JP2004334461A - Character recognition device and character recognition program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a technique that recognizes characters in an input image displaying characters differing in size. <P>SOLUTION: An input image is divided by an area section part 1 into subareas of a first division mode and divided also by an area section part 2 into subareas of a second division mode differing in size from the subareas of the first division mode. A projection section extraction part 3 next extracts black sections from both subareas. A character string area extraction part 4 then merges the black sections to form character string areas suitable to the respective division modes, a character pickup part 5 executes picking in character units, and a character recognition part 6 executes character recognition. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a character recognizing device and a character recognizing program for recognizing characters in an image. In particular, when a character string composed of characters of different sizes exists in an image, a character string area is formed from the image. It relates to a technology for efficient extraction.
[0002]
[Prior art]
A conventional character recognition device that recognizes characters in an image estimates a character pattern size, cuts out an area (character string area) in which a character string exists, and extracts a pixel pattern and a character existing in the character string area. It was to match the pattern. In such a character recognition device, a method of estimating the size of a character pattern serving as a basis for the size of a character string region to be cut out is important.
[0003]
As a method of estimating the size of such a character pattern, there is a technique of acquiring a distribution state of pixels constituting a character in a partial region at a specific position in an image and estimating a character size from the distribution state. (For example, Patent Document 1).
[0004]
[Patent Document 1]
JP-A-63-292381, "Character line detection device" (FIG. 1, page 3-5)
[0005]
[Problems to be solved by the invention]
As described above, the conventional character recognition device estimates the character size and line spacing from a partial area of the input image. Therefore, even if information of a reference character is obtained from such an area, it cannot be correctly recognized if a character having a size different from that of the character exists in another area. there were.
[0006]
The present invention has been made to solve such a problem, and a character recognition device that appropriately detects and recognizes a character string even when a plurality of character strings having different sizes exist. The purpose is to provide.
[0007]
[Means for Solving the Problems]
A character recognition device according to the present invention divides an input image into regions in a first division mode suitable for characters of a predetermined size and a region in a second division mode suitable for characters having a size different from the size. Black section extracting means for further dividing, as a black section, a region having a predetermined number of black pixels or more from each region of the first division mode and each region of the second division mode,
A character area extracting unit that merges the black sections to form a character string area that conforms to those division forms, and cuts out a character area from the character string area.
And character recognition means for recognizing a character pattern in the character area.
[0008]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments according to the present invention will be described.
Embodiment 1 FIG.
FIG. 1 is a block diagram showing a configuration of a character recognition device according to Embodiment 1 of the present invention. In the figure, a region partitioning unit 1 and a region partitioning unit 2 divide an input image captured by an image input unit such as a camera (not shown) into a plurality of regions. The size of the area divided by the area dividing section 1 is different from the size of the area divided by the area dividing section 2. The projection section extraction unit 3 is configured to extract an area (black section) where a black pixel exists, based on the pixel value of the area in the input image. The character string area extraction unit 4 combines the black sections extracted by the projection section extraction unit 3 to form a character string area candidate that is an area where a character string may exist, and furthermore, the character string area candidate This is a part for determining whether or not the area is a character string area based on the size. The character cutout unit 5 cuts out each character region from the character string region formed by the character string region extraction unit 4. The character recognizing unit 6 is a part for recognizing a region of each character cut out by the character cutting unit 5. Here, the area section 1 and the area section 2 and the projection section extracting section 3 constitute black section extracting means, and the character string area extracting section 4 and the character extracting section 5 constitute character area extracting means. It is. Further, the character recognition unit 6 corresponds to a character recognition unit.
[0009]
Next, the operation of the character recognition device will be described. FIG. 2 is a flowchart showing the operation of the character recognition device. As a premise of this processing, it is assumed that an input image 7 as shown in FIG. 3 has been captured. The input image 7 is a binary image composed of white and black pixels. As shown in the figure, the input image 7 has a pattern 8 which is not a character string and three horizontal writings composed of character strings 9 to 11. String exists.
[0010]
First, the region partitioning unit 1 and the region partitioning unit 2 divide the entire input image 7 into partial regions (step S1). The region partitioning unit 1 divides the input image 7 into partial regions having equal areas. A state in which the input image 7 is divided by the region partitioning unit 1 is referred to as a first division mode. Further, the area partitioning section 2 divides the same input image 7 into partial areas having different areas from the partial areas according to the first division mode and having equal areas to each other. Such a division state by the area partitioning unit 2 is referred to as a second division mode.
[0011]
FIG. 4 is a diagram showing an example of a partial area setting (first division mode) by the area division unit 1, and FIG. 5 is a diagram showing an example of a partial area setting (second division mode) by the area division unit 2. It is. Compared to the second division mode, the first division mode is divided into narrower partial regions.
[0012]
Generally, in order to recognize a character in an image, it is basically necessary to divide an image into a plurality of partial areas and obtain a distribution of pixels for each partial area. In order to perform accurate character recognition, it is necessary to appropriately set the partial area. By the way, in many cases, an object other than a character string and its shadow are captured in the image, or the character string rotates (the character string is captured diagonally with respect to the horizontal or vertical coordinate axes of the image). As a result, noise (pixels that do not constitute a character) is mixed. Therefore, by setting a large partial area as much as possible, the influence of the pixel distribution due to such noise can be relatively reduced. However, when small characters are present in the image, even such small characters may be eliminated as noise.
[0013]
Therefore, the character recognition device according to the first embodiment also performs area division for the purpose of recognizing small characters. This is because the area division for small characters enables character recognition even for a pixel distribution that is excluded as noise in the area division for large characters.
[0014]
The first division mode shown in FIG. 4 indicates a state where a relatively small character is divided into partial regions for the purpose of recognition. In addition, the second division mode shown in FIG. 5 indicates a state where the image is divided into partial areas for the purpose of recognizing a character larger than the character corresponding to the first division mode.
[0015]
Regarding the direction in which the image is divided, when recognizing a horizontally written character string, it is better to divide the input image into vertically long strips. This is because noise is rejected in units of the width of the partial area. On the other hand, in order to recognize a vertically written character string, the input image may be divided into horizontally long strips. If it is not possible to determine whether the image is horizontally or vertically written, the input image is divided into a partial area close to a square. FIGS. 4 and 5 show the input image divided into vertically long strip-shaped partial areas.
[0016]
In addition, the partial areas of the first and second divided forms are determined based on the assumed size of characters in the input image for simplicity of description. This is because, if the image is divided into partial areas having extremely different character sizes and sizes, noise cannot be rejected correctly or a part of the character is detected as being missing. Unlike this example, if the size of the character in the input image cannot be predicted, it is necessary to prepare a division form corresponding to the character size in several stages and an area partition corresponding to the division form. Good. Therefore, naturally, the image may be divided into three or more division forms.
[0017]
Next, the projection section extraction unit 3 calculates a projection value by projecting, for both the first division mode and the second division mode, for each column of pixels in a direction in which the input image is divided ( Step S2). The projection value refers to a sum of pixel values on a row of pixels in a certain direction (horizontal direction or vertical direction) in a certain area. In this example, since the input image is divided in the horizontal direction (so that the partial area forms a vertically elongated strip), the sum of the pixel values of each partial area is calculated for each column of pixels in the horizontal direction.
[0018]
Subsequently, a black section is extracted based on the projection value calculated in step S2 (step S3). Specifically, each projection value is compared with a predetermined threshold value, and is binarized to 1 when the projection value is equal to or more than the predetermined threshold value and to 0 when the projection value is less than the predetermined threshold value. Next, as a binarized projection value, an area where 1s are continuous is a black section, and an area where 0s are continuous is a white section. As a result, FIG. 6 shows an example of a black section extracted from the first division mode. In the narrow black section 14 of the first division mode, the area of the large character string 9 is divided. FIG. 7 shows an example of a black section extracted from the second division mode. In the wide black division 15 of the second division mode, the regions of the character strings 10 and 11 which are small and close to each other are one black division. In addition, the area of the pattern 8 of the input image 7 in FIG. 3 is divided in the black section 14 of the first division mode, but is entirely one section in the black section 15 of the second division mode. The height (height) is almost the same as the character strings 9 and 10.
[0019]
Next, the character string area extracting unit 4 extracts a black section constituting the character string area from each of the black sections of the first division mode and the second division mode extracted by the projection section extraction unit 3 (step S4). ). That is, the following processing is performed. First, for the first division mode, the size of a character (referred to as the first character size) stored in a storage device (not shown) is obtained. Here, the standard height of the character is acquired as the size of the first character. Next, the height of each black section in the first division mode is compared with the standard height of the character acquired as the size of the first character. In this case, for example, the allowable minimum magnification is set to 90% and the allowable maximum magnification is set to 110% as an allowable range, and the height of the black section is 90% or more of the standard height of the character and within 110%. Then, the black section is adopted as a black section constituting the character string area. In the second division mode, the size of the second character is similarly obtained and compared. Since the second division mode is larger than the first division mode, the size of the second character is also set to be larger than the size of the first character.
[0020]
As a result, the black section corresponding to the pattern 8 existing at the lower left of FIGS. 6 and 7 is not selected. The reason is that in the first division mode, these black sections are smaller than the allowable range of the size of the first character, and in the second division mode, these black sections are the first section. , But is outside the allowable range for the size of the second character. In this way, even in the case of an input image in which characters of different sizes are mixed, noise caused by pixels that do not constitute characters is removed, and erroneous detection is prevented.
[0021]
In the above example, in order to detect a horizontally written character string, the height of the black section existing in the vertically divided area was compared with the standard height of the character. On the other hand, when recognizing a vertically written character string, the area is divided in the horizontal direction. In this case, the width of the black section existing in each area may be compared with the standard width of the character. When an input image in which a vertical writing character string and a horizontal writing character string are mixed is divided into partial areas in a square shape in order to be subjected to character recognition, both the height and the width may be compared.
[0022]
As a result, the character string area extraction unit 4 sets black sections falling within the permissible range as character string area candidates (step S5). After that, the character string area extracting unit 4 combines the character string area candidates to form a character string area (step S6). That is, character string region candidates existing in adjacent partial regions and having a difference in vertical coordinate between each other less than or equal to a predetermined threshold are regarded as one character string region. On the other hand, when there is no character string area candidate whose upper end and lower end are close to an adjacent partial area, the character string area candidate is not formed as a character string area. FIG. 10 is an example of a character string area formed from black sections in the first division mode. Character string areas 21 to 22 corresponding to the character strings 10 to 11 in FIG. 3 are formed. FIG. 11 is an example of a character string area formed from black sections in the second division mode, and a character string area 24 corresponding to the character string 9 in FIG. 3 is obtained.
[0023]
The character cutout unit 5 determines a character cutout target area for each of the character string areas extracted by the character string area extraction unit 5, and performs character cutout in the same procedure as in the related art (step S7). The character cutout target area is, for example, a range in which the upper, lower, left, and right sides of the character string area are extended by a predetermined value so that a character pattern protruding from the area does not occur. An example set for the character string area 24 in FIG. 11 is the character extraction target area 25 in FIG. Thereafter, the character recognition unit 6 performs character recognition in the same procedure as in the related art (step S8).
[0024]
In addition, from the state of the image distribution, it is conceivable that the character string region of the second division mode for processing large characters and the character string region of the first division mode for processing small characters overlap. For example, the kanji character "chi" is composed of a partial "arrow" and a side "mouth", but it is also possible to treat only the partial and the side as a single kanji. In such a case, “arrow” and “mouth” are detected from the processing result of the first division mode, and “knowledge” is detected from the processing result of the second division mode. The processing results will be inconsistent.
[0025]
Therefore, in such a case, the calculation result of the second division mode for processing a large character is given priority. As a result, the processing results of the plurality of division modes are integrated. Note that such integration processing is performed by either the character cutout unit 5 or the character recognition unit 6.
[0026]
As is clear from the above, according to the character recognition device of the first embodiment, the size of the character string to be extracted is determined in association with the size of the partial area, and this size is compared with the size of the extracted black section. Then, the character string area is extracted and further divided into partial areas of different sizes, and such processing is performed for each partial area of each size, so that characters of different sizes are included. Even for an input image, erroneous recognition can be prevented and proper character recognition can be performed.
[0027]
Further, since the entire input screen is divided into partial areas, the character string can be appropriately detected and recognized regardless of the display position and size of the character string.
[0028]
In the above processing, the size of the black section of each divided form is compared with the standard size of the character, and only the suitable black section is selected, and then the character string area is formed from the selected black section. . However, besides this method, a method is also conceivable in which first, adjacent black sections are merged to form a character string area candidate, and then whether or not the character string area candidate is a character string area. In this case, as described above, when the character string is written horizontally, a method of selecting a character string area candidate based on the standard height of the character (when the character string is written vertically, the width of the character, In addition to the two methods for mixing writing and horizontal writing, the following character string region candidate selection method may be adopted.
[0029]
That is, when the width of the character string area candidate (when the character string is written horizontally) is divided by the standard width of the character corresponding to this division mode, and the result of the division becomes a value close to an integer value (discrete value), The character string area candidate is determined to be a character string area. If the string is vertical, divide by the standard height of the character. If the characters are mixed, one of the standard height and the standard width of one of the characters is selected and divided. Further, a standard area obtained by multiplying the standard width by the standard height may be used as a reference.
[0030]
In addition, in the present embodiment, in order to detect and recognize black characters, a black section is obtained from the projection value of the input image. However, white characters can be detected and recognized by first inverting the black and white of the input image. .
[0031]
Further, it is naturally possible to realize as a computer program that causes a computer to execute the character recognition function performed by the character recognition device according to the first embodiment. In this case, the functions corresponding to the functions of the respective sections of the area section 1, the area section 2, the projection section extracting section 3, the character string area extracting section 4, the character extracting section 5, and the character recognizing section 6 are executed. What is necessary is just to make it a computer program which executes a computer program sequentially.
[0032]
Embodiment 2 FIG.
In the first embodiment, the entire input image is divided into partial regions by a plurality of division modes. On the other hand, in step S1 of the flowchart in FIG. 2, one input image may be divided by combining a plurality of division forms.
[0033]
For example, as shown in FIG. 13, a camera 26a that captures an input image is installed at the upper end of a column 26b, and when reading a number printed on a license plate 27 or 28 of an automobile or the like, when a camera reads a number printed on a license plate 27 or 28, Is as shown in FIG. In FIG. 14, a character string 30 on the license plate 28 is displayed relatively small above the input image 29. Further, a character string 31 on the license plate 27 is displayed relatively large below the input image 29. As described above, the size of the characters on the license plates 27 and 28 is originally substantially the same size, but the character string 30 on the license plate 27 at a position far from the camera 26a is small above the input image 29, and The character string 31 on the license plate 28 at a position close to the camera 26a is displayed large below the input image 29.
[0034]
In such a case, if, for example, the upper half of the input image 29 is divided into regions by the first division form and the lower half is divided into regions by the second division form, the character string is similar to the first embodiment. Can be divided into character string regions suitable for the size of the characters constituting.
[0035]
As is clear from the above, according to the character recognition device according to the second embodiment, the size of the character string in the input image is determined in advance from the relationship between the imaging position of the camera and the position of the object on which the character string is displayed. When prediction is possible, detection and recognition can be performed by optimally combining the division forms of the input image.
[0036]
Also, the same input image is partially divided by combining a plurality of division modes, as compared with the method of performing the process of dividing the entire input image into partial regions in a single division mode multiple times as in the first embodiment. As a result, it is not necessary to process a large number of pixels, so that the performance is improved and the computer resources can be saved.
[0037]
Needless to say, the method of dividing the input image may be changed according to the positional relationship between the camera position and the object displaying the character string. For example, when capturing a license plate installed in the right direction from a camera installed on the left side, a target area of a large divided form of the partial area is set on the left side of the input image, May be set to the right of the input image.
[0038]
Embodiment 3 FIG.
Next, a character recognition device according to a third embodiment of the present invention will be described. Although the character recognition device according to the first embodiment uses an input image as a binary image composed of white and black pixels, the character recognition device according to the third embodiment uses a multivalued image or a multi-tone image as an input image. Different.
[0039]
FIG. 15 is a block diagram illustrating a configuration of a character recognition device according to the third embodiment. In the figure, a differential image extraction unit 101 is a part that creates a differential image from a multi-layer image. The other components that are denoted by the same reference numerals as those of the character recognition device according to the first embodiment are the same as those in the first embodiment, and thus description thereof is omitted.
[0040]
Next, the operation of the character recognition device according to the third embodiment will be described. FIG. 2 is used for the flowchart showing the processing of this character recognition device as in the first embodiment. First, as in the first embodiment, a multi-level image is captured and captured by an image input unit (not shown). For example, it is assumed that this image is a grayscale image of 8 bits per pixel. FIG. 16 is a diagram showing an example of such an input image. In the input image 36, a white character string 37 and a black character string 38 are mixed. Indicates a column 39 and a column 40.
[0041]
First, as in the first embodiment, in step S1, the area partitioning sections 1 and 2 are divided into areas according to the first division mode and the second division mode (step S1). Subsequently, a projection value of the input image 36 is calculated (step S2). Unlike the character recognition device according to the first embodiment, the input image of the character recognition device is a multi-layer image. Therefore, in this step, first, the differential image extracting unit 101 creates a differential image, and then binarizes the differential image to calculate a projection value. Differentiation of an image is described in, for example, “Introduction to Computer Image Processing” published by Soken Publishing, pp. Although various methods described in 119 to 122 can be used, here, the method by the Sobel operator is used. For the binarization of the differential value, for example, a method of applying a fixed threshold can be used. As a result, as shown in FIG. 17, an image is obtained in which only the outline of the displayed object and character remains. By calculating the projection value from the differential binary image, a black section is extracted not only from a black character but also from a white character string.
[0042]
Note that, in FIG. 17, the boundaries of the divisions divided by the region divisions 1 and 2 are omitted for easy viewing. Alternatively, the processing order of step S1 and step S2 may be reversed, and the area division may be performed after obtaining the differential binary image first.
[0043]
Next, in the same procedure as in the first embodiment, a black section is extracted for each of the first and second division modes based on the projection values calculated in step S2 (step S3). In the case of a differential binary image, since the distribution of black pixels is small in an area with a small change in density, many black sections are detected in a contour portion of a character or an object with a large change in density. FIG. 18 is a diagram illustrating an example of a black section extracted from the first division mode. In the black section 44 of the narrow first division mode, the area of the large character string 37 is divided. FIG. 19 is a diagram showing an example of a black section extracted from the first division mode. In the wide black section 45 of the second division mode, the area of the character string 38 is adjacent to the adjacent columns 39 and 40. Is extremely large black section.
[0044]
Next, character string area candidates that can constitute a character string area are selected from the extracted black sections (steps S4 and S5), and then the character string area candidates are merged to form a character string area (step S6). ). These processes are the same as those in the first embodiment, and a description thereof will not be repeated.
[0045]
Subsequently, the character string cutout unit 5 cuts out a character pattern as in the first embodiment (step S7). However, since the input image in the third embodiment is a multi-layer image, it is first determined whether the characters included in each character string area are black characters or white characters. For this purpose, a determination target region is set based on the position of the character string region, the determination target region in the input image is binarized and projected in the character string direction, and a determination is made based on the projection value.
[0046]
Note that the term “character string direction” means a direction in which characters constituting a character string are arranged (vertical or vertical / horizontal or horizontal), and “projects in the character string direction”. For example, in the case of a horizontally written character string, projection data is calculated for each pixel row in the horizontal direction. Therefore, for example, when a horizontally written character is displayed in a character string area having 20 (vertical) × 128 (horizontal) pixels, there are 20 horizontal pixel rows of 128 pixels having the same vertical coordinates. Will do. In such a case, performing projection in the character string direction means calculating projection data for each of the 20 horizontal pixel rows.
[0047]
The determination area is an area including a character string area. For example, a range in which the character string area is extended by a predetermined amount in a direction perpendicular to the character string direction so that the end of the character is included in the determination area. Area. FIG. 20 is a diagram conceptually showing such a determination area, and shows a determination area 48 for the white character string 37 and a determination area 49 for the black character string 38.
[0048]
The process of determining whether a character displayed in this region is a black character or a white character (black character / white character determination) will be described below taking the determination region 48 as an example. FIG. 21 shows the distribution of the projection values calculated for the determination area 48. In the figure, a projection value distribution 50 indicates the projection values in the horizontal direction of the entire area. 51 is the center position of the character string area 37, and 52 and 53 are the lower and upper limits of the preset projection value, respectively. The lower limit value 52 and the upper limit value 53 are, for example, values obtained by multiplying the length of the determination target area in the character string direction by a predetermined ratio.
[0049]
In this case, starting from the projection value at the center position 51 of the character string area, then upward (to the left in the case of a vertically written character string) and downward (to the right in the case of a vertically written character string) Then, the projection values are sequentially acquired one pixel at a time, and it is checked whether each projection value does not fall below the lower limit value or further exceeds the upper limit value. As a result, for the first time, the position of a pixel below the lower limit or above the upper limit is regarded as the end of the character string. When the value falls below the lower limit first, this character string is regarded as a black character string, and when the value exceeds the upper limit first, this character string is regarded as a white character string.
[0050]
In the example of FIG. 21, the character string exceeds the upper limit value 53 before falling below the lower limit value 52 in both the up and down directions, so that this character string is determined to be a white character string. On the other hand, in the example shown in FIG. 22, when the projection value 54 is referred to from the center of the character string area, it falls below the lower limit 55 before exceeding the upper limit 56, and is determined to be a black character string.
[0051]
After that, the character cutout unit 5 determines a character cutout target area in the same procedure as in the first embodiment, and then binarizes the input image of the area to create a binary image for character cutout. Further, as a result of the black character / white character determination, if the determination result is a white character, an image obtained by inverting the binary image in black and white is used for character extraction. Subsequent processing is the same as in the first embodiment, and a description thereof will not be repeated.
[0052]
As is apparent from the above description, according to the character recognition device of the third embodiment, a multi-layer image is converted into a differential binary image, and then a black section is extracted and compared with the size of the character string. Since the row region is selected, individual characters can be correctly extracted and recognized even from an image in which white and black character strings are mixed without greatly increasing the processing amount.
[0053]
In the above description, after performing the area division, the differential binarization is performed for each of the division forms. However, the differential binarization may be performed before step S3 for extracting the black section. For example, the input image may be binarized differently, and region division may be performed on the image after the binarization.
[0054]
Embodiment 4 FIG.
Next, a character recognition device according to a fourth embodiment will be described. The character recognition device according to the fourth embodiment is characterized by a method of forming a partial area in the second division mode. Also, the method of handling multi-tone images is different from that of the character recognition device according to the third embodiment, and the input image contains a rotated character string.
[0055]
FIG. 15 is used as a block diagram showing the configuration of the character recognition device according to the fourth embodiment. However, in the character recognition device according to the fourth embodiment, the area section 2, the projection section extraction section 3, and the character string area extraction section 4 are different from the third embodiment. The area dividing section 2 is configured to form a partial area according to the second division mode by merging the partial areas according to the first division mode divided by the area dividing section 1. The projection section extraction unit 3 combines the black sections extracted from the partial areas of the first division mode to form black sections of the partial areas of the second division mode. The character string area extraction unit 4 forms a character string area by eliminating the influence of the shift between black sections caused by the rotation of the character string. The other components are the same as those in the third embodiment, and a description thereof will not be repeated.
[0056]
FIG. 23 is an example of an input image on which the character recognition device of the fourth embodiment performs character recognition. In the input image 58 shown in the figure, a character string 59 is displayed in white on a black background. Reference numeral 62 denotes a character or an elliptical figure which is not a character string, and character strings 60 and 61 are character strings in which rotation occurs.
[0057]
Next, the operation of the character recognition device according to the fourth embodiment will be described. The processing in the character recognition device according to the fourth embodiment is shown by the flowchart in FIG. 2 as in the first to third embodiments. First, the area division unit 1 divides the input image 58 into partial areas according to the first division mode, and then the area division unit 2 forms a partial area according to the second division mode based on these partial areas (step S1). S1). That is, first, the area partitioning unit 1 divides the input image 58 into small partial areas. The partial area divided by the area dividing unit 1 is stored in a memory as a first division mode. Next, the area partitioning unit 2 forms a large partial area by merging two adjacent partial areas among the small partial areas. In the first and third embodiments, the area partitioning sections 1 and 2 independently divide the input image into partial areas, but in the fourth embodiment, the first division mode is used to divide the input image into the partial areas. Is different in that
[0058]
In this description, for the sake of simplicity, of the partial regions according to the first division mode, adjacent partial regions are merged two by two to form a partial region according to the second division mode. The method of forming the partial area by the two division forms is not limited to this. For example, a method of merging three adjacent partial regions may be adopted, or a method of merging three adjacent partial regions and then forming a partial region by a method of dividing into two equal parts.
[0059]
Next, a projection value is calculated from the partial area according to the first division mode and the partial area according to the second division mode, and a black section is extracted (steps S2 and S3). Since the input image in the fourth embodiment is a multi-tone image, the differential binarization is performed as in the third embodiment, and then the black section is extracted. However, the fourth embodiment differs from the third embodiment in the following points. That is, after performing differential binarization on the partial area according to the first division mode and further extracting a black section, the black sections are merged to form a black section according to the second division mode. is there.
[0060]
Specifically, the following processing is performed. First, as in the third embodiment, calculation of a differential binary image and a projection value and extraction of a black section are performed for the first division mode. FIG. 24 is an example of the differential binary image obtained here. However, the display of the boundary line between the partial areas is omitted. FIG. 25 is an example of a black section obtained based on the first division mode.
[0061]
Further, if black sections are adjacent to each other in the first divided form that has been merged to form the partial areas in the second divided form, the black sections are merged. This merging process is performed, for example, by any of the following methods.
[0062]
(1) Calculate the minimum rectangle surrounding the black section in the adjacent first division mode, take the ratio of the sum of the areas of the black sections surrounded by this rectangle to the area of this rectangle, and set this ratio to a predetermined value. If the value is equal to or greater than the value, the entirety of the minimum rectangle is set as a black section in the second division mode.
(2) When the length of the boundary line between adjacent black sections in the first division mode is equal to or greater than a predetermined value, the entirety of the smallest rectangle surrounding these black sections is defined as the black section in the second division mode.
[0063]
FIG. 26 is an explanatory diagram showing such a black section merging process. The figure shows how the black sections 111 and 112 according to the first division mode are merged into the black section 113 according to the second division mode. Further, the black section 116 according to the first division mode is adjacent to both the black sections 114 and 115 belonging to the same partial area. In such a case, the smallest rectangle surrounding all of the black sections 114, 115, and 116 becomes one black section 117.
[0064]
On the other hand, the black sections 118 and 119 according to the first division mode are also adjacent to each other. In such a case, the black sections according to the second division mode are formed by any of the methods (1) and (2). Is not formed. As described above, in order to absorb the rotation of the character string, if the deviation of the black section is allowed, the deviation of the black section other than the deviation of the black section caused by the rotation of the character string may be included. However, if adjacent black sections are merged based on the criteria of (1) and (2), such a case can be eliminated.
[0065]
Thus, by combining the black sections of the first division mode to form the black sections of the second division mode, the differential binary image and the projection value are calculated with respect to the second division mode, and the black section is calculated. Since it is not necessary to perform the process of performing the extraction, the process can be performed at high speed.
[0066]
Next, the character string area extracting unit 4 forms a character string area by merging the black sections (step S4). In the fourth embodiment, instead of selecting a black section based on the standard size of a character, a black section is selected based on the pixel density of an input image in each black section area. For example, both black characters and white characters are allowed for the black section in the first division mode, and only white characters are allowed for the partial area in the second division mode.
[0067]
The process of selecting a black section is performed as follows. That is, first, the average value of the maximum pixel value and the minimum pixel value of the input image in the black section area is calculated as the binarization threshold. Next, the number of pixels having a value smaller than the binarization threshold value is compared with the number of pixels having a value equal to or greater than the binarization threshold value. If the latter is large, it is determined to be a white character. If the determination result of the black character / white character matches the character color (black / white) defined in the division mode of the black partition, the black partition is selected. If they do not match, the black section is rejected. The formation of the character string area (step S5) is performed by merging only the selected black sections.
[0068]
In a general character string image area, since there are more background pixels than character pixels, according to the above-described method, black and white characters can be determined by appropriately setting the binarization threshold. In the method described in the third embodiment, since the projection in the character string direction is used, it may not be possible to correctly determine that the rotation angle of the character string is very large. However, in this method, since the grayscale distribution is used, the rotation is not performed. It can be determined without any restrictions on the angle.
[0069]
The processing after step S5 is the same as in the third embodiment, and a description thereof will not be repeated.
[0070]
As is clear from the above, according to the character recognition device of the fourth embodiment, since the second division mode is obtained based on the first division mode, the amount of calculation can be significantly reduced. In addition, since the black sections are merged based on the above criteria (1) and (2), character recognition that is strong against rotation of a character string can be performed.
[0071]
【The invention's effect】
A character recognition device according to the present invention divides an input image into regions in a first divisional form, and also divides the input image into regions in a second divisional form having a size different from the region in the first divisional form. Since a character string area suitable for each of the divided forms is formed from the black sections extracted from the areas of both divided forms, even if a plurality of character strings of different sizes exist, the This has an extremely excellent effect that a character string can be detected and recognized.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of a character recognition device according to Embodiment 1 of the present invention.
FIG. 2 is a flowchart of the character recognition device according to the first embodiment of the present invention.
FIG. 3 is a diagram illustrating an example of an input image according to the first embodiment of the present invention;
FIG. 4 is a diagram showing a first division mode according to the first embodiment of the present invention.
FIG. 5 is a diagram showing a second division mode according to the first embodiment of the present invention.
FIG. 6 is a diagram showing an example of a black section extracted from the first division mode according to the first embodiment of the present invention.
FIG. 7 is a diagram illustrating an example of a black section extracted from the second division mode according to the first embodiment of the present invention.
FIG. 8 is a diagram showing an example of a black section constituting a character string area extracted from the first division mode according to the first embodiment of the present invention.
FIG. 9 is a diagram showing an example of a black section constituting a character string area extracted from the second division mode according to the first embodiment of the present invention.
FIG. 10 is a diagram showing an example of a character string region candidate extracted from the first division mode according to the first embodiment of the present invention.
FIG. 11 is a diagram showing an example of a character string region candidate extracted from the second division mode according to the first embodiment of the present invention.
FIG. 12 is a diagram showing an example of a character string cutout area according to the first embodiment of the present invention.
FIG. 13 is an explanatory diagram illustrating a positional relationship between a camera that captures an input image and a license plate according to the second embodiment of the present invention.
FIG. 14 is a diagram illustrating an example of an input image according to the second embodiment of the present invention;
FIG. 15 is a block diagram showing a configuration of a character recognition device according to Embodiment 3 of the present invention.
FIG. 16 is a diagram showing an example of an input image according to the third embodiment of the present invention.
FIG. 17 is a diagram showing an example of a differential binary image according to the third embodiment of the present invention.
FIG. 18 is a diagram illustrating an example of a black section extracted from the first division mode according to the third embodiment of the present invention.
FIG. 19 is a diagram showing an example of a black section extracted from the second division mode according to the third embodiment of the present invention.
FIG. 20 is an explanatory diagram of a determination area according to the third embodiment of the present invention.
FIG. 21 is a diagram showing a distribution of projection values of a determination area for white characters according to Embodiment 3 of the present invention.
FIG. 22 is a diagram showing a distribution of projection values of a determination area for black characters according to Embodiment 3 of the present invention.
FIG. 23 is a diagram illustrating an example of an input image according to Embodiment 4 of the present invention;
FIG. 24 is a diagram showing an example of a differential binary image according to the fourth embodiment of the present invention.
FIG. 25 is a diagram illustrating an example of a black section extracted from the first division mode according to the fourth embodiment of the present invention.
FIG. 26 is an explanatory diagram showing a black partition merging process according to the fourth embodiment of the present invention;
[Explanation of symbols]
1, 2 area division
3 Projection section extraction unit
4 Character string area extraction unit
5 Character cutout
6 Character recognition unit
101 Differential image extraction unit

Claims (16)

The input image is divided into regions in a first division mode suitable for characters of a predetermined size, and further divided into regions in a second division mode suitable for characters having a size different from the size. Black section extraction means for extracting, as a black section, a region having a predetermined number of black pixels or more from each region of the form and each region of the second division form;
A character area extracting unit that merges the black sections to form a character string area that conforms to those division forms, and cuts out a character area from the character string area.
A character recognition unit that recognizes a character pattern in the character area. The black section extracting means may divide a predetermined portion of the input image into regions according to the first division mode, and divide a portion of the input image excluding the predetermined portion into regions according to the second division mode. The character recognition device according to claim 1, wherein 2. The character recognition device according to claim 1, wherein the black section extraction unit merges a plurality of adjacent areas of the first division mode to form an area of the second division mode. 3. The black section extraction unit extracts the black section from the area of the first division mode, and then merges the plurality of areas of the first division form and the black section adjacent to each other to form the second division mode. 2. The character recognition device according to claim 1, wherein each of the regions and a black section thereof are formed. 2. The black section extracting unit according to claim 1, wherein, instead of the input image, an image obtained by differentiating and binarizing the input image is divided into an area of a first division mode and an area of a second division mode. The character recognition device according to any one of claims 1 to 4. The black section extracting means may differentiate the area of the first division mode and the area of the second division mode into differential binarization, and extract the black section from the differential binarized area. Item 5. The character recognition device according to any one of Items 1 to 4. When the size of the black section is within a predetermined range as the size of a character corresponding to the division mode of the black section, the character area extracting unit merges the black sections to form the character string area. The character recognition device according to claim 1, wherein the character recognition device is formed. The character area extracting means forms the character string area by merging the black sections when a length of a boundary line between the plurality of adjacent black sections is equal to or greater than a predetermined value. Item 7. The character recognition device according to any one of Items 1 to 6. The character area extracting means merges the black sections and forms the character string area when a ratio of a total area of the plurality of black sections adjacent to each other and an area of a rectangle surrounding the black section is a predetermined value or more. The character recognition device according to any one of claims 1 to 6, wherein 2. The character area extracting unit according to claim 1, wherein the character string area candidate is selected as the character string area when a character string area candidate formed by merging the black sections satisfies a predetermined condition. 7. The character recognition device according to any one of claims 6 to 6. The character area extracting unit may be configured such that the size of the character string area candidate formed by merging the black sections is close to a value obtained by multiplying the size of the character corresponding to the division mode of the black section by a natural number. The character recognition device according to claim 10, wherein the character string area candidate is selected as the character string area. The character area extracting unit may be configured to, when a pixel density of a character string area candidate formed by merging the black sections is substantially the same as a pixel density corresponding to the division mode of the black section, determine the character string area candidate. The character recognition apparatus according to claim 10, wherein the character string area is selected. When the pixel density of the character string area is equal to or less than a predetermined value, the character area extracting means inverts the pixels of the character string area to generate an inverted character string area, and extracts the character area from the inverted character string area. The character recognition device according to claim 1, wherein the character is cut out. The character area extracting means calculates projection data in the character string direction for the binary image of the character string area, and calculates the projection data for each pixel string in the character string direction from the center position of the character string area. The data is sequentially acquired toward the end of the character string area and compared with a predetermined upper limit value and a lower limit value. When the projection data becomes equal to or more than a predetermined upper limit value before becoming equal to or less than a predetermined lower limit value, the inversion is performed. 13. The character recognition device according to claim 12, wherein a character string area is generated. The character area extracting means calculates an average value of the maximum density and the minimum density of the pixels in the character string area, and calculates the number of pixels having a density lower than the average value and the pixels having a density equal to or higher than the average value in the character string area. 13. The character recognition device according to claim 12, wherein whether or not to generate the inverted character string area is determined based on a magnitude relationship with a number. The input image is divided into regions in a first division mode suitable for characters of a predetermined size, and further divided into regions in a second division mode suitable for characters having a size different from the size. A black section extraction procedure for extracting, as a black section, an area having a predetermined number of black pixels or more from each area of the form and each area of the second division form;
A character area extraction procedure for merging the black sections to form a character string area conforming to the division mode, and for extracting a character area from the character string area,
And a character recognition procedure for recognizing a character pattern in the character area.

Images