JPH11110485A - Device and method for reading character - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make surely readable characters without being affected by a background pattern or the fluctuation of a luminance value, by extracting the line area of an inputted image, binarizing the image of the line area, recognizing the character segmented from the provided binarized image and reading the character information of the input image. SOLUTION: At a photoelectric transducing part 101, the image of a surface where lot numbers are printed on a postcard conveyed by a conveyer line is inputted as a multilevel image. At a line range extracting part 102, a line range is extracted from the provided multilevel image. At a binarizing pot 103, only the provided range of lines is binarized. At a character detecting and segmenting part 104, a character pattern stream is extracted as each character pattern. At a character recognizing part 104, each extracted character pattern is matched with a character pattern registered on a prepared dictionary, for example, and character recognizing processing is performed.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character reading apparatus for reading characters printed on a multicolor printed background such as a postcard with lottery.

[0002]

2. Description of the Related Art As a character reading procedure of a conventional character reading apparatus, first, a character string written on the surface of an object such as paper or metal is input as a multivalued image, and a line range is determined from the multivalued image. Extract (row range extraction process). The line range is a range in which a character string exists in a direction perpendicular to the arrangement of characters.
Then, a binary image is generated only for the extracted line range (binary processing), individual character patterns are extracted (character detection cutout processing), and the extracted character patterns are recognized as characters. Outputs a character string as a reading result.

[0003] Hereinafter, a conventional method for a line range extraction process, a binarization process, and a character detection cutout process will be described in detail.

(1) Row range extraction processing First, a minimum projection is obtained from an input multi-valued image. The minimum projection is a one-dimensional waveform obtained by scanning a multivalued image in the direction in which the character patterns are arranged and obtaining the minimum value of the luminance value for each line. Next, the differential value is calculated. Finally, the maximum value of the absolute value of the derivative is obtained, and a section between the two maximum values is extracted as a row range.

(2) Binarization processing First, a luminance histogram is generated from a multi-valued image. Next, a threshold for binarization is determined from the luminance histogram.
Here, a method using inter-class variance is mentioned as a typical method of determining a threshold value for binarization.

The determination of the binarization threshold is a process of dividing the luminance value into two classes, a class larger than the threshold and a class smaller than the threshold. The inter-class variance when the threshold is k is
Find for all k. K that maximizes the inter-class variance is obtained as a threshold for binarization.

This method is intuitively a method of determining a threshold value so that the two classes are well united and the separation between the classes is improved, and is stable without adjusting the parameters. The feature is that an appropriate binarization threshold can be obtained.

Finally, in the multi-valued image, a pixel larger than the threshold value and a pixel smaller than the threshold value are replaced with white and black, respectively, to generate a binary image.

(3) Character detection cutout processing First, labeling is performed on a binary image. Next, by label filtering, of the labels obtained by labeling, those that do not satisfy the criterion of the size as the character pattern are removed. This is performed for the purpose of removing extremely small noise labels as character patterns and extremely large background pattern labels. Finally, a character pattern extraction process is performed. Labels that are input to this process include:
Background noise and background patterns that cannot be removed by label filtering may be included. Therefore,
Based on information obtained in advance such as the size of the character pattern and the printing pitch, only the label corresponding to the character pattern is extracted from the input label.

[0010]

The conventional line range extraction process has a problem that the line range is difficult to extract, especially in the case of a multivalued image of a character printed on a multicolored background. Was. That is, as shown in FIG. 1, when the background pattern exists in the multi-valued image, a plurality of local maxima appear in the absolute value of the differential waveform of the minimum projection, and the boundary of the row range appears as a local maximal value. Absent. In the conventional row range extraction method, a maximum value pair having a large value is found in the absolute value of the differential waveform of the minimum projection. In such a case,
Line range cannot be extracted correctly.

In the conventional binarization processing for determining a threshold value using inter-class variance, an appropriate threshold value is set when the background area for the character pattern string is large and the luminance value of the character pattern string and the background are close. There was a problem that was not obtained. This is because the conventional threshold determination method assumes that the area of the character pattern is equal to the area of the background. When the area of the background with respect to the character pattern string is large and the luminance values of the character pattern string and the background are close, the luminance histogram is as shown in FIG. 2, for example. In such a case, the threshold obtained by the threshold determination method is, as shown in FIG.
A binary image is generated that is brighter than an appropriate threshold value, that is, closer to the background side, and thus includes background noise.

Further, in the conventional character detection cutout process, contact between the background pattern and the character pattern is not assumed, and the process of cutting out the character pattern from the character pattern touching the background pattern is not performed. Therefore, when the background pattern and the character pattern come into contact with each other, there is a problem that the character pattern cannot be recognized. This is because a character pattern in contact with a background pattern may be removed by label filtering as not satisfying a criterion as a character pattern and may not be included in a character pattern string obtained as a result of the character detection cutout process. It is.

In view of the above-mentioned problems, the present invention provides a high-precision high-precision which can reliably read even a character printed on a multicolor printed background without being affected by a change in a background pattern or a luminance value. It is an object to provide a character reading device and a character reading method.

[0014]

According to the present invention, there is provided a character reading apparatus for scanning a partial image of an input image in a predetermined character string direction to obtain minimum projection information; High-frequency component removing means for removing high-frequency components from the minimum projection information, row extracting means for extracting a row area based on the minimum projection information from which the high-frequency component has been removed, and a row area extracted by the row extracting means. A binarizing unit for binarizing an image, a character extracting unit for extracting characters from the binarized image obtained by the binarizing unit, and a character extracting unit for recognizing the character extracted by the character extracting unit. With the provision of the character reading means for reading the character information of the input image, it is possible to extract a line range that is hardly affected by a change in the background pattern or the luminance value. Therefore, even a character printed on a multicolor printed background can be reliably read without being affected by a change in the background pattern or the luminance value.

According to a second aspect of the present invention, there is provided a character reading apparatus for scanning a partial image of an input image in a predetermined character string direction to obtain minimum projection information; High frequency component removing means for removing components,
A line extracting means for differentiating the minimum projection information from which the high-frequency component has been removed, and extracting a region in which the projection value does not change between a position where the differential value is negative and a position where the differential value is positive as a line region. Binarizing means for binarizing the image of the line area extracted by the line extracting means, character extracting means for extracting characters from the binarized image obtained by the binarizing means, and character extracting means. Character reading means for recognizing the characters cut out by the output means and reading the character information of the input image, whereby the line range which is hardly affected by the background pattern and the fluctuation of the luminance value can be extracted. . Therefore, even a character printed on a multicolor printed background can be reliably read without being affected by a change in the background pattern or the luminance value.

According to a third aspect of the present invention, there is provided a character reading apparatus for scanning a partial image of an input image in a predetermined character string direction to obtain minimum projection information; High frequency component removing means for removing components,
A line extracting unit that extracts a line region based on the minimum projection information from which the high-frequency component has been removed, a unit that obtains a luminance distribution of an image of the line region extracted by the line extracting unit, Means for determining a binarization threshold based on a luminance distribution obtained by deleting a distribution area corresponding to the luminance of the image according to a predetermined criterion, and binarizing the image of the row area using the binarization threshold. Binarizing means, character extracting means for extracting characters from the binarized image obtained by the binarizing means, and characters of the input image by recognizing the characters extracted by the character extracting means With the provision of the character reading means for reading the information, the binarization threshold can be determined without being influenced by the background pattern or the fluctuation of the luminance value. Therefore, even a character printed on a multicolor printed background can be reliably read without being affected by a change in the background pattern or the luminance value.

According to a fourth aspect of the present invention, there is provided a character reading apparatus for scanning a partial image of an input image in a predetermined character string direction to obtain minimum projection information; High frequency component removing means for removing components,
A line extracting unit for extracting a line region based on the minimum projection information from which the high frequency component has been removed, a binarizing unit for binarizing an image of the line region extracted by the line extracting unit, and a binarizing unit Labeling means for performing labeling on the binarized image obtained by the means; and detecting a label of a character in contact with a background image based on the size of each label obtained by the labeling means, and detecting peripheral characteristics of the label. A first character extracting unit that extracts only characters from the character label based on a predetermined character size, and a character extracting unit that extracts characters based on a predetermined character size and a character interval from the label obtained by the labeling unit. 2 character extracting means, and character reading means for recognizing characters extracted by the first and second character extracting means and reading character information of the input image. By the, even when the character pattern and the background pattern is in contact, allows excision of the characters.
Therefore, even a character printed on a multicolor printed background can be reliably read without being affected by a change in the background pattern or the luminance value.

A character reading apparatus according to the present invention is characterized in that a partial image of an input image is scanned in a predetermined character string direction to obtain minimum projection information, and a high frequency High frequency component removing means for removing components,
A line extracting means for differentiating the minimum projection information from which the high-frequency component has been removed, and extracting a region in which the projection value does not change between a position where the differential value is negative and a position where the differential value is positive as a line region. Means for obtaining a luminance distribution of an image of a line area extracted by the line extracting means; and a luminance distribution obtained by deleting a distribution area corresponding to background luminance from the luminance distribution according to a predetermined reference. Means for determining a binarization threshold value on the basis of: a binarization means for binarizing the image of the row area using the binarization threshold value; and a binarization image obtained by the binarization means. Labeling means for performing labeling;
When a label of a character in contact with the background image is detected based on the size of each label obtained by the labeling means, a first character that cuts out only characters from the character label based on the peripheral characteristics of the label and a predetermined character size is used. Character extracting means, and a second character extracting means for extracting a character based on a predetermined character size and character interval when a character label is detected based on the label size obtained by the labeling means, Character reading means for recognizing the character extracted by the first and second character extracting means and reading the character information of the input image, whereby the character printed on the multicolor printed background is provided. However, it is possible to reliably read the image without being affected by the fluctuation of the background pattern or the luminance value.

According to a character reading method of the present invention, a partial image of an input image is scanned in a predetermined character string direction to obtain minimum projection information; Removing the row region based on the minimum projection information from which the high-frequency component has been removed;
Digitizing, extracting characters from the binarized image, and recognizing the extracted characters and reading the character information of the input image. A line range that is not easily affected by fluctuations of the line can be extracted. Therefore, even a character printed on a multicolor printed background can be reliably read without being affected by a change in the background pattern or the luminance value.

The character reading method according to the present invention (claim 8) includes a step of scanning a partial image of an input image in a predetermined character string direction to obtain minimum projection information; Removing the component, and differentiating the minimum projection information from which the high-frequency component has been removed, and defining a region in which the projection value does not change between a position where the differential value is negative and a position where the differential value is positive is a row region. Extracting, extracting the character of the input image by recognizing the extracted character, extracting the character from the binarized image, The step of reading makes it possible to extract a line range that is not easily affected by a change in the background pattern or the luminance value. Therefore, even a character printed on a multicolor printed background can be reliably read without being affected by a change in the background pattern or the luminance value.

According to a ninth aspect of the present invention, there is provided a character reading method comprising: scanning a partial image of an input image in a predetermined character string direction to obtain minimum projection information; Removing a component, extracting a row region based on the minimum projection information from which the high-frequency component has been removed, obtaining a luminance distribution of an image of the extracted row region, and Determining a binarization threshold based on a luminance distribution obtained by deleting a distribution area corresponding to background luminance in accordance with a predetermined criterion, and converting the image of the row area into two using the binarization threshold Digitizing, extracting a character from the binarized image, and recognizing the extracted character to read character information of the input image. Background pattern and can determine the binarization threshold without being affected by the variation in the luminance values. Therefore, even a character printed on a multicolor printed background can be reliably read without being affected by a change in the background pattern or the luminance value.

The character reading method of the present invention (claim 10)
Scanning a partial image of the input image in a predetermined character string direction to obtain minimum projection information; removing high frequency components from the minimum projection information; and Extracting a row region based on the projection information; binarizing the image of the extracted row region; labeling the binarized image; and labeling each label obtained by the labeling. When detecting a label of a character in contact with the background image based on the size, a first character extraction step of extracting only characters from the character label based on the peripheral characteristics of the label and a predetermined character size, and the labeling A second character extraction step of extracting characters from the obtained label based on a predetermined character size and character spacing; A step of recognizing the character extracted in the first and second character extraction steps and reading the character information of the input image. Can be cut out. Therefore, even a character printed on a multicolor printed background can be reliably read without being affected by a change in the background pattern or the luminance value.

The character reading method of the present invention (claim 11)
Scanning a partial image of the input image in a predetermined character string direction to obtain minimum projection information; removing high frequency components from the minimum projection information; and Differentiating the projection information, extracting a region in which the projection value does not change between a position where the differential value becomes a positive maximum value and a position where the differential value becomes a negative maximum value as a row region, and Determining a luminance distribution of the image of the row region; and determining a binarization threshold based on the luminance distribution obtained by deleting a distribution region corresponding to the luminance of the background from the luminance distribution according to a predetermined reference. A step of binarizing the image of the row region using the binarization threshold, a step of labeling the binarized image, and a step of labeling each label obtained by the labeling. When a label of a character in contact with the background image is detected based on the label, a first character extraction step of extracting only characters from the character label based on the peripheral characteristics of the label and a predetermined character size, and the labeling When a character label is detected based on the obtained label size, a second character extracting step of extracting a character based on a predetermined character size and a character interval, and the first and second character extracting steps are performed. Recognizing the cut-out character and reading the character information of the input image, even if the character is printed on the multicolor printed background, is not affected by the change in the background pattern or the luminance value. Can be reliably read.

[0024]

Embodiments of the present invention will be described below with reference to the drawings.

According to the present invention, characters are read from an object having characters printed on a multicolor printed background, for example, a lottery number of a postcard with lottery or a ticket issue number.

FIG. 3 shows a schematic configuration of a lottery and postcard sorting device to which the character reading device of the present invention is applied. The transport line 1, the stacking unit 3, the reading unit 5, the recognizing unit 7,
The control unit 13 includes sorting gates 9 to 11 and stacking units 15 to 17.

The transport line 1 is for transporting the postcards 2 from the stacking unit 3 for sorting out New Year's postcards with lottery (hereinafter simply referred to as "postcards") to the above-mentioned sections for sequentially taking out and transporting the postcards 2 one by one.

The reading section 5 is constituted by, for example, a CCD camera or the like which is provided in the middle of the transport line 1 and reads the lottery number described on the postcard 2 being transported. The character information read by the reading unit 5 is sent to the recognition unit 7.

The recognition unit 7 compares the character information (lottery number) read by the reading unit 5 with a winning number stored in advance, and determines the postcard 2 currently being transported on the transport line 1.
It is determined whether or not the lottery number matches the winning number.

A control unit 13 is connected to the recognizing unit 7. The control unit 13 selects sorting gates 9, 10, 10.
11 is driven and controlled.

The sorting gate 9 guides the postcard to the stacking unit 15, the sorting gate 10 guides the postcard to the stacking unit 16, and the sorting gate 11 determines that the postcard cannot be read. Is led to the other accumulation unit 17.

In such a configuration, the recognizing section 7 determines whether the lottery number read from the postcard 2 being conveyed by the reading section 5 and a predetermined winning number are determined by, for example, the composite similarity, and the recognition section 7 determines whether the numbers match. If it is judged as "winning", sorting gate 10
In contrast, if they do not match, the command signal for driving the sorting date 9 is output to the control unit 13, and as a result, the conveyed postcard is determined. It is guided to the out-of-stack accumulating unit 15 or the hit accumulating unit 16 based on the respective pass / fail.

With such a configuration, it is possible to automatically determine whether or not a lottery is required without manual intervention, so that the determination as to whether or not a large number of lottery postcards can be performed quickly.

FIG. 4 shows an example of the configuration of a character reading device used in the reading unit 5 of the lottery postcard sorting device shown in FIG.
It comprises a value conversion unit 103, a character detection cutout unit 104, and a character recognition unit 105.

The character reading apparatus of the present invention is particularly effective when reading characters printed on a multicolor printed background on an object such as paper, such as a lottery number printed on a lottery postcard. is there. For example, an image of an area where characters to be read (a lottery number in the case of a postcard with lottery) is printed is an image as shown in FIG. 5 and has the following features that make the character reading process difficult.

There are a plurality of background patterns.

The background pattern may come into contact with the character pattern.

The contrast between the background and the character pattern is low.

Next, an outline of the processing operation of the character reading apparatus of FIG. 4 will be described with reference to FIG.

First, in the photoelectric conversion unit 101, the image of the surface on which the lottery number of the postcard conveyed on the conveyance line 1 is printed is input as a multi-valued image.

The row range extraction unit 102 extracts a row range from the obtained multi-valued image. The binarization unit 103 binarizes only the obtained row range. The character detection cutout unit 104 extracts a character pattern string as individual character patterns. The character recognition unit 105 performs a character recognition process by matching the extracted individual character patterns with, for example, character patterns registered in a prepared dictionary.

Next, the line range extracting unit 102, the binarizing unit 103, and the character detecting and extracting unit 104, which are features of the present invention, will be described in detail.

(1) Row Range Extraction Unit The processing operation of the row range extraction unit 102 will be described with reference to FIG.

An image of a predetermined print character area of the input multivalued image G (x, y), that is, a partial image g (x, y)
The following minimum projection P _min (y) is obtained for y) (step S1). The minimum projection is the minimum projection value of the line obtained as a result of scanning the partial image g (x, y) in the direction in which the character patterns are arranged.

[0045]

(Equation 1)

Here, as the partial image g (x, y), an image having a limited range in the x direction is used for speeding up the processing.
In FIG. 7, G is used as the partial image g (x, y).
A middle image obtained by dividing (x, y) into three equal parts in the x direction is used.

At the minimum projection P _mim , the range of the row appears as a valley with a large width and depth as shown in FIG. The valleys appearing not only correspond to the row range, but also include valleys corresponding to the background pattern and high-frequency components of the image.

[0048] Therefore, the purpose of removing the trough corresponding to the non-character pattern string by removing high frequency components from the P _mim (y) to calculate the smoothed waveform P _"mim (y) (step S2) .P Before _{obtaining mim} (y), the following waveform P ′ _mim (y) is calculated once.

[0049]

(Equation 2)

[0050] calculated from the P _'mim of P _"mim is carried out in the following manner.

[0051]

(Equation 3)

Here, w is the width of the smoothing, and is set slightly smaller than the height of the character pattern string. The height of the character pattern depends on the size of the font and the photoelectric conversion unit 101.
Can be predicted from the input resolution at.

Next, a row range is extracted by detecting a valley from P ″ _mim (y) (step S3). Specifically, as shown in FIG. 8, the first derivative of P ″ _mim (y) , And a range in which the value of the first derivative is “0” (no change in the projection value) between the negative value and the positive value is extracted as a row range.

(2) Binarization Unit The processing operation of the binarization unit 103 will be described with reference to FIG.

First, a luminance histogram h (z) of a multivalued image for the extracted row range is obtained (step S1).
1).

Next, the distribution area corresponding to the luminance of the background is deleted from the luminance histogram according to a predetermined reference (step S12). That is, instead of directly applying the threshold determination method using interclass variance to h (z), h
A histogram cut is performed on (z) to obtain h ′ (z), and a conventional threshold value determination method is applied to h ′ (z) (step S13).

The histogram cut is performed as follows.
h included in the range of z (brightness value) satisfying z> z _area
(Z) _Find z _area such that the area is T _area % of the entire area of the histogram. For z satisfying z> z _area , the value of h (z) is replaced with “0”. The obtained waveform is defined as h ′ (z).

The purpose of the histogram cut is to reduce the area of the portion corresponding to the background in the histogram,
The purpose is to obtain an appropriate threshold value by a threshold value determination method using inter-class variance.

In the histogram, the value of T _area % is set to a value that satisfies the following two conditions in the histogram: (1) the portion corresponding to the background is not completely removed; (2) the area of the portion corresponding to the background is sufficiently reduced. deep.

Finally, binarization is performed on the area of the row range of the input multi-valued image at the determined threshold value (step S1).
4).

(3) Character Detection and Extraction Unit The processing operation of the character detection and extraction unit 104 will be described with reference to FIG.

First, labeling is performed on the binary image (step S21). Next, in the contact character pattern cutout processing, for example, a processing of cutting out a character pattern in contact with the background pattern, such as the second “0” from the right in FIG. 10, from the background pattern (step S22). Here, the size of the character (character pattern) to be read is predetermined. Therefore, the size of the label can also be determined in advance for each character. When the label is in contact with the background pattern, the label becomes larger than the predetermined size. Therefore, the character pattern in contact with the background pattern can be determined based on the size of the label.

Finally, in the character pattern extraction processing, individual character patterns constituting a character pattern sequence are extracted from the label based on information obtained in advance such as the size of the character pattern and the printing pitch ( Step S2
3). At this time, only the character pattern is extracted from the label of the character pattern that is in contact with the background pattern (in FIG. 10, the second “0” from the right).

Here, focusing on the character pattern (the second “0” from the right in FIG. 10) in contact with the background pattern, FIG. 11 shows the cutout processing in step S22 and the character pattern extraction processing in step S23. Reference is made in detail.

First, a label formed by contact between a character pattern and a background pattern is found. This is obtained based on the size and position of the circumscribed rectangle of the label. If the background pattern is above the character pattern, cut out the obtained label as follows, and if the background pattern is below the character pattern, reverse the following algorithm upside down. And cut it out.

In the character extraction processing, first, a peripheral pattern PERI (x) is obtained (step S31). The peripheral pattern is scanned from the bottom to the top in the circumscribed rectangle of the label as shown in FIG.
As shown in (b), the distance from the lower end of the circumscribed rectangle to the first black pixel that hits is taken as the value of the peripheral pattern.

Then, the horizontal range of the character pattern is obtained from the peripheral pattern (step S32). PE
A horizontal range in which RI (x) has a value larger than the threshold value _Hth is extracted as a horizontal range. The threshold value _Hth is a value that can be determined in advance because the character size is known.

Subsequently, a range from the value “0” of the peripheral pattern to the threshold H _std is extracted as a vertical range of the character pattern (step S33). Threshold H
_std is a value that can be determined in advance because the character size is known.

Finally, a range corresponding to the cut-out rectangular range is cut out from the label image (step S34).

In step S22 of FIG. 10, if the label size is substantially the same as the predetermined size, the character pattern extraction process of step S23 does not perform the process shown in FIG. The character pattern area may be cut out based on the determined size and printing pitch of various character patterns.

As described above, according to the above-described embodiment, it is possible to extract a line range which is hardly affected by a change in a background pattern or a luminance value. Also, there is an effect that the adjustment becomes easy. This is because there are few adjustment parameters, and the parameters need only be set in consideration of the height of the character pattern string with little change.

Also, in the binarization process, a simple operation on the luminance histogram allows a conventional threshold such as a case where the background area for the character pattern row is large and the luminance value of the character pattern row and the background are close to each other. In the determination method, an appropriate threshold can be obtained even for a multivalued image for which an appropriate threshold cannot be obtained. Thereby, the performance of the binarization processing is improved.

Further, even when a character pattern and a background pattern come into contact with each other such that detection and extraction cannot be performed by the conventional method, a character pattern string can be detected and extracted. This leads to an improvement in the recognition rate of the character pattern string.

[0074]

As described above, even characters printed on a multicolor printed background can be reliably read without being affected by variations in the background pattern or luminance value.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining a problem of a conventional row range extraction process.

FIG. 2 is a diagram for explaining a problem of a conventional binarization threshold value determination method.

FIG. 3 is a diagram showing a schematic configuration of a lottery and postcard sorting device to which the character reading device of the present invention is applied.

FIG. 4 is a diagram showing a schematic configuration of a character reading device of the present invention.

FIG. 5 is a view showing an example of a multi-value image input to the character reading device of the embodiment.

FIG. 6 is a view for explaining an outline of a processing operation of the character reading apparatus of FIG. 4;

FIG. 7 is a diagram for explaining a processing operation of a row range extraction unit.

FIG. 8 is a diagram for explaining the processing method of step S3 in FIG. 7 in more detail;

FIG. 9 is a diagram for explaining a processing operation of a binarization unit;

FIG. 10 is a diagram for explaining a processing operation of a character detection cutout unit;

FIG. 11 is a diagram for explaining in more detail the extraction processing of step S22 and the extraction processing of the character pattern of step S23 in FIG. 10, paying attention to the character pattern in contact with the background pattern.

FIG. 12 is a view for explaining a method of calculating a peripheral pattern in step S31 of FIG. 11;

[Explanation of symbols]

 101: photoelectric conversion unit 102: row range extraction unit 103: binarization unit 104: character detection cutout unit 105: character recognition unit

Claims (12)

[Claims] A means for scanning a partial image of the input image in a predetermined character string direction to obtain minimum projection information; a high frequency component removing means for removing high frequency components from the minimum projection information; A row extracting means for extracting a row area based on the minimum projection information from which the high-frequency component has been removed; a binarizing means for binarizing an image of the row area extracted by the row extracting means; Character extracting means for extracting characters from the binarized image obtained in the above, and character reading means for recognizing the characters extracted by the character extracting means and reading the character information of the input image. Character reading device. 2. A means for scanning a partial image of an input image in a predetermined character string direction to obtain minimum projection information; a high frequency component removing means for removing high frequency components from the minimum projection information; Line extraction means for differentiating the minimum projection information from which the high-frequency component has been removed, and extracting a region in which the projection value does not change between a position where the differential value is negative and a position where the differential value is positive as a line region, A binarizing unit for binarizing the image of the line area extracted by the line extracting unit; a character extracting unit for extracting a character from the binarized image obtained by the binarizing unit; Character reading means for recognizing characters cut out by the means and reading character information of the input image. A means for scanning a partial image of the input image in a predetermined character string direction to obtain minimum projection information; a high frequency component removing means for removing high frequency components from the minimum projection information; A row extracting unit that extracts a row region based on the minimum projection information from which the high-frequency component has been removed; a unit that obtains a luminance distribution of an image of the line region extracted by the line extracting unit; Means for determining a binarization threshold based on a luminance distribution obtained by deleting a distribution area corresponding to luminance in accordance with a predetermined criterion, and binarizing an image of the row area using the binarization threshold Character extracting means for extracting characters from the binary image obtained by the binarizing means; character information of the input image by recognizing the character extracted by the character extracting means Character reader to read When, character reading apparatus characterized by comprising a. 4. A means for scanning a partial image of an input image in a predetermined character string direction to obtain minimum projection information; a high frequency component removing means for removing high frequency components from the minimum projection information; A row extracting means for extracting a row area based on the minimum projection information from which the high-frequency component has been removed; a binarizing means for binarizing an image of the row area extracted by the row extracting means; Labeling means for labeling the binarized image obtained in the step (a), and when a label of a character in contact with the background image is detected based on the size of each label obtained by the labeling means, the peripheral features of the label and the A first character extracting unit that extracts only characters from the character label based on the determined character size; and a predetermined character size and a predetermined size based on the label obtained by the labeling unit. A second character cutout unit that cuts out characters based on a character interval and a character reading unit that recognizes a character cut out by the first and second character cutout units and reads character information of the input image; A character reading device comprising: 5. A means for scanning a partial image of an input image in a predetermined character string direction to obtain minimum projection information; a high frequency component removing means for removing high frequency components from the minimum projection information; Line extraction means for differentiating the minimum projection information from which the high-frequency component has been removed, and extracting a region in which the projection value does not change between a position where the differential value is negative and a position where the differential value is positive as a line region, Means for calculating the luminance distribution of the image of the line area extracted by the line extracting means; and a luminance distribution obtained by deleting a distribution area corresponding to the luminance of the background from the luminance distribution according to a predetermined reference. Means for determining a binarization threshold value based on the binarization threshold value; binarization means for binarizing the image of the row region using the binarization threshold value; Labeling means for performing labeling; When a label of a character in contact with the background image is detected based on the size of each label obtained by the labeling means, only characters from the character label are detected based on the peripheral characteristics of the label and a predetermined character size and character spacing. First character extracting means for extracting, and second character extracting means for extracting a character based on a predetermined character size and character interval when a label of a character is detected based on the size of the label obtained by the labeling means. And a character reading means for recognizing characters cut out by the first and second character cutting means and reading character information of the input image. 6. The apparatus according to claim 1, wherein said high-frequency component removing means removes a high-frequency component from said minimum projection information by a parameter determined based on a predetermined character height. The character reading device according to any one of the above. 7. A step of scanning a partial image of the input image in a predetermined character string direction to obtain minimum projection information; a step of removing a high-frequency component from the minimum projection information; Extracting a line area based on the minimum projection information from which the line area has been removed, binarizing the image of the extracted line area, cutting out characters from the binarized image, and cutting out the characters Recognizing the input image and reading the character information of the input image. 8. A step of scanning a partial image of the input image in a predetermined character string direction to obtain minimum projection information; a step of removing a high-frequency component from the minimum projection information; Differentiating the minimum projection information from which the difference has been removed, and extracting a region in which the projection value does not change between a position where the differential value is negative and a position where the differential value is positive as a line region; and Binarizing the image of the region, cutting out characters from the binarized image, and reading character information of the input image by recognizing the cut out characters. Character reading method. 9. A step of scanning a partial image of the input image in a predetermined character string direction to obtain minimum projection information; a step of removing a high-frequency component from the minimum projection information; Extracting a row region based on the minimum projection information from which the line region has been removed; obtaining the luminance distribution of the image of the extracted line region; and determining, in advance, a distribution region corresponding to background luminance among the luminance distribution. Determining a binarization threshold based on a luminance distribution obtained by deleting according to the obtained criterion; binarizing the image of the row region using the binarization threshold; A character reading method, comprising: a step of cutting out characters from a character string; and a step of recognizing the cut out character and reading character information of the input image. 10. A step of scanning a partial image of an input image in a predetermined character string direction to obtain minimum projection information; a step of removing a high frequency component from the minimum projection information; Extracting a line region based on the minimum projection information from which the line region has been removed, binarizing the image of the extracted line region, labeling the binarized image, and labeling the binarized image. A first character extracting step of extracting only a character from the character label based on the peripheral characteristics of the label and a predetermined character size when detecting a label of a character in contact with the background image based on the size of each label thus obtained. And a second sentence that cuts out characters based on a predetermined character size and character spacing from the label obtained by the labeling. A character reading method, comprising: a character extracting step; and a step of recognizing the character extracted in the first and second character extracting steps and reading character information of the input image. 11. A step of scanning a partial image of the input image in a predetermined character string direction to obtain minimum projection information; a step of removing a high frequency component from the minimum projection information; Differentiating the minimum projection information removed, and extracting a region in which the projection value does not change between a position where the differential value becomes a positive maximum value and a position where the differential value becomes a negative maximum value as a row region; and Obtaining a luminance distribution of the image of the extracted row region; and, based on a luminance distribution obtained by deleting a distribution region corresponding to the luminance of the background from the luminance distribution according to a predetermined reference. Determining a binarization threshold; binarizing the image of the row region using the binarization threshold; performing labeling on the binarized image; A first character extracting step of extracting only a character from the character label based on the peripheral characteristics of the label and a predetermined character size when detecting a label of a character in contact with the background image based on the size of each label thus obtained. A second character extracting step of extracting a character based on a predetermined character size and a character interval when a character label is detected based on the label size obtained by the labeling; and the first and second character extracting steps. Recognizing the character extracted in the character extracting step and reading the character information of the input image. 12. The character according to claim 7, wherein a high-frequency component is removed from the minimum projection information within a range determined based on a predetermined character height. Reading method.