortest distance between a black pixel included in one connected component and a black pixel included in another connected component is, for example, in the character string (vertical writing, row) direction. Connected component integrating means for integrating a combination within a certain range (for example, less than or equal to a threshold value) with respect to a separately determined reference value as one extraction target, and the action of the connected component extracting means and the connected component integrating means The character recognition information extracting means for extracting the information about the character to be recognized, for example, the black pixel of only one character, from the finally extracted connected component is provided.

According to the third aspect of the present invention, the character string extracting means for extracting information about the position of the character string from the read data of the image in which the character is written, and the character string extracting means for the image data are extracted. In the area indicated by the position information of the character string, a kanji character "2" is selected from the combination of the connected component extracting means for extracting the connected component formed by connecting the black pixels and each connected component extracted by the connected component extracting means. "
Or the equal sign “=” of a mathematical symbol, etc., but the coordinate values in the column direction of the character of a predetermined figure containing one connected component and the character string of a predetermined figure containing another connected component (vertical If it is written, the connected component integration means that integrates a combination having an inclusion (overlap etc.) relationship with the coordinate values in the row direction as one extraction target, and the operation of the connected component extraction means and the connected component integration means Character recognition information extraction means for extracting information about the character to be recognized from the finally extracted connected component.

According to the invention of claim 4, a character string extracting means for extracting information on the position of the character string from the read data of the image in which the character is described, and the character string extracted by the character string extracting means. In the area indicated by the information about the position of the complex pixel, among the combinations of the connected component extracting means for extracting the connected component formed by connecting the black pixels and the respective connected components extracted by the connected component extracting means, a complicated kanji character or The letters "H", "A", "E", etc. can occur, but a predetermined figure containing one connected component, such as a horizontal line, and a predetermined figure containing another connected component, such as a vertical line. Line, the degree of overlap with a threshold value that is separately determined or more is integrated component for integrating as one extraction target, and the connected component extraction means and the connected component integration means It is characterized in that a character recognition information extracting means for extracting character information from the manner extracted connected components.

According to a fifth aspect of the present invention, a character string extracting means for extracting information about the position of the character string from the read data of the image in which the character is described, and the character string extracted by the character string extracting means. In the region indicated by the information about the position of the symbol, among the combinations of the connected component extracting means for extracting the connected components formed by connecting the black pixels and the respective connected components extracted by the connected component extracting means, the symbol “: , German umlaut, alphabetical letter “j”, etc., but the distance value between a predetermined figure including one connected component and a predetermined figure including another connected component is equal to or less than a threshold value, and The area ratio of both figures is within a certain range with respect to the separately determined reference value (for example, within the range of 0.8 to 1.2 for the symbol “;”, and below the threshold of 0.3 for “j”). One extraction Connected component integrating means for integrating as an elephant, and character recognition information extraction for extracting information about a character to be recognized from the connected components finally extracted under the action of the connected component extracting means and the connected component integrating means. And means are provided.

According to a sixth aspect of the present invention, a character string extracting means for extracting information about the position of the character string from the read data of the image in which the character is written, and the character string extracted by the character string extracting means. Connected component extracting means for extracting a connected component formed by connecting black pixels existing in an area having a predetermined number of pixels on both sides orthogonal to the character string direction than the area indicated by the information about the position, and the connected component extracting means. Among the connected components extracted by, the start or end coordinates in the direction orthogonal to the character string of the predetermined graphic including the connected component have the existence range of the position of the character string extracted by the character string extraction means, for example, , Which are given by the coordinate values in the upper and lower directions and the number of pixels in the direction orthogonal to the character string, are extracted as connected components that also form the characters of the character string. The connected component deleting means to be deleted from the object, and the information about the recognition target, for example, individual characters, is extracted from the connected component finally extracted under the action of the connected component extracting means and the connected component deleting means. And character recognition information extracting means for

According to a seventh aspect of the invention, a character string extracting means for extracting information about the position of the character string from the read data of the image in which the character is described, and the character string extracted by the character string extracting means. In the area indicated by the information about the position of the connected component, one connected component is selected from the connected component extracting means for extracting the connected component formed by connecting the black pixels, and each connected component extracted by the connected component extracting means. A connected component that integrates a combination of a straight line connecting the center of a predetermined graphic including the above and the center of a predetermined graphic including another connected component within a certain range with respect to the character string as one extraction target Character recognition information for extracting information about a character to be recognized from the connected component finally extracted under the action of the integrating unit, the connected component extracting unit, and the connected component integrating unit. A straight line connecting the center of a predetermined graphic including one connected component and the center of a predetermined graphic including another connected component among the combinations of the output means and the respective connected components extracted by the connected component extraction means. The character recognition auxiliary dictionary unit that has registered in advance information about a character or a combination of characters in which a combination in which the inclination with respect to the character string is within a certain range, and the connected component integrating unit,
When the judgment processing for integrating the combination having the inclination within a certain range as one extraction target is made, upon receiving this notification, at least one of the judgment results, the extraction information from the character recognition information extraction means. When character recognition is performed on the basis of the above, the character recognition section is characterized by having a connected component integrated information reflecting character recognition section that utilizes information registered in the character recognition auxiliary dictionary section.

According to an eighth aspect of the present invention, a character string extracting means for extracting information about the position of the character string from the read data of the image in which the character is described, and the character string extracted by the character string extracting means. In the area indicated by the information about the position, the connected component extracting means for extracting the connected component formed by connecting the black pixels, and one connected component among the combinations of the connected components extracted by the connected component extracting means Connected component integrating means for integrating a combination in which a shortest distance between a black pixel included in the black pixel and a black pixel included in another connected component is within a certain range with respect to a separately determined reference value as one extraction target, A character recognition information extraction means for extracting information about a character to be recognized from the connected component finally extracted under the action of the connected component extracting means and the connected component integrating means; Among the combinations of connected components extracted by the component extracting means, the shortest distance between the black pixel included in one connected component and the black pixel included in another connected component is constant with respect to a separately determined reference value. The auxiliary dictionary unit for character recognition in which information about characters or combinations of characters in which a combination within a range is registered in advance, and the connected component integrating means are within a certain range with respect to the shortest distance or the reference value. If the judgment processing for integrating a certain combination as one extraction target is made, the notification and at least one of the judgment results are received, and character recognition is performed based on the extracted information from the character recognition information extraction means. In this case, it is characterized in that it has a connected component integrated information reflecting character recognition unit that uses information registered in the auxiliary dictionary unit for character recognition.

According to the present invention, the character string extracting means for extracting information about the position of the character string from the read data of the image in which the character is written, and the character string extracting means for the image data are extracted. In the area indicated by the position information of the character string, one connected component among the connected component extracting means for extracting the connected component formed by connecting the black pixels, and each connected component extracted by the connected component extracting means. A combination having an inclusive relation between the coordinate values in the column direction of a character of a predetermined figure that includes and the coordinate values in the column direction of a character of a predetermined figure that includes other connected components is integrated as one extraction target. Character recognition information extraction for extracting information about a character to be recognized from the finally extracted connected component under the action of the connected component integrating means, the connected component extracting means and the connected component integrating means And a predetermined figure including the other connected components and the column direction coordinate value of the character of the predetermined figure including one connected component among the combinations of the connected components extracted by the connected component extracting means. The auxiliary dictionary unit for character recognition in which information about a character or a combination of characters in which a combination having an inclusive relation with the column-direction coordinate value of the character occurs is registered in advance, and the connected component integrating means includes the above inclusive relation. If a determination process for integrating a certain combination as one extraction target is made, the notification and at least one of the determination results are received, and character recognition is performed based on the extraction information from the character recognition information extraction means. When performing, it is characterized in that it has a connected component integrated information reflecting character recognition unit that uses information registered in the auxiliary dictionary unit for character recognition.

According to the tenth aspect of the invention, a character string extracting means for extracting information about the position of the character string from the read data of the image in which the character is described, and the character string extracted by the character string extracting means. In the area indicated by the information about the position of the connected component, one connected component is selected from the connected component extracting means for extracting the connected component formed by connecting the black pixels, and each connected component extracted by the connected component extracting means. Connected component integrating means for integrating a combination in which the degree of overlap between a predetermined figure including the above and a predetermined figure including another connected component is equal to or more than a separately determined threshold value as one extraction target, and the above connected component extracting means And a character recognition information extracting means for extracting character information from the finally extracted connected component under the action of the connected component integrating means, and the connected component extracting means. Among the combinations of the respective connected components, the degree of overlap between a predetermined figure including one connected component and a predetermined figure including another connected component is equal to or greater than a separately determined threshold value If the character recognition auxiliary dictionary unit that has registered information about combinations in advance, and the connected component integrating means performs a determination process of integrating combinations in which the degree of overlap is a threshold value or more as one extraction target, Upon receiving this notification and at least one of the judgment results, when the character recognition is performed based on the extracted information from the character recognition information extracting means, the information registered in the auxiliary dictionary unit for character recognition is used. It is characterized by having a connected component integrated information reflecting character recognition unit.

According to an eleventh aspect of the present invention, a character string extracting means for extracting information about the position of the character string from the read data of the image in which the character is described, and the character string extracted by the character string extracting means. In the area indicated by the information about the position, the connected component extracting means for extracting the connected component formed by connecting the black pixels, and one connected component among the combinations of the connected components extracted by the connected component extracting means A connected component in which a distance value between a predetermined figure including the above and a predetermined figure including another connected component is less than or equal to a threshold value, and an area ratio of the two figures is within a certain range with respect to a separately determined reference value. Connected component integrating means for integrating the combination of as one extraction target, and the connected component finally extracted under the action of the connected component extracting means and the connected component integrating means to a character to be recognized. Of the character recognition information extracting means for extracting the information, and a predetermined graphic including one connected component and a predetermined graphic including another connected component among the combinations of the connected components extracted by the connected component extracting means. The distance value to the figure is less than or equal to the threshold value, and the area ratio of both figures is within a certain range with respect to the separately defined reference value. The character recognition auxiliary dictionary unit and the connected component integration means perform a determination process to integrate a combination in which the distance value is equal to or less than a threshold value and the area ratio is within a certain range with respect to a reference value as one extraction target. In response to this notification and at least one of the judgment results, when performing character recognition based on the extracted information from the character recognition information extracting means, the character recognition auxiliary dictionary unit is registered. It is characterized by having a coupling component integrated information reflecting a character recognition unit that utilizes the information being.

According to the twelfth aspect of the present invention, a character string extracting means for extracting information about the position of the character string from the read data of the image in which the character is described, and the character string extracted by the character string extracting means. Connected component extracting means for extracting a connected component formed by connecting black pixels existing in regions having a predetermined number of pixels on both sides orthogonal to the character string direction than the region indicated by the information about the position, and the connected component extracting device. Among the connected components extracted by, the start or end coordinates in the direction orthogonal to the character string of the predetermined figure including the connected component are the existence range of the position of the character string extracted by the character string extracting means. Connected component deleting means for deleting a deviated connected component from a target for extracting as a connected component which also constitutes a character of the character string, the connected component extracting means and the connected component The character recognition information extracting means for extracting information about the character to be recognized from the finally extracted connected component under the action of the removing means, and among the connected components extracted by the connected component extracting means, A character recognition auxiliary dictionary unit in which information about a character or a combination of characters that is likely to have a connected component for which the connected component deleting means exhibits its effect is registered.
When the connected component deletion means deletes the connected component from one extraction target, upon receiving this notification, the deleted connected component is added to the end based on the extraction information from the character recognition information extraction means. It is characterized in that it has a connected component deletion information reflecting character recognition unit that utilizes information registered in the auxiliary dictionary unit for character recognition when recognizing characters in other columns included as a part. In the thirteenth and fourteenth aspects of the present invention, the connected component integrating means is a quadrangle that circumscribes the connected component and whose sides are in the character string direction or in a direction orthogonal to this as a predetermined figure including the connected component. Is characterized by having a circumscribing quadrangle creation unit for creating

[0021]

With the above structure, in the invention of claim 1,
Character string extraction means, from the read data of the image in which the characters are described, after grasping the overall shape of the image data of the paper in which the characters are described, etc., information about the position of the character string,
For example, the left and right edges, the upper and lower edges, etc. are extracted by using the coordinate value indicated by the number of pixels, the coordinate value indicated by the ratio to the entire image data, and the like. The connected component extracting means extracts only a continuous portion of "1" from the data consisting of "0" and "1" in the area indicated by the information about the position of the character string extracted by the character string extracting means. Then, a connected component which is a character component formed by connecting black pixels is extracted. The connected component integrating means sets the center of a predetermined graphic including one connected component and the center of a predetermined graphic including another connected component among the combinations of the connected components extracted by the connected component extracting means. Combinations in which the inclination of the connecting straight line with respect to the character string is within a certain range, such as 70 ° to 110 °, are integrated as one extraction target. The character recognition information extraction means, under the action of the connected component extraction means and the connected component integration means, finally extracts information about the character of the recognition target from the connected component extracted as one recognition target, Further, it is sent to the character recognition unit.

According to the second aspect of the present invention, the character string extracting means extracts the information about the position of the character string from the read data of the image in which the characters are described. The connected component extracting means extracts a connected component formed by connecting black pixels in the area indicated by the information about the position of the character string extracted by the character string extracting means. Connected component integration means
Among the combinations of the connected components extracted by the connected component extracting means, the shortest distance between the black pixel included in one connected component and the black pixel included in another connected component with respect to a separately determined reference value. Within a certain range, for example, a combination that is a threshold value of 0.3 or less with respect to the height of the character string is integrated as one extraction target. The character recognition information extraction means extracts information about the character to be recognized from the finally extracted connected component under the action of the connected component extracting means and the connected component integrating means. In the third aspect of the invention, the character string extracting means extracts information about the position of the character string from the read data of the image in which the characters are written. The connected component extracting means extracts a connected component formed by connecting black pixels in an area of the image data indicated by the position information of the character string extracted by the character string extracting means. The connected component integrating means includes, among the combinations of connected components extracted by the connected component extracting means, predetermined coordinate values including the coordinate values in the column direction of characters of a predetermined graphic including one connected component and other connected components. A combination having an inclusive relation with the coordinate values of the characters of the figure in the column direction is integrated as one extraction target. The character recognition information extraction means extracts information about the character to be recognized from the finally extracted connected component under the action of the connected component extracting means and the connected component integrating means.

According to the fourth aspect of the invention, the character string extracting means extracts information on the position of the character string from the read data of the image in which the characters are described. The connected component extracting means extracts a connected component formed by connecting black pixels in the area indicated by the information about the position of the character string extracted by the character string extracting means. Connected component integration means
Among the combinations of the connected components extracted by the connected component extracting means, the degree of overlap between a predetermined figure including one connected component and a predetermined figure including another connected component is not less than a separately determined threshold value. A certain combination is integrated as one extraction target. The character recognition information extracting means extracts character information from the finally extracted connected component under the action of the connected component extracting means and the connected component integrating means.

According to the fifth aspect of the invention, the character string extracting means extracts information about the position of the character string from the read data of the image in which the characters are described. The connected component extracting means extracts a connected component formed by connecting black pixels in the area indicated by the information about the position of the character string extracted by the character string extracting means. Connected component integration means
Among the combinations of connected components extracted by the connected component extracting means, the distance value between a predetermined figure including one connected component and a predetermined figure including another connected component is equal to or less than a threshold value. A combination of connected components, in which the area ratio of the graphic is within a certain range with respect to a separately determined reference value, for example, a threshold value or more, is integrated as one extraction target. The character recognition information extraction means extracts information about the character to be recognized from the finally extracted connected component under the action of the connected component extracting means and the connected component integrating means.

In the invention of claim 6, the character string extracting means extracts information about the position of the character string from the read data of the image in which the characters are described. The connected component extracting means has a predetermined value determined by the situation of the character string data of the original image on both sides orthogonal to the character string direction with respect to the area indicated by the information about the position of the character string extracted by the character string extracting means. The number of pixels, for example, one pixel, a connected component formed by connecting black pixels existing in a large area is extracted. The connected component deleting means extracts, from the connected components extracted by the connected component extracting means, start or end coordinates in a direction orthogonal to a character string of a predetermined graphic including the connected components by the character string extracting means. The connected component formed of a line or the like that deviates from the existing range of the position of the character string is deleted from the target to be extracted as the connected component that also constitutes the character of the character string. The character recognition information extracting means extracts information about the character to be recognized from the finally extracted connected component under the action of the connected component extracting means and the connected component deleting means.

According to the invention of claim 7, in addition to the operation of the invention of claim 1, the following operation is performed. In the auxiliary dictionary unit for character recognition, among the combinations of the connected components extracted by the connected component extracting means, the center of a predetermined graphic including one connected component and the center of a predetermined graphic including another connected component Information about a character or a combination of characters in which a combination of a straight line connecting with and the inclination of the character with respect to the character string is within a certain range, for example, "i", "j", "=".
"÷", German umlaut, and other reverse information, such as "∴", uppercase "P", "T", and lowercase "c", "a", "e", etc. I have registered. If the connected component integrated information reflecting character recognition unit performs the determination process of integrating the combination in which the connected component integrating means has the inclination within a certain range as one extraction target,
Upon receiving this notification, at least one of the judgment results, and performing character recognition based on the extracted information from the character recognition information extracting means, the information registered in the character recognition auxiliary dictionary unit is set to the normal information. It is used as a reference when it is unclear in the recognition process.

In the eighth aspect of the invention, the following action is performed in addition to the action of the second aspect of the invention. In the character recognition auxiliary dictionary unit, the shortest distance between the black pixel included in one connected component and the black pixel included in another connected component among the combinations of the connected components extracted by the connected component extracting unit. Information about a character or a combination of characters in which a combination is within a certain range with respect to a separately determined reference value, for example, "i", "j", "...",
“...”, “;”, “:”, “=”, “H”, “D”, etc. in case of special typeface and separation are registered in advance. If the connected component integrated information reflecting character recognizing unit performs the determination process of integrating the combination in which the connected component integrating means has the shortest distance within a certain range with respect to the reference value as one extraction target, Upon receiving the notification, at least one of the judgment results, and performing character recognition based on the extracted information from the character recognition information extracting means, the information registered in the character recognition auxiliary dictionary unit is used as a candidate character Used for weighting, etc.

In the invention of claim 9, in addition to the operation of the invention of claim 3, the following operation is performed. In the character recognition auxiliary dictionary unit, among the combinations of the connected components extracted by the connected component extracting means, the coordinate values in the column direction of the characters of a predetermined figure including one connected component and other connected components are included. Information about a character or a combination of characters in which there is an inclusive relationship with the coordinate value of the character of the predetermined figure in the column direction, for example, "=", "?"
"!", "...", "...", two "・" of umlaut,
Other uppercase letters such as "P" as reverse information (plurality of characters that can be vertically related) and a lowercase letter "c" adjacent to them
Etc., lowercase letters "f" and "y" are registered in advance. If the connected component integrated information reflecting character recognition unit performs the determination process of integrating the combination having the inclusion relation as one extraction target, this notification, at least one of the determination results is given. When performing character recognition based on the extracted information from the character recognition information extracting means, it is possible to directly use the information registered in the auxiliary dictionary unit for character recognition or narrow down the candidates. To use.

In the invention of claim 10, claim 11 is provided.
In addition to the function of the invention described above, the following function is performed. In the character recognition auxiliary dictionary unit, among the combinations of the connected components extracted by the connected component extracting means, the degree of overlap between a predetermined figure including one connected component and a predetermined figure including another connected component. Information about a character or a combination of characters in which a combination is equal to or greater than a separately determined threshold value, for example, “X”, “H”, “E”, other postal symbols, etc. "Y" etc. are registered in advance. If the connected component integrated information reflecting character recognition unit performs the determination process of integrating the combination in which the overlapping degree is the threshold value or more as one extraction target, this notification, or the result of the determination Upon receiving at least one and performing character recognition based on the extracted information from the character recognition information extraction means, the information registered in the auxiliary dictionary unit for character recognition is used.

According to the eleventh aspect of the invention, the following action is performed in addition to the action of the fifth aspect of the invention. A distance value between a predetermined figure including one connected component and a predetermined figure including another connected component among the combinations of the connected components extracted by the connected component extracting means in the auxiliary dictionary unit for character recognition. Is less than or equal to a threshold value, and the area ratio of the two figures is within a certain range with respect to a separately determined reference value.
For example, if the area ratio is 0.7 or more, then "i", "j",
"?", ":", Etc. if it is less than this. In addition to this, as reverse information, uppercase letters "P", "T" and lowercase letters "c", etc. are multiple characters, but this relationship may occur depending on the condition. , Etc. are registered in advance.

The connected component integrated information reflecting character recognition unit judges that the connected component integrating means integrates a combination in which the distance value is equal to or less than a threshold value and the area ratio is within a certain range with respect to a reference value as one extraction target. After the processing, upon receiving this notification and at least one of the judgment results, when performing character recognition based on the extracted information from the character recognition information extraction means, it is registered in the character recognition auxiliary dictionary section. Use the information provided.

In the twelfth aspect of the invention, the following action is performed in addition to the action of the sixth aspect of the invention. In the character recognition auxiliary dictionary unit, among the connected components extracted by the connected component extraction unit, a character or a character that is likely to have a connected component that is the target of the action of the connected component deleting unit. Information about combinations, such as horizontal writing “n”, “g”, “q”, “f”, and vertical writing cursive, such as “water”, is registered in advance. When the connected component deleting means deletes the connected component from the one extraction target, the connected component deletion information reflecting character recognition unit receives this notification and displays the information registered in the character recognition auxiliary dictionary unit. , Which is used when recognizing a character in another column including the deleted connected component as a part at the end based on the extracted information from the character recognition information extracting means.

In the thirteenth and fourteenth aspects of the present invention, in the circumscribed quadrangle creation unit of the connected component deleting means, a predetermined figure including the connected component is circumscribed to the connected component and each side is a character string. In the direction or in the direction orthogonal thereto, for example, a rectangle is created if the character to be recognized is an alphabet, and a square is created if it is a Chinese character.

[0034]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A character cutting device according to the present invention will be described below based on embodiments. (First Embodiment) FIG. 1 is a block diagram of an embodiment of a character slicing device according to the present invention. In the figure, 10 is an input means for inputting a document converted into image data. 1
Reference numeral 1 is a character string extracting means for extracting position information of a character string from image data in which characters are described. Reference numeral 12 is a connecting section extracting means for checking the pixel value on the designated scanning line of the image data and outputting the start and end positions of the section in which black pixels are one-dimensionally connected. 13 is an extraction on a scanning line continuous in the character string direction in an area larger by one pixel on both sides orthogonal to the direction of the character string than the area indicated by the position information of the character string extracted by the character string extraction means 11. Based on the connectivity of the connected sections of the black pixels, the connected component extracting means extracts the black pixel connected components that are originally composed of points and lines that form a character. The reference numeral 14 indicates a character extracted from the connected component of the black pixel by the connected component extracting means 13 in which the start or end coordinates in the direction orthogonal to the character string direction of the rectangle circumscribing the connected component are extracted by the character string extracting means. It is a connected component deleting unit that deletes a black pixel connected component that deviates from the range of column position information from the target of character extraction. 15
Is a combination of black pixel connected components that are extracted by the connected component extraction unit 13 and not deleted by the connected component deletion unit 14 and are combined as one extraction target. It is a means for integrating connected components. Reference numeral 16 denotes a connected component deleting unit 14 of the connected components extracted by the connected component extracting unit 13.
And character extraction means for extracting the character image information for each character from the black pixel connected components excluding the connected components invalidated by the connected component integrating means 15. Reference numeral 200 denotes a character recognition device main body that performs character recognition based on the image information for each character extracted by the character extraction means 16, and incorporates a recognition dictionary (template, template) 20 and the like. There is.

In addition to the above, a normalization processing unit for compensating for differences in printed characters and typefaces, character size, etc., an average character size calculation unit for obtaining the average size of each character from the height of a character string, a drawing, A non-character pixel exclusion unit that eliminates ruled lines, etc., a graphic document that extracts some words or part of lines independently from other lines or other parts of the same line, if necessary, Although it has a control unit or the like for technical papers, these are not directly related to the gist of the present invention, and are well-known and obvious techniques, so description thereof will be omitted.

Next, the operation of the character slicing device configured as described above will be described with reference to FIGS. 2 to 14. FIG. 2 shows a basic procedure of the character string extraction means 11 for extracting position information of a character string from image data in which characters are described. This procedure is based on the fact that if there is at least one black pixel in a pixel row arranged in the x- and y-axis directions or in the horizontal and vertical directions, a character string (row) exists in the pixel row. , To find the coordinates of a pixel column (row) in which a character string exists. Since this is a known technique, detailed description of each step is omitted. In practice, this procedure alone is not sufficient, so the following processing is performed prior to or in addition to this.

Black pixels are normally projected in the vertical direction and the horizontal direction of the image data-formed paper to find the character writing position, the writing direction, and roughly the columns and rows. If necessary, the length and width of the paper may be half or 1/5, for example.
The above operation is performed by taking each of them, and the character string or the character string including the headword is found more accurately. The marginal distribution in the column direction of the character is taken to more accurately determine the position of the character column in the column direction.

For each position of the character string, the marginal distribution is taken in the direction orthogonal to the character string direction to determine the circumscribed rectangular coordinates of the character string. If necessary, the image is compressed in the character string direction by means such as, for example, if even one of the eight pixels in the character string direction has a black pixel, all eight pixels are regarded as black pixels (the same as in FIG. 2). The connected components are extracted, and the circumscribed rectangle of each connected component is further extracted.

In addition, the circumscribed rectangles are integrated as recognition targets and extraction targets according to a rule such that rectangles that are close to each other in the direction in which the character string is described are the same character or character string that is the same recognition target. At this time, for example, if the characters to be extracted are alphabets of a typeface, all uppercase letters have the same height, and even lowercase letters "h" and "l" have the same height as the uppercase letters. There is an upper limit for the ratio of, and the principle is used. However, these processes are not directly related to the gist of the present invention, but rather are premised so-called publicly known techniques, and therefore further description will be omitted.

An example of the result of the operation of the character string extracting means 11 is shown in FIG. This figure shows in (b) the position information of the character string extracted from the image data shown in (a). Here, the thick black line portion of the y coordinate axis at the left end of (a) indicates the portion (y coordinate value) where the character string is present in the procedure shown in FIG. Further, in (b), in order to indicate the position information of the character string in the image data, the x coordinate of the upper left point of the rectangle circumscribing the character string is s (start) x, the y coordinate is sy, and the lower right point. X-coordinate of e (end) x, y-coordinate of e
y. The rectangle that circumscribes the character string is used as the position information because it matches the shape of the printed alphabet, and the process described later is simple.
This is because the basic program can be used for italics. Further, in this embodiment, as shown in FIG.
The coordinates of the upper left corner of the image data are the origin (0, 0), the character string (right side) direction is the x-axis direction, and the vertical downward direction is the y-axis direction when orthogonal to the character string direction.

The connected section extracting means 12 checks the pixel value on the designated scanning line in the image data, and outputs the start and end positions of the section where the black pixels are continuous in the horizontal (x-axis) direction. The procedure will be described with reference to the flowchart of FIG. In the following description, the first pixel is counted as the 0th pixel in the order of pixels on the scanning line of the image data.

The connecting section extracting means 12 (step a01) sets the number of pixels on the designated scanning line to the variable N as the final pixel number of the column to be investigated. (Step a02) (0) is set to the variable i as the pixel number to be investigated, and the initial value (0) is also set to the variable j as the number of the connected section.

(Step a03) It is checked whether or not the i-th pixel on the designated scanning line of the image data is a black pixel. If it is a black pixel, the process proceeds to step a04, and if it is not a black pixel, the process proceeds to step a10. (Step a04) Set the start position of the connection section (j) to the variable i
Set to the value of. (Step a05) (1) is added to the value of the variable i.

(Step a06) It is checked whether or not the value of the variable i is equal to the value of the variable N. If they are equal, the process has proceeded to step a12 because all pixels have been checked, and if they are not equal, the process proceeds to step a07. (Step a07) It is checked whether or not the i-th pixel on the designated scanning line of the image data is a white pixel. If it is a white pixel, the process proceeds to step a08, and if it is not a white pixel, the process proceeds to step a0.
Return to 5.

(Step a08) The end position of the connection section (j) is set to the value of the variable i. (Step a09) (1) is added to the value of the variable j. (Step a10) Add (1) to the value of the variable i. (Step a11) It is checked whether or not the value of the variable i is equal to the value of the variable N. If they are not equal, the process returns to step a03, and if they are equal, the process ends.

(Step a06) If the value of the variable i is equal to the value of the variable N, the process proceeds to step a12. (Step a12) Set the end position of the connected section (j) to the variable i.
Set to the value of. (Step a13) (1) is added to the value of the variable j, and the process ends. Through the above processing, the connected section extracting unit 12 checks all the pixels on the designated scanning line of the image data, and the total number j of connected sections of black pixels and the connected section (0) to the connected section (j-1). The start position and end position up to are output.

An example of the operation result of the connected section extracting means 12 is shown in FIG. (A) of the figure shows the connected section extracting means 1
2 shows an example of pixel-by-pixel data on a designated scanning line in the image data given in FIG. Then, the black portion becomes a stroke or the like that constitutes a character. In this figure, the position of the pixel at the left end is set to 0, and the coordinate value increases in the right direction.

(B) of the figure is information on the connection section extracted from the pixel data shown in (a), and the connection section number, and thus the total number, and the start coordinates of the connection section of each number, in order from the left column. A value, similarly the coordinate value of the end. The connected component extracting means 13 has black pixels in an area larger than the area indicated by the position information of the character string in the image data extracted by the character string extracting means 11 by one pixel on each side orthogonal to the direction. Of the black pixel component and the circumscribed rectangle are output by distinguishing each of them. The procedure will be described with reference to the flowcharts of FIGS. Although these three figures are originally one leaf, they are divided into three because of size limitation.

First, before explaining the procedure itself, the symbols used in the flow chart will be described. For the connection section extracted by the connection section extraction means 12, the number of connection sections on the scanning line immediately before the scanning line to be acted on (upper, y value is small) is pn, and is set as the current object of operation. The number of connected sections on the scanning line is represented by cn. Further, each connected section itself is represented by p (0) to p (pn-1) and similarly c (0) to c (cn-1). Next, the number of the connected component that includes the i-th connected section p (i) is represented in the form of Rp (i), and the number of the connected component that also includes the i-th connected section c (i) is Rc (i ) Format.

Further, as information for identifying the connected component R,
The x coordinate sx, y coordinate sy of the upper left point of the circumscribed rectangle for indicating the position information of the connected component R, and the x coordinate ex, y of the lower right point
It is assumed that the start coordinates sx and the end coordinates ex of the black pixel for each scanning line for indicating the spread of the black pixel of the coordinate ey and the connected component R are used. Therefore, in the following description, when updating the information on the connected components, the position information s of the circumscribed rectangle is
x, sy, ex, ey and the start coordinates sx and end coordinates ex of the black pixels for each scanning line are updated.

Based on the above, the procedure itself will be described below. (Step b01) The character string extraction means 11 extracts the variable y indicating the coordinate of the scanning line of interest in the image data in the y-axis direction within a range not exceeding the start coordinate of the image data in the y-axis direction. A coordinate value that is one less than the value of the start y coordinate of the character string rectangular coordinate is set. It should be noted that this is to check the connection section existing in the scanning line on the upper side by one pixel. Further, the value of the variable k indicating the number of the connected component being extracted (and thus the character number) is initialized to (0), and at the same time, the connection in the scan line immediately preceding the scan line of interest is connected. The value of the variable pn indicating the number of sections is initialized to (0).

(Step b02) Connected section extracting means 12
Is used to extract the total number of connected sections cn of the scanning line y of the image data and the start and end coordinates of each connected section c (0) to c (cn-1). (Step b03) The value of the variable i that represents the connected section on which the action is to be performed on the scanning line to be investigated is initialized to (0).

(Step b04) The value of the variable i is the scanning line y.
It is determined whether or not the number of connected sections is equal to cn. If they are equal, the process proceeds to step b21 to process the next scanning line, and if not, the process proceeds to step b05. Hereafter, step b05
From step b20 to the connection section c (i) on the scanning line of interest, the connection sections p (0) to p (p on the scanning line immediately preceding the scanning line of interest are processed.
The connectivity with n-1), that is, whether to extract as one recognition target is checked. If there is a connection, in step b09 to step b14, the area of the connection section c (i) is added to the existing connected component, or the existing connected components are integrated. A process for creating a new connected component will be performed.

(Step b05) The value of the variable Rc (i) indicating the number of the connected component including the connected section c (i) is changed to (-).
Initialize to 1). (Step b06) Of the connected sections on the previous scanning line, the value of the variable j representing the number of the object whose connectivity is to be checked is initialized to 0. (Step b07) It is judged whether or not the value of the variable j is equal to the number of connected sections pn of the immediately preceding scanning line, and if they are equal, the process proceeds to step b16, and if they are not equal, the process proceeds to step b08.

(Step b08) Whether or not the connecting section c (i) on the scanning line of interest and the connecting section p (j) of the scanning line immediately preceding the scanning line of interest are connected. To judge. That is, the relationship between the start and end coordinate values of the connection section c (i) and the start and end coordinate values of the connection section p (j) is checked to see if there is a section common (overlapping) in the x coordinate direction. . Then, if they are connected, step b0
9. If not connected, the process proceeds from step b09 to step b15 without performing the processing of step b14.

(Step b09) It is judged whether or not the value of the connected component Rc (i) including the connected section c (j) is (-1), that is, it is invalid. If it is invalid, the existing connected component Rp is found.
To add the area of the connected section c (i) to (j), proceed to step b10, and if not invalid, the existing connected component Rp
Go to step b11 to integrate (j) with the existing connected component Rc (i).

(Step b10) The region of the connected section c (i) is added to the connected component Rp (j) including the connected section p (j) to form the connected component Rp (j), and the process proceeds to Step b14. (Step b11) Connected component Rc (i) including connected section c (i) and connected component R including connected section p (j)
It is determined whether or not p (j) is equal. If they are equal, there is no need to integrate again, so the process proceeds to step b14, and if they are not equal, the process proceeds to step b12.

(Step b12) The connected component Rp (j) containing the connected section p (j) is added with the region of the connected component Rc (i) containing the connected section c (i) to obtain the connected component Rp.
(J). (Step b13) The connected component Rc (i) is invalidated. (Step b14) The same value as Rp (j) is set to the variable Rc (i) indicating the connected component including the connected section c (i).

(Step b15) In order to check the connectivity with the next connecting section on the scanning line immediately preceding the scanning line of interest, (1) is added to the value of the variable j and the process returns to step b07. (Step b07) If the value of the variable j is equal to the variable pn, that is, if the connectivity with all connected sections on the scan line immediately preceding the scan line of interest has been checked, the process proceeds to step b16. To reach.

(Step b16) The connected section c (i) is
Whether it is not included in any connected component yet,
That is, the connection section p (0) to p (p
n-1) whether or not there is a connection with any of the connection section c
It is determined whether or not the variable Rc (i) indicating the connected component including (i) is (-1) indicating that it is invalid. If it is invalid, the following steps b17 to b19 are performed to create a new connected component, and if it is not invalid, the process proceeds to step b20.

(Step b17) A connected component k is newly set from the information on the connected section c (i). That is, the start coordinate value and the end coordinate value of the connected section c (i) are input to sx and ex of the connected component k, respectively. Also, sy and e
The coordinate value y of the scanning line of interest is substituted for y. Moreover, the coordinate values of the start and end of the connection section c (i) are input as the coordinate values of the start and end of the black pixel on the scanning line y.

(Step b18) The value k is set to the variable Rc (i) indicating the connected component including the connected section c (i). (Step b19) Add (1) to the value of the variable k. (Step b20) In order to process the next connected section on the scanning line of interest, (1) is added to the value of the variable i and the process returns to step b04. And step b04
Then, when the value of the variable i is equal to the variable cn, that is, when the processing for all the connection sections of the scanning line of interest is completed, the process reaches step b21. Next step b2
From 1 to step b24, the information regarding the connection section of the scanning line which is currently focused is regarded as the connection section of the scanning line immediately before the next scanning line in order to process the next scanning line. .

(Step b21) The value of the connected section number cn of the scanning lines, which has been focused so far, is input as the connected section number pn of the immediately preceding scanning line. (Step b22) From the position information c (0) to c (cn-1) of each connected section of the scanning line, which has been focused so far, from the position information p (0) of each connected section of the preceding scanning line. p (cn
-1) Enter as.

(Step b23) A variable Rc indicating a connected component including each connected section of the scanning line which has been focused so far.
From (0) to Rc (cn-1), variables Rp (0) to Rp indicating the connected component including the connected section of the previous scanning line are included.
Input as (cn-1). (Step b24) In order to advance the scanning line of interest by one, 1 is added to the value of the variable y.

(Step b25) It is judged whether or not the value of the variable y is equal to the value of the end y coordinate of the character string rectangular coordinates extracted by the character string extracting means 11 (2). If they are not equal, the process returns to step b02 and the process for the next scanning line is performed. If they are equal to each other, it means that the examination of the connecting section in the scanning line below by one pixel has been completed, so the process is completed. In addition, the reason why "+2" is set here is
This is consistent with the input of "-1" in the previous step b01.

FIG. 9 shows an operation example of the connected component extracting means 13. As shown in (a) of this figure, a rectangle L2 that is one pixel extra in both the vertical direction and the y-coordinate of the circumscribing rectangle L1 of the character string extracted by the character string extracting means 11 (internal characters are omitted). The black pixel connected component is extracted for the image range. However, the start and end y-coordinates of the rectangle L2 are
Of course, the range of the original image data should not be exceeded.

The three connected components R1 shown in FIG.
The extraction results for R2 and R3 are shown in (c) of this figure.
The connected component deletion means 14 is a character whose start or end coordinates in the direction orthogonal to the character string of the circumscribed rectangle are extracted from the combination of the black pixel connected components extracted by the connected component extraction means 13 by the character string extraction means. Black pixel connected components that deviate from the rectangular coordinate range of the column are deleted. That is, FIG.
In (a) of (3), elements having elements in the upper and lower L2 sections are excluded from the abstraction target of the character of the character string.

Next, the procedure of the connected component deleting means 14 will be described with reference to the flowchart of FIG. (Step c01) The total number of connected components extracted by the connected component extracting means 13 is set in the variable N. (Step c02) An initial value (0) is set to the variable i indicating the number of connected components existing in the row to be investigated.

(Step c03) Connected component extracting means 13
Whether the starting coordinate ys (i) of the circumscribing rectangle of the connected component (i) extracted in step y is smaller than the starting coordinate ysL of the circumscribing rectangle of the character string extracted by the character string extracting unit 11 in the y direction. To judge. If the condition is satisfied, the process proceeds to step c05, and if not, the process proceeds to step c04. (Step c04) End coordinate ye in the y direction of the circumscribed rectangle of the connected component (i) extracted by the connected component extracting means 13.
It is determined whether (i) is larger than the end coordinate yeL in the y direction of the circumscribed rectangle of the character string extracted by the character string extraction means 11. If the condition is satisfied, the process proceeds to step c05, and if not, the process proceeds to step c06.

(Step c05) When it is determined in step c03 or step c04 that the connected component (i) deviates from the circumscribed rectangle of the character string, the connected component (i) is invalidated in this step. . (Step c06) Add (1) to the variable i. (Step c07) Compare the value of the variable i with the value of the variable N,
If the values do not match, the process returns to step c03, and if the values match, the process ends.

FIG. 11 shows an operation example of the connected component deleting means 14. For image data as shown in (a) of this figure,
The result of extracting the image of the character described in the second line,
Also shown in (b). In (b), the connected component extraction means 13 extracts connected components from a region indicated by a rectangle L2 that is one pixel wider in the direction perpendicular to the direction of the character string than the circumscribed rectangle L1 of the character string, and the rectangle is extracted from the rectangle. The connected component R deviating from the region of L1 is judged by the connected component deleting means 14 to be a part of the character "p" included in the character string of another line, which is included in the upper line in this figure. Then, it is deleted from the extraction target of this line.

The connected component integrating means 15 selects a combination of connected components satisfying a predetermined condition among the connected component combinations extracted by the connected component extracting means 13 and not deleted by the connected component deleting means 14. Integrate as one extraction target. Next, the processing of the connected component integration means 15 will be described using the flowchart of FIG. 12 and an example of integration conditions shown in FIG.

(Step d01) The number of connected components extracted by the connected component extracting means 13 is substituted for the variable N, and the value (0) is set for the variable i indicating the number of the connected component to be acted on.
Set. (Step d02) It is checked whether or not the connected component (i) is valid. If it is valid, the procedure proceeds to step d03, and if it is invalid, the procedure proceeds to step d09.

(Step d03) The variable j is set to the value of (i + 1). (Step d04) Check whether the connected component (j) is valid,
If valid, proceed to step d05, and if invalid, step d0.
Proceed to 7. (Step d05) For the connected component (i) and the connected component (j), it is determined whether or not any one of the integration conditions (1) to (5) shown in (a) of FIG. 13 is satisfied, If 1 is satisfied, the process proceeds to step d06, and if not, the process proceeds to step d07. The determination procedure at this time is shown in FIG.

(Step d06) The connected component (i) and the connected component (j) are integrated and stored as a new connected component (i), and accordingly, the connected component (j) is excluded from a single extraction target. Invalidate. (Step d07) 1 is added to the value of the variable j. (Step d08) The value of the variable j is compared with the value of the variable N,
If the values do not match, the process returns to step d04, and if the values do match, the process proceeds to step d09.

(Step d09) 1 is added to the value of the variable i. (Step d10) The value of the variable i is compared with the value of (N-1). If the values do not match, the process returns to step d02, and if the values match, the process ends. FIG. 14 shows an operation example of the connected component integrating means 15. Conditions (1) to (5) shown in FIG.
14 (a) to (e) of FIG. 14, respectively.

In FIG. 14A, the absolute value of the inclination of the straight line connecting the center of gravity of the circumscribed rectangle R1 of the connected component C1 and the center of gravity of the circumscribed rectangle R2 of the connected component C2 is 2 or more, so that in FIG. The condition of 1) is satisfied. In FIG. 14B, the shortest distance between the black pixels of the connected component C1 and the connected component C2, that is, the distance at the point B is 1 pixel or less, so the condition of (2) in FIG. 13 is satisfied.

In (c) of FIG. 14, there is an inclusive relation between the coordinates of the circumscribed rectangle R1 of the connected component C1 in the x-axis direction and the coordinates of the circumscribed rectangle R2 of the connected component C2 in the x-axis direction.
The condition (3) of is satisfied. In (d) of FIG. 14, the overlap between the x-axis direction of the circumscribed rectangle R1 of the connected component C1 and the x-axis direction of the circumscribed rectangle R2 of the connected component C2 is shorter in the x-axis direction, that is, the circumscribed rectangle R2. 1 of the x-axis length
Since it is / 2 or more, and the same applies to the y-axis direction, the condition (4) of FIG. 13 is satisfied.

In (e) of FIG. 14, the distance between the circumscribed rectangle R1 of the connected component C1 and the circumscribed rectangle R2 of the connected component C2 is 2 pixels or less, and the area ratio is 5 or more.
The condition (5) of is satisfied. As described above, in (a) to (e) of FIG.
The information on 1 and the information on the connected component C2 are integrated to create a new connected component R3.

The character extracting means 16 respectively extracts, from the connected components extracted by the connected component extracting means 13, the connected components excluding the connected components invalidated by the connected component deleting means 14 and the connected component integrating means 15. Is converted into the final information of the image of the one character to be recognized and is output. That is, the x coordinate sx, y coordinate sy of the upper left point for indicating the position information of the connected component R, the x coordinate ex of the lower right point,
The y coordinate ey and the start coordinate sx and end coordinate ex of the black pixel for each scanning line for indicating the spread of the black pixel of the connected component R
Therefore, it was possible to accurately extract and create the image data including the black pixels of the connected component R in the original image data. (Second Embodiment) FIG. 15A is a block diagram of an embodiment of the character recognition apparatus according to the present invention. In (a) of this figure, 100 is a character slicing device. Reference numeral 200 denotes a main body of the character recognition device, which includes an auxiliary dictionary part 22 for character recognition and a connected component information reflecting character recognition control part 23 in addition to the recognition dictionary 20 and the recognition processing part 21 required as an ordinary character recognition device. Have

The character cutting device 100 is the same as that shown in the previous embodiment. However, if the connected component integrating means and the connected component deleting means have fulfilled their functions, the effect and how the function has been exerted is sent to the connected component information reflecting character recognition control section 22. Means 1
The difference is that it has 7. Auxiliary dictionary unit 23 for character recognition
Is registered in association with each character that is likely to cause a difficulty in extracting a connected component as a recognition target for a character to be recognized and a cause of erroneous extraction. Specifically, the information is created by reflecting the following facts.

Upper part (horizontal writing), right part (vertical writing) of Kanji
The small and simple characters with a small number of black pixels described in Section 1 above are highly likely to be syllabary characters. If it is a Chinese sentence, it is a burial point (symbol). In the alphabet, it is the combination of uppercase letters "P", "T", etc. and the lowercase letters "a", "c", "e" etc. in the vicinity that are likely to cause vertical overlap. It also occurs in foreign languages such as German, which are cited in part of English sentences, italics, and typeface similar to cursive. In this case, in particular, lowercase letters “f”, “l” and the like are likely to cause vertical overlap.

The separating characters are lowercase letters "i", "j" and the like in the case of alphabets, and other German umlauts and the like. Further, as the separated characters that are likely to occur due to the difficulty of reading (erroneous reading), there are characters in which black pixels are connected in a narrow portion of the character line, such as German characters corresponding to M and N. In Chinese characters, the center line of the vertical separating portion is not largely biased with respect to the vertical line. For example,
"Chang", "Call", etc. In addition, "bamboo crowns" and "grass crowns" have a substantially constant vertical ratio that occupies the entire kanji.

Characters with upper and lower subscripts are often found in technical articles and academic papers. In this case, the upper subscript is an Arabic numeral such as "2", "p", "x", "y", "M",
It is often found in lowercase letters of certain alphabets such as "n". Subscripts are often attached to certain alphabets such as "P" and Greek letters. Also, small upper subscripts and lower subscripts themselves are often found in Arabic numerals such as "3" and in specific alphabets such as "i" and "m".

Characters that partly protrude above the normal character string are "f", "h", etc. in lowercase alphabets regardless of italic type or typeface, and conversely as characters protruding downward. , Lowercase letters "j", "n", "q", etc. As for the combination of characters that come into contact with each other, if it is an alphabet, lowercase letters "w", "y", etc., such as those having a line (stroke) extending to the upper left or right end of the character, and those and above "t" etc. It tends to occur between characters that have horizontally extending and protruding lines. In capital letters, "A" that projects downward and "H", "R", "L", etc. where the left or right is vertical. In particular, as shown in FIG. 16 (c), there are many fonts in which a small horizontal projection is provided at the lower end of the character. Also,
By handwriting, "13" may sometimes resemble "B". Similarly, in handwritten vertical writing, “Ma” and “shi” are used.

In addition to the above, various readings can be made depending on the field of science or technology such as mathematics, the type of language such as Chinese or Japanese, the writing style or the typeface, or the typeface if it is the typeface. There are characters and symbols that are difficult to extract and combinations thereof, but the characters and combinations of characters that cause the difficulty are roughly determined. Specifically, in any language, the point added to the word to alert the reader is just above the letters that make up the word in horizontal writing, right in vertical writing, and This point is often a black circle with a certain size.

Kanji separated horizontally are "Kawa" and "Aki".
For example, the number of Chinese characters that are vertically separated is small, such as "two" and "three." The number of separations is usually up to three. Compared to Kanji, Kana are often written smaller. Dashes often found in technical documents are attached to the right shoulder of letters. For this reason, the character recognition auxiliary dictionary unit 23 registers the cause and the character causing the cause in advance according to the language and typeface of the document to be recognized. A part thereof is shown in FIG.

The connected component information reflecting character recognition control unit 22
When the connected component integrating means and the connected component deleting means have fulfilled their functions and have received information on how they have worked, the character recognition auxiliary dictionary unit 23 is referred to obtain information about the corresponding characters. It is sent to the recognition processing unit 21. The recognition processing unit 21 reflects this information when recognizing the extracted character. However, it is needless to say that how to reflect this information when selecting or determining a candidate character by template matching or the like, the weighting value, etc. are appropriately selected according to the document to be recognized. is there.

Specifically, when the original image data is dirty and the paper is smeared here and there, the weight of the notification from the connected component deleting means is lowered, and a word consisting of a combination of characters read before and after is read. Raises the weight of whether or not exists in the recognition word dictionary (not shown). If the character spacing and the row and column spacing are sufficiently vacant as in the specification of the patent application, the character recognition auxiliary dictionary is hardly referenced.

The present invention has been described above based on the embodiments, but the present invention is not limited to the above embodiments. That is, for example, the following may be performed. (1) In this embodiment, the method of extracting the connected component by extracting the connected section included in the scanning lines of the image data and examining the connectivity between the connected sections included in the continuous scanning lines has been described. However, for example, a method of extracting a connected component by tracing a contour line is used.

(2) In the embodiment, the predetermined figure including the connected component is a circumscribing rectangle corresponding to the printed alphabet. If it is Japanese or Chinese, it is square. Further, the typeface, language, etc. are detected during the operation of extracting the character string in the first line, and these are automatically performed.

(3) For the convenience of manufacturing, a plurality of essential items (means) 1 of the present invention may be provided in a plural number, or conversely, a plurality of items may be integrated, or these may be appropriately combined. (4) In the case of alphabets, the width and height of one character, whether upper case or lower case, are within a certain range. For this reason, when adjacent characters become one connected component for the convenience of reading, etc., a means for dividing this into two parts from the upper limit of the character width is adopted.

(5) A weight is used for extracting and recognizing handwritten characters. (6) In cursive script, the characters in the upper and lower lines may largely protrude. Therefore, in order to deal with this, the out-of-column region investigated by the connected component deleting means is not one pixel in the direction orthogonal to the character string. , Multiple pixels.

[0094]

As described above, the invention of claim 1 can correctly extract the separated characters such as alphabetic characters i and j. According to the second aspect of the present invention, it is possible to correctly extract a character or the like separated from a part of a faint image that should be originally connected and is missing.

According to the third aspect of the present invention, it is possible to correctly extract a character or the like separated from a part of a faint image that should have been originally connected and is missing. According to the fourth aspect of the present invention, it is possible to correctly extract a character or the like separated from a part of a faint image that should be originally connected and is missing. According to the fifth aspect of the invention, it is possible to correctly extract a character or the like separated from a part of a faint image that should be originally concatenated.

According to the sixth aspect of the invention, a part of the characters included in another character string is projected in the rectangular area of the extracted character string due to the inclination of the image or the narrow space between the character strings. If some of the characters on other lines are not extracted by mistake. Further, the inventions of the above claims are integrated,
This contributes to improving the accuracy of extracting character information.

The inventions of claims 7 to 12 can effectively reflect the extraction results of the character images in the inventions of claims 1 to 6 in character recognition. Therefore, the character recognition rate is improved especially when the original image is handwritten or dirty. According to the thirteenth and fourteenth aspects of the invention, when the inventions of the first to the twelfth aspects are carried out, it is easy to create a program and process data. Further, it is possible to deal with the extraction and recognition processing of italic fonts etc. without making a large change to the program created as the main object of the typeface.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a first embodiment of a character clipping device according to the present invention.

FIG. 2 is a basic operation flow chart of character string extraction in the above embodiment.

FIG. 3 is a diagram showing an example of an operation result of a character string extracting means in the above embodiment.

FIG. 4 is a diagram showing a procedure of an operation of the connected section extracting means in the above embodiment.

FIG. 5 is a diagram showing an example of an operation result of the connected section extracting means in the above embodiment.

FIG. 6 is ⅓ of the diagram showing the procedure of the operation of the connected component extracting means in the embodiment.

FIG. 7 is 2/3 of the diagram showing the procedure of the operation of the connected component extracting means in the embodiment.

FIG. 8 is 3/3 of the diagram showing the procedure of the operation of the connected component extracting means in the embodiment.

FIG. 9 is a diagram showing an example of the operation content of the connected component extraction means in the above embodiment.

FIG. 10 is a diagram showing a procedure of an operation of the connected component deleting means in the above embodiment.

FIG. 11 is a diagram showing an example of the content of the operation of the connected component deleting means in the above embodiment.

FIG. 12 is a diagram showing a procedure of an operation of the connected component integrating means in the above embodiment.

FIG. 13A is a diagram showing an example of integration conditions of the connected component integration means in the above-described embodiment. (B) is a diagram showing a procedure for determining whether or not the condition is satisfied.

FIG. 14 is a diagram showing an operation example of a connected component integrating means in the above embodiment.

FIG. 15 is a configuration diagram of an embodiment of a character recognition device according to the present invention.

FIG. 16 is an example of a Western sentence that is difficult to extract by the conventional technique.

[Explanation of symbols]

 10 Input Means 11 Character String Extracting Means 12 Concatenated Section Extracting Means 13 Concatenated Component Extracting Means 14 Concatenated Component Deleting Means 15 Concatenated Component Integrating Means 16 Character Extracting Means 17 Concatenated Deletion Integrated Information Sending Means 20 Recognition Dictionaries 21 Recognition Processing Units 22 Recognition Components 22 Information reflected character recognition control unit 23 Character recognition auxiliary dictionary unit

 ─────────────────────────────────────────────────── --- Continuation of the front page (72) Inventor Ho Ho Takakura 1006 Kadoma, Kadoma City, Osaka Prefecture Matsushita Electric Industrial Co., Ltd.

Claims (14)

[Claims] 1. A character string extracting means for extracting information about a position of a character string from read data of an image including a character, and information about a position of a character string extracted by the character string extracting means. In a region, a connected component extracting means for extracting a connected component formed by connecting black pixels, and a center of a predetermined figure including one connected component among combinations of connected components extracted by the connected component extracting means Connected component integrating means for integrating a combination in which a straight line connecting the center of a predetermined graphic including other connected components with respect to the character string within a certain range as one extraction target, and the connected component extraction Means and character recognition information extracting means for extracting information about a character to be recognized from the connected component finally extracted under the action of the connected component integrating means. Character cutting device. 2. A character string extracting means for extracting information about a position of a character string from read data of an image including a character, and information about a position of a character string extracted by the character string extracting means. In the region, a connected component extracting means for extracting a connected component formed by connecting black pixels, and a combination of each connected component extracted by the connected component extracting means, a black pixel included in one connected component and another Connected component integration means for integrating a combination whose shortest distance from a black pixel included in a connected component is within a certain range with respect to a separately determined reference value as one extraction target, the connected component extraction means and the connected A character slicing device comprising: character recognition information extraction means for extracting information about a character to be recognized from a connected component finally extracted under the action of the component integration means. Place. 3. Character string extraction means for extracting information about the position of a character string from read data of an image containing characters, and position information of the character string extracted by the character string extraction means of image data. In a region, a connected component extracting means for extracting a connected component formed by connecting black pixels, and a predetermined graphic character including one connected component among combinations of connected components extracted by the connected component extracting means Connected component integrating means for integrating a combination having an inclusive relation between the coordinate values in the column direction of the column and the coordinate values in the column direction of a character of a predetermined graphic including other connected components as one extraction target; Character recognition information extracting means for extracting information about a character to be recognized from the finally extracted connected component under the action of the component extracting means and the connected component integrating means. Character cutting device. 4. A character string extracting means for extracting information about a position of a character string from read data of an image including a character, and information about a position of a character string extracted by the character string extracting means. In the region, a connected component extracting means for extracting a connected component formed by connecting black pixels, and a predetermined figure including one connected component among other combinations of connected components extracted by the connected component extracting means, Connected component integrating means for integrating as a single extraction target a combination in which the degree of overlap with a predetermined figure including the connected component is greater than or equal to a separately determined threshold, and the operation of the connected component extracting means and the connected component integrating means And a character recognition information extraction means for extracting character information from the connected component finally extracted under the following conditions. 5. A character string extraction unit for extracting information about the position of a character string from the read data of an image containing characters, and information about the position of the character string extracted by the character string extraction unit. In the region, a connected component extracting means for extracting a connected component formed by connecting black pixels, and a predetermined figure including one connected component among other combinations of connected components extracted by the connected component extracting means, A combination of connected components whose distance value to a predetermined figure including the connected component is less than or equal to a threshold value and the area ratio of the two figures is within a certain range with respect to a separately determined reference value is set as one extraction target. Connected component integrating means for integrating, and character recognition information for extracting information about a character to be recognized from the connected components finally extracted under the action of the connected component extracting means and the connected component integrating means. A character slicing device comprising: a report extracting unit. 6. A character string extracting means for extracting information about a position of a character string from read data of an image including a character, and information about a position of a character string extracted by the character string extracting means. A connected component extracting means for extracting a connected component formed by connecting black pixels existing in an area having a predetermined number of pixels on both sides orthogonal to the character string direction, and among the connected components extracted by the connected component extracting means , A connected component whose start or end coordinates in a direction orthogonal to a character string of a predetermined graphic including the connected component deviates from the existence range of the position of the character string extracted by the character string extraction means. Finally, the connected component deleting means for deleting from the object to be extracted as the connected component forming the character of the character string, and the connected component extracting means and the connected component deleting means finally extract A character slicing device comprising character recognition information extraction means for extracting information about a character to be recognized from the output connected component. 7. A character string extracting means for extracting information about a position of a character string from read data of an image including a character, and information about a position of a character string extracted by the character string extracting means. In a region, a connected component extracting means for extracting a connected component formed by connecting black pixels, and a center of a predetermined figure including one connected component among combinations of connected components extracted by the connected component extracting means Connected component integrating means for integrating a combination in which a straight line connecting the center of a predetermined graphic including other connected components with respect to the character string within a certain range as one extraction target, and the connected component extraction Means and the character recognition information extraction means for extracting information about the character to be recognized from the connected component finally extracted under the action of the connected component integrating means, and the connected component extraction Among the combinations of the connected components extracted by the step, the inclination of the straight line connecting the center of the predetermined figure including one connected component and the center of the predetermined figure including the other connected component with respect to the character string is The auxiliary dictionary unit for character recognition in which information about a character or a combination of characters in which a combination within a certain range occurs is registered in advance, and the connected component integrating means selects one of the combinations whose inclination is within a certain range. If the judgment process of integrating as the extraction target of the character is performed, the character recognition is performed on the basis of the extracted information from the character recognition information extraction means in response to this notification and at least one of the judgment results. A character recognition device comprising a connected component integrated information reflecting character recognition unit that uses information registered in a recognition auxiliary dictionary unit. 8. A character string extracting means for extracting information about a position of a character string from read data of an image including a character, and information about a position of a character string extracted by the character string extracting means. In the region, a connected component extraction unit that extracts a connected component formed by connecting black pixels, and a combination of each connected component extracted by the connected component extraction unit, a black pixel included in one connected component and another Connected component integrating means for integrating a combination whose shortest distance from a black pixel included in a connected component is within a certain range with respect to a separately defined reference value as one extraction target; the connected component extracting means and the connected A character recognition information extracting means for extracting information about a character to be recognized from the connected component finally extracted under the action of the component integrating means, and the connected component extracting means. Among combinations of connected components, a combination in which the shortest distance between a black pixel included in one connected component and a black pixel included in another connected component is within a certain range with respect to a separately determined reference value occurs. An auxiliary dictionary unit for character recognition in which information about a character or a combination of characters is registered in advance, and the connected component integrating means selects one combination that is within a certain range from the shortest distance or the reference value as one extraction target. If the judgment process of integrating the characters is performed, the notification and at least one of the judgment results are received, and when the character recognition is performed based on the extracted information from the character recognition information extracting means, the character recognition auxiliary A character recognition device comprising a connected component integrated information reflecting character recognition unit that uses information registered in a dictionary. 9. A character string extracting means for extracting information about the position of a character string from read data of an image containing characters, and position information of the character string extracted by the character string extracting means of image data. In a region, a connected component extracting means for extracting a connected component formed by connecting black pixels, and a predetermined graphic character including one connected component among combinations of connected components extracted by the connected component extracting means Connected component integrating means for integrating a combination having an inclusive relation between the coordinate values in the column direction of the column and the coordinate values in the column direction of a character of a predetermined graphic including other connected components as one extraction target; A character recognition information extracting means for extracting information about a character to be recognized from the finally extracted connected component under the action of the component extracting means and the connected component integrating means; and the connected component extracting means. Among the combinations of the extracted connected components, the coordinate values in the column direction of the character of the predetermined figure including one connected component and the coordinate values in the column direction of the character of the predetermined figure including the other connected component An auxiliary dictionary unit for character recognition in which information about a character or a combination of characters in which a combination having an inclusive relation occurs is registered in advance, and the connected component integrating means, the combination having the inclusive relation is one extraction target. If you make a judgment process to integrate as
Upon receiving this notification and at least one of the judgment results, when the character recognition is performed based on the extracted information from the character recognition information extracting means, the information registered in the auxiliary dictionary unit for character recognition is used. A character recognition device having a connected component integrated information reflecting character recognition unit. 10. A character string extraction means for extracting information about a position of a character string from read data of an image including a character, and information about a position of a character string extracted by the character string extraction means. In the region, a connected component extracting means for extracting a connected component formed by connecting black pixels, and a predetermined figure including one connected component among other combinations of connected components extracted by the connected component extracting means, Connected component integrating means for integrating as a single extraction target a combination in which the degree of overlap with a predetermined figure including the connected component is greater than or equal to a separately determined threshold, and the operation of the connected component extracting means and the connected component integrating means Character recognition information extraction means for extracting character information from the finally extracted connected component under the following, and a combination of each connected component extracted by the connected component extracting means Information about a character or a combination of characters in which a degree of overlap between a predetermined figure including one connected component and a predetermined figure including another connected component is equal to or more than a separately determined threshold is registered in advance. If the character recognition auxiliary dictionary unit and the connected component integration means perform a determination process of integrating a combination having an overlapping degree of a threshold value or more as one extraction target, this notification, at least the determination result thereof. In response to the above, when performing character recognition based on the extracted information from the character recognition information extraction means, a connected component integrated information reflecting character recognition section that uses information registered in the character recognition auxiliary dictionary section is provided. A character recognition device having. 11. A character string extracting means for extracting information about a position of a character string from read data of an image including a character, and information about a position of a character string extracted by the character string extracting means. In the area, a connected component extracting means for extracting a connected component formed by connecting black pixels, and a predetermined figure including one connected component among the combinations of the connected components extracted by the connected component extracting means and others The distance value to a predetermined figure including the connected component is less than or equal to a threshold value, and the combination of the connected components in which the area ratio of the two figures is within a certain range with respect to a separately determined reference value is set as one extraction target Connected component integrating means for integrating, and character recognition for extracting information about a character to be recognized from the connected components finally extracted under the action of the connected component extracting means and the connected component integrating means. A distance value between a predetermined figure including one connected component and a predetermined figure including another connected component is equal to or less than a threshold value among the combinations of the information extracting means and the respective connected components extracted by the connected component extracting means. And an auxiliary dictionary unit for character recognition in which information about a character or a combination of characters in which a combination of connected components is generated in which the area ratio of the two figures is within a certain range with respect to a separately determined reference value is registered in advance. If the connection component integrating means performs a determination process of integrating a combination in which the distance value is equal to or less than a threshold value and the area ratio is within a certain range with respect to a reference value as one extraction target, this notification, the Upon receiving at least one of the judgment results and performing character recognition based on the extracted information from the character recognition information extracting means, a sequence of using the information registered in the auxiliary dictionary unit for character recognition is used. Character recognition device, characterized in that a component integrated information reflecting the character recognition unit. 12. A character string extracting means for extracting information about a position of a character string from read data of an image including a character, and information about a position of a character string extracted by the character string extracting means. A connected component extracting means for extracting a connected component formed by connecting black pixels existing in a region having a predetermined number of pixels larger on both sides orthogonal to the region than the region, and among the connected components extracted by the connected component extracting device , A connected component whose start or end coordinates in a direction orthogonal to a character string of a predetermined graphic including the connected component deviates from the existence range of the position of the character string extracted by the character string extraction means. Finally, under the action of the connected component deleting means for deleting from the object to be extracted as the connected component forming the character of the character string, the connected component extracting means and the connected component deleting means. A character recognition information extraction unit that extracts information about a character to be recognized from the extracted connected component, and among the connected components extracted by the connected component extraction unit, the connected component deletion unit is a target for exerting the action. A character recognition auxiliary dictionary unit that pre-registers information about a character or a combination of characters that is likely to have a connected component, and the connected component deleting unit deletes the connected component from one extraction target. If this is done, upon receiving this notification, when recognizing the character in the other column including the deleted connected component as a part at the end based on the extracted information from the character recognition information extracting means, the character A character recognition device having a connected component deletion information reflecting character recognition unit that uses information registered in a recognition auxiliary dictionary unit. 13. The circumscribed quadrangle creation unit, wherein the connected component integrating means creates a quadrangle circumscribing the connected component and each side of which is a character string direction or a direction orthogonal thereto as a predetermined figure including the connected component. Claim 1, claim 2, claim 3, claim 4, claim 5, claim 6, claim 7, claim 8, claim 9, or claim 10 characterized in that Character cutting device. 14. A circumscribed quadrangle creation unit that creates a quadrangle circumscribing the connected component and each side of which is in the character string direction or in a direction orthogonal to this as the predetermined figure including the connected component. The character slicing device according to claim 6 or 12, further comprising: