JPH04169986A - Character recognizing device - Google Patents

Abstract

PURPOSE:To improve the recognition accuracy of a specific small character or the like by changing the segmenting of a character in accordance with its character sort. CONSTITUTION:A character image is segmented in each line and each segmented line image is segmented in each character image. A character sort deciding means 6 decides the character sort of the segmented character image based upon the size of the character image and a reference character frame segmenting part 7 segments a character image corresponding to the character image based upon the decided result, changes the size of the segmented character image and then segments the character image again from the line image concerned. Thus, the segmented character image is recognized based upon its character size. Thus, a character can properly be segmented in accordance with its character sort and the recognition accuracy of a specific small character e.g. can be improved.

Description

[Detailed description of the invention] [Industrial application field]

The present invention relates to a character recognition device that inputs a document image, recognizes the document image, and converts the document image into a character code.

[Conventional technology]

Conventional character recognition devices photoelectrically read a document using a scanner or the like, and perform a so-called cutting process in which the text is separated into individual characters. For example, as shown in FIG. 2, characters are cut out using a character frame of a constant size. In this case, a histogram of each character width is taken in the line direction (the direction in which characters are arranged), and the line width is determined from the width of this histogram. Then, the size of the character frame is determined based on this line width. After normalizing the position of the character image cut out in this way, it is enlarged or reduced and converted to a fixed size, and the size of the character is normalized. However, such a method requires normalization of the position, and also causes variations in the size of the character frame corresponding to each character due to the inclination of the character. For this reason, there is a problem that the relative size of the characters to the character frame varies (this means that it is difficult to properly normalize the size of the characters in the true sense of the word.Next, such problems In order to improve the character image, as shown in Figure 3, a circumscribed rectangle for each character is found for each character, and the character image is enlarged or reduced to a fixed size within this circumscribed rectangle and converted. One possible method is to normalize the size of the characters. According to this method, there is no need to normalize the position of the character frame, and since the character frame is not used, the enlargement or reduction ratio of the same character is constant. More appropriate normalization is possible. However, this method does not allow punctuation marks (.,
), periods (,), and other special lowercase characters, important positional information for these characters (relative positional information within the character frame) is lost. In addition, such special lowercase letters have a large enlargement ratio (as shown in Figure 4), so there is a problem that the outline of the character pattern appears jagged and the character image is distorted. Distinguishing between (,) and bullets (・) 1. It may be difficult to distinguish between a period (.) and a white circle (○), or between a comma (,) and a repetition symbol (A), or they may be misrecognized as other characters. These special lowercase characters, such as punctuation marks, are important characters that are used as clause separation information for post-processing parsing that is normally performed by character recognition devices. Misrecognition becomes a serious problem in character recognition.

[Problem to be solved by the invention]

As explained above, in the conventional character cutting method,
In the first method shown in FIG. 2 described above, it is necessary to normalize the position of the character frame, and the size of the character frame varies due to the inclination of the characters. As a result, the relative size of the character to the character frame varies, causing a problem in which it is difficult to normalize the size correctly. In addition, in the second method shown in Figure 3, for special lowercase characters such as punctuation marks (., ,), commas (1), and periods (,), important positional information (relative within the character frame) is used as a feature of these characters. Location information) and size information may be lost, or other characters may be misrecognized. The present invention has been made in view of the above-mentioned Conventional Example 1, and an object of the present invention is to provide a character recognition device that improves the recognition accuracy of special lowercase letters by changing the extraction of characters depending on the character type.

[Means to solve the problem]

In order to achieve the above object, the character recognition device of the present invention has the following configuration. That is, a character recognition device inputs a document image and performs character recognition on the document image, comprising a line cutting means for cutting out the document image line by line, and a line cutting means for cutting out each line image cut out by the line cutting means. a character cutting means for cutting out each character image; a determining means for determining the character type of the character image based on the size of the character image cut out by the character cutting means; A cutting means for cutting out a character image corresponding to the image again from the corresponding line image by changing the cutting size, and a character recognition means for recognizing the character cutting means or the character image cut out by the cutting means. do.

[Effect]

In the above configuration, a document image is cut out line by line, and each line image is cut out in character image units. Based on the size of the character image extracted in this way, the character type of the character image is determined, and according to the determination result, the character image corresponding to the character image is cropped again from the corresponding line image by changing the cropping size. It works like this. In this way, based on the cut out character image, the character recognition of the character image is performed.

【Example】

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. FIG. 1 is a block diagram showing the configuration of a character recognition device according to a first embodiment of the present invention. In FIG. 1, a page memory 1 stores image data for at least one page of a document inputted from a scanner (not shown) or the like. A line buffer 2 stores image data of each line cut out by the line cutter 4 out of the image stored in the page memory 1. A character buffer 3 stores an image of a character portion cut out by the circumscribed rectangle cutout unit 5. Reference numeral 4 denotes a line cutting unit that can cut out each line of data in the page memory 1. Reference numeral 5 denotes a circumscribed rectangular cutting section which cuts out an image of a character portion of the image data of the page memory 1, for example, using a rectangular frame as shown in FIG. Reference numeral 6 denotes a character type determining unit which determines whether the character type cut out by the circumscribed rectangle cutting unit 5 is a standard character or a special lowercase character such as a punctuation mark. Reference numeral 7 denotes a standard character frame cutting unit, which performs character cutting processing again when the cut out character h is a special lowercase character. 8 is a normalization unit which normalizes the extracted character image. 9 is a feature extraction part,
The features of the character image are extracted, and the recognition unit 10 uses the dictionary unit 1
1, this character image is recognized. 12 is a code buffer that can store recognized character information. The operation of the character recognition device with the above configuration will be explained below. The page memory 1 stores one page worth of document images (image data) read from a scanner or the like (not shown). Then, the page memory 1 is
Count the number of black pixels in the document image in the row direction and create a histogram. Next, by cutting the image data in the row direction at the valleys of this histogram, 1
One line image data is cut out from the page image and the line data (image data) is stored in the line buffer 2. Next, as shown in FIG. 3, the line image data in the line buffer 2 is cut out by the circumscribing rectangle cutting unit 5 into a character image for each character using a circumscribing rectangular frame that circumscribes each character. Stores image data for one character. At this time, the height b+ (i=1.2.・
), the maximum value hM is determined, and this value is transferred to the register of the character type determining section 6. The character type determination unit 6 calculates hl/h for each character,
It is checked whether this value is less than or equal to a predetermined threshold value S (for example, 1/2). If (hr/hM)<S, it is determined that the first character is a special lowercase character, and a re-cutting signal is sent to the standard character cutting section 7. As a result, the standard character frame cutting section 7 recuts the first character in the hMxh, l size character frame and inserts it back into the character buffer 3, as shown in FIG. On the other hand, the character type determination unit 6 selects (bl /hM
)≧S, the standard character frame cutting section 7 does not operate, and the data in the character buffer 3 is the data itself cut out by the circumscribed rectangle cutting section 5. The normalization unit 8 enlarges or reduces the character image in the character buffer 3 to a fixed size (for example, 64×64 dots). Next, after the feature extraction section 9 extracts the features of the character image using a known method, the identification section 10 and the dictionary section 11
Characters are identified by The character code of the character thus recognized is stored in the code buffer 12. As explained above, according to the character recognition device of this embodiment,
Standard size full-width characters are normalized to a constant size that does not change due to character cutting. On the other hand, in the case of special lowercase letters, the character size can be normalized while retaining position information and size information, making it possible to achieve highly accurate character recognition that allows the mixture of multiple fonts. <Other Embodiments (Fig. 6)> Fig. 6 is a block diagram showing a schematic configuration of a character recognition device according to a second embodiment. Parts common to those in Fig. 1 are indicated by the same numbers, and their explanations are given below. Omitted. A recognition unit 62 includes a simple feature extraction unit and a simple identification unit (not shown). Here, simple means that the same features as some or all of the feature amounts extracted by the feature extraction unit 9 are extracted, and the extracted features are used. 63 is dictionary ■ which is also a simple recognition dictionary. Reference numeral 64 denotes a character type determination unit that determines whether it is a character or a character portion. 72 is a synthesis section. Following the same procedure as in the first embodiment described above, the character image is cut out into a circumscribing rectangle by the circumscribing rectangle cutting unit 5, and then simple recognition is performed in the recognition unit I62 using a dictionary technician according to a known method. That is, here, the distance d between the standard pattern included in the dictionary (2) and the extracted character image is calculated. Next, the character type determination unit 64 checks whether the distance d is greater than a predetermined threshold. If the distance d is greater than a predetermined threshold, it is determined that the character is not a character, that is, it is a part of a character, and an operation signal is sent to the synthesis unit 72. As a result, the compositing unit 72 combines the current image (image data) and the next image (
(next image data cut out by the circumscribed rectangle cutting unit 5) to create one new character image and store it in the character buffer 3. Then, similar to the first embodiment, processing after normalization is performed. If the distance d is not greater than the threshold, the character type determining unit 64 considers that the image being processed is a character image, and immediately performs normalization and subsequent processing without sending an operation signal to the compositing unit 72. As described above, according to the other embodiments, it is possible to prevent separated characters (such as い, す, 情, 正, etc.) from being separated and cut out due to character cutting. As described above, according to this embodiment, it is possible to cut out characters appropriately according to the character type, and the recognition accuracy of character recognition can be greatly improved.

【Effect of the invention】

As explained above, according to the present invention, since characters can be appropriately cut out according to the character type, the recognition accuracy of, for example, special lowercase characters can be improved.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing a schematic configuration of a character recognition device according to a first embodiment of the present invention, FIG. 2 is a diagram for explaining an example of conventional character segmentation, and FIG. 3 is a diagram showing a character recognition device of this embodiment. FIG. 4 is a diagram showing an example of extraction, FIG. 4 is a diagram illustrating a state in which image distortion occurs due to conventional character image normalization, FIG. 5 is a diagram illustrating a character extraction method for special lowercase letters in this embodiment, and FIG. 6 is a block diagram showing a schematic configuration of a character recognition device according to a second embodiment. In the figure, 1... page memory, 2... line buffer, 3
... Character buffer, 4... Line cutting section, 5... Circumscribing rectangle cutting section, 6... Character type determination section, 7... Standard character frame cutting section, 8... Normalization section, 9 ...Feature extraction unit, 10... Identification unit, 11... Dictionary, 12... Code buffer, 62... Recognition unit ■, 63... Dictionary technician, 64... Character type determination unit 72...Synthesis section. Unexploded day and month 1 child, length '8 kan, island island \n meru. RO Kunimi Doro ward, ward, country - Hajime Miguchi ■ Fig. 2 The present invention 1, Bunji ko L iron 4g, η, riru. Trouble F Old Bow Roro Menu Mekumetsu day, day, day, day, day, day, day, day, day, day, day, day, day, day, day, day of Russia, day of day, day of Russia, day of day, day of Russia, 80°11 decoy translation, day of day, day and age of Russia. Figure 3 Figure 5

Claims (3)

[Claims] (1) A character recognition device that inputs a document image and performs character recognition on the document image, comprising a line cutting means for cutting out the document image line by line, and each line image cut out by the line cutting means. character cutting means for cutting out the text into character image units; determining means for determining the character type of the character image based on the size of the character image cut out by the character cutting means; a cutting means for cutting out a character image corresponding to the character image again from the corresponding line image by changing the cutting size; a character recognition means for recognizing the character cutting means or the character image cut out by the cutting means; A character recognition device comprising: (2) The determining means determines that the character image is of a different character type when the size of the character image is less than or equal to a predetermined value compared to the maximum character size of the line, and the cutting means converts the determined character image into the 2. The character recognition device according to claim 1, wherein the character recognition device is configured to cut out the character at a size corresponding to a maximum value. (3) A character recognition device that inputs a document image and performs character recognition on the document image, comprising a line cutting means for cutting out the document image line by line, and each line image cut out by the line cutting means. a character cutting means for cutting out the character image into character image units; a determining means for determining whether a distance between a character image cut out by the character cutting means and a character stored in a dictionary is greater than or equal to a predetermined value; and recognition means for combining and recognizing a character image corresponding to the character image and a character image to be next cut out by the character cutting means when it is determined that the distance is equal to or greater than a predetermined value. A character recognition device featuring: