JPH0765125A - Recognition device for object character string whose character array is already known - Google Patents

Abstract

PURPOSE:To correctly segment characters even in the case of dirt, a blur, misbinarization, etc., by scanning an extraction window corresponding to the character array which is already known and determining a character segmentation position. CONSTITUTION:Binary image data have a blur 20 and a noise 21 resulting from input conditions and a failure in binarization. The extraction window 22 is scanned for the binary image data and feature quantities are extracted by scanning points 25. The extraction window 22 has size and arrangement corresponding to the character array which is already known; and hatched parts 23 correspond to character areas and their peripheral blank part 24 corresponds to a background area. The feature quantities are represented as the result obtained by subtracting the number of all black pixels in the background area 24 from the number of all black pixels in the character areas 23. The position 26 of the extracted window 22 a where the feature quantity is minimum is regarded as the character segmentation position. Then characters in the character areas 23 in the extracted window 22 at the position 26 are segmented.

Description

Detailed Description of the Invention

[0001]

TECHNICAL FIELD The present invention relates to an apparatus for character recognition of a character string whose character array is known in an input image.

[0002]

BACKGROUND ART A conventional character string recognition process in which a character array is known is a labeling process that binarizes image data around a character string and extracts connected characters on a binarized image, or vertical and horizontal directions. By using the projection length data of, the character cutting position is determined, and the character recognition processing is performed from the result obtained.

However, in such a method, a correct result can be obtained when the input image is good and stable, but the environmental change in the outdoors and the change around the target character string,
It was difficult to deal with the case where the separation between characters was not clear due to dirt, blurring, or an error in binarization.

[0004]

DISCLOSURE OF THE INVENTION The present invention is applicable to a target character string whose character sequence is already known, even when a change in input conditions outdoors, a change around the character string, dirt, blurring, or a binarization error occurs. A character recognition device capable of cutting out a character is provided.

[0005]

A character recognition apparatus according to the present invention comprises an input means for inputting an image containing a target character string whose character arrangement is known, a binarizing means for binarizing the input image, and the above-mentioned 2 A feature amount calculation unit that scans an extraction window having a size and an arrangement corresponding to a known character array with respect to the image data binarized by the value conversion unit, and calculates a feature amount for each position of the extraction window. The cutout position determining means for determining the character cutout position based on the feature amount obtained by the feature amount calculating means, and the image data cut out in accordance with the extraction window of the position determined by the cutout position determining means are stored in advance. A character recognition means for performing a character recognition process based on the reference pattern, and an output means for outputting the result obtained by the character recognition means, It includes those were.

A character recognition method according to the present invention receives an input of an image containing a target character string whose character arrangement is known, performs binarization processing on the received input image, and converts the input image into binarized image data. On the other hand, the extraction window having the size and arrangement corresponding to the known character array is scanned, the feature amount is calculated for each position of the extraction window, and the character cutout position is determined based on the calculated feature amount. The character recognition processing is performed on the image data cut out according to the extraction window of the position based on the reference pattern stored in advance, and the obtained result is output.

According to the present invention, the extraction position for each character is not extracted as the extraction position determining means, but the extraction window corresponding to the character array is utilized by utilizing the known arrangement and size of the character array. By scanning,
Determines the character cutout position in the character string. Therefore, environmental changes outdoors, changes around the target character string, dirt, blurring,
It is possible to correctly cut out a character even if there is a mistake in binarization or the separation between characters is not clear, and it is possible to perform correct character recognition processing.

In another embodiment of the character recognition device according to the present invention, a cutting position adjusting means for finely adjusting the cutting position for each character in the vicinity of the position determined by the cutting position determining means is added.

According to this embodiment, after the CPU determines the character cutout position, the cutout position is finely adjusted for each character in the vicinity of the determined cutout position. Therefore, it is correct even when there are variations in the size and position of each character, when the font such as "1" and "I" is thinner than other characters, and when uppercase and lowercase letters of the alphabet are mixed. Character recognition processing can be performed.

In still another embodiment of the character recognition apparatus according to the present invention, a plurality of types of extraction windows are used according to the size of the recognition target.

According to this embodiment, when the CPU scans the extraction window, it scans a plurality of extraction windows corresponding to a plurality of character arrays assumed in advance, and extracts the corresponding character array and cutout from the obtained feature amount. Determine the position. Therefore, even in a target character string such as a license plate in which a plurality of character arrays of different sizes exist, the correct determination of the corresponding character array and the correct character recognition processing can be performed.

In still another embodiment of the character recognition device according to the present invention, a cutout position candidate extracting means for extracting a plurality of cutout position candidates based on the feature amount obtained by the feature amount calculating means, and the cutout position. A cut-out position candidate selecting means for calculating suitability as a cut-out position for each candidate extracted by the candidate extracting means and selecting a candidate having the highest suitability is added.

According to this embodiment, the CPU extracts a plurality of cutout position candidates from the feature amount obtained by scanning the extraction window, and calculates the suitability of each cutout position candidate.
Character recognition is performed on the most suitable candidate. Therefore, correct character recognition can be performed even when the feature amount of the correct cutout position does not take the maximum value due to a bad input state or a problem of binarization.

In still another embodiment of the character recognition device according to the present invention, a plurality of cutout position candidates are extracted based on the feature amount obtained by the feature amount calculating means,
The cut-out position candidate priority order determining means for giving priority to each candidate, and whether or not the cut-out position is appropriate according to the priority order determined by the cut-out position candidate priority order determining means are sequentially determined. A cutout position candidate determination means for determining the cutout position when it is determined to be appropriate at that time is added.

According to this embodiment, the CPU extracts a plurality of cutout position candidates from the feature amount obtained by scanning the extraction window, and sets the priority order to each cutout position candidate.
Whether or not each candidate is appropriate as the cut-out position is sequentially determined according to the priority order, and character recognition is performed when there is an appropriate one. Therefore, correct character recognition can be performed even when the feature amount of the correct cutout position does not take the maximum value due to a bad input state or a problem of binarization, and the processing time can be further shortened.

In still another embodiment of the character recognition device according to the present invention, a plurality of extraction windows having different sizes are prepared, and extraction window switching means for switching the extraction window to be used according to the position of the image is added. To be done.

According to this embodiment, a plurality of extraction windows having different sizes are prepared in advance, and the CPU
Switches the extraction window to be used according to the position of the image when scanning the extraction window. Therefore, since the target character string is moving in a known fixed direction with respect to the image input means, when the size of the character in the image changes depending on the timing of capturing the input image, for example, it approaches. Correct character recognition processing can be performed even for things such as vehicle license plates.

In still another embodiment of the character recognition device according to the present invention, the edge data is extracted from the image input by the input means, and the character area is limited based on the extracted edge data. The binarizing means for binarizing the image of the area limited by the area limiting means,
Further, a inverting means for determining whether the binary image binarized by the binarizing means is a normal character or a reverse character and inverting the binary image in the case of a reverse character is added.

According to this embodiment, the CPU detects a reversed character from the binarized character line peripheral image, and in the case of a reversed character, reverses the binarized image. Therefore, the correct character recognition process can be performed even when a reverse character such as a license plate is included (for example, a white character on a green background).

Another character recognition apparatus according to the present invention comprises an input means for inputting an image including a target character string whose character arrangement is known, a binarizing means for binarizing the input image, and the binarizing means. Project the binarized image horizontally,
Horizontal projection length data creating means for creating horizontal projection length data, character line cutout means for determining a cutout position of a character line based on the horizontal projection length data created by the horizontal projection length data creating means, With respect to the vertical projection length data of the image cut out by the character line cutting means, a one-dimensional extraction window corresponding to a known character array is horizontally scanned, and a feature amount is calculated for each position of each extraction window. Characteristic amount calculating means, a cutting position determining means for determining a vertical direction cutting position based on the characteristic amount calculated by the characteristic amount calculating means, a character line cut by the character line cutting means and the vertical direction cutting position The image data within the range determined by the vertical cutting position determined by the determination means is stored in advance. Output means for outputting a result obtained by the character recognition means, and said character recognition means performs character recognition processing based on the quasi-pattern, but with a.

According to the present invention, the CPU cuts the character line to narrow the scanning range of the extraction window and scans the extraction window in which the character array is one-dimensional. Therefore, the character recognition processing time can be shortened.

Still another character recognizing device according to the present invention is an input means for inputting an image including a target character string whose character arrangement is known, horizontal edge data is extracted from the image input by the input means, Character line cutout position determining means for determining the cutout position of the character line, character line binarizing means for binarizing the character line image cut out by the character line cutting means, and binary by the character line binarizing means A feature amount calculation means for horizontally scanning a one-dimensional extraction window corresponding to a known character array with respect to the vertical projection length data of the converted image, and calculating a feature amount for each position of each extraction window, Cutout position determining means for determining a vertical cutout position based on the feature amount calculated by the feature amount calculating means, the character line cut out by the character line cutting means, and the hanging line. The image data within the range determined by the vertical cut position determined by the direction cropping position determining means, the character recognition means for performing character recognition process on the basis of the reference pattern stored in advance,
And output means for outputting the result obtained by the character recognition means.

According to the present invention, the CPU cuts out character lines to narrow the binarization range and scanning range of the extraction window, and scans the extraction window in which the character array is one-dimensional. Therefore, the character recognition processing time can be shortened, and at the same time, the amount of memory can be saved.

The character recognition device according to the present invention is more practically applicable to an extraction window in which a plurality of known character arrays are unified in the vertical direction with respect to a plurality of character lines, and a target character whose character array is known. Input means for inputting images including columns, character area limiting means for extracting edge data from the image input by the input means, and limiting character areas based on the extracted edge data, area limited by the character area limiting means Character area binarizing means for binarizing the image of the above, and whether the binary image binarized by the character area binarizing means is a normal character or a reverse character is determined. Is an inversion means for inverting the binary image, and a projection length data creation for creating a vertical projection length data by projecting the binary image obtained by the inversion means in the vertical direction with respect to the character line. Means, the corresponding one-dimensional extraction window is selected from the position of the area obtained by the character area limiting means, and the one-dimensional extraction window selected for the vertical projection length data obtained by the projection length data creating means is selected. A feature amount calculation means for scanning the feature amount to calculate a feature amount, a plurality of cutout position candidates extracted from the feature amount obtained by the feature amount calculation means, and a cutout position candidate priority determination means for prioritizing each candidate, According to the priority order determined by the cut-out position candidate priority determining means, while finely adjusting the cut-out position of each character determined in the vicinity thereof, it is sequentially determined whether or not it is appropriate as the cut-out position, and the one determined to be appropriate is Cut-out position candidate adjustment determination means for determining the cut-out position when it is determined to be appropriate, and the cut-out position candidate adjustment determination A character recognition means for performing character recognition processing on an image cut out by a step based on a reference pattern stored in advance, and an output means for outputting a result obtained by the character recognition means. is there.

According to the present invention, the CPU limits the character area and determines the reverse character so that the correct character recognition processing can be performed even when the reverse character such as a license plate is included, and a plurality of kinds of character arrangements can be obtained. By scanning the corresponding one-dimensional extraction window, correct character recognition processing can be performed even when multiple types of character arrangements such as number plates are conceivable, and the one-dimensional extraction window can be selected depending on the position of the limited area. By switching, the correct character recognition processing can be performed even when the size of the character string changes depending on the position in the image. Noise is extracted by extracting multiple cut-out position candidates and adjusting and judging according to the priority order. When the feature amount at the correct cutout position is not maximized due to the influence of, for example, one-dimensional extraction window Even if the actual position of the character is misaligned, the cut-out position can be determined at high speed and correct character recognition processing can be performed, and the binarized character string image was obtained by projecting in the direction perpendicular to the character string. The character cut-out position can be determined by scanning the one-dimensional extraction window with respect to the projection length data.

[0026]

1 is a layout diagram of an embodiment in which a character recognition device is applied to read a license plate of a vehicle traveling on a road.

The pillars 3 are erected on both sides of the road 4, the horizontal shafts 2 are passed over the pillars, and the television (video) camera 1 is installed on the horizontal shafts 2. The camera 1 images a license plate of a vehicle 5 traveling in a direction toward the camera 1 as shown by an arrow in a predetermined range on the road 4, and outputs a video signal representing the captured image. The video signal output from the camera 1 is the I / O control device 11
Through the character recognition device described later. The pillar 3 may be erected on either side of the road 4 and the camera 1 may be installed above it. It is also possible to attach the camera 1 to an overpass or other structure and take an image of the license plate of a vehicle traveling on the road 4.

FIG. 2 is a view of the arrangement shown in FIG. 1 as seen from the side of the road 4.

As shown in FIG. 2, since the visual angle θ of the camera 1 installed on the horizontal shaft 2 passed on the road 4 is set to be relatively narrow, the image from the camera 1 to the recognition target object in one image is set. There is almost no difference in size depending on the location depending on the distance, and the image of the license plate 6 is not distorted.

This character recognition device can be widely applied to other than the license plate of the vehicle as long as the arrangement of the target character string is known. For example, reading ICs on a printed circuit board, model numbers of products flowing on a line, numbers of railway vehicles, letters and symbols indicating delivery destinations displayed on deliveries in the distribution industry, and character strings written on paper. And its confirmation, etc.

First embodiment

FIG. 3 shows the electrical construction of the character recognition device.

The CPU 10 sets the input image data captured by the camera 1 to the input image data 2 to be described later.
It performs value conversion, scanning of extraction windows, calculation of feature quantities, determination of cutout positions, and character recognition processing of cutout images.

The I / O control device 11 takes in an image input by the camera 1 and stores it in the image storage RAM 12 in a bus.
Input control transmitted via line 19, and CPU1
Output control is performed to output the result of recognition processing by 0 to, for example, a display device.

The image storage RAM 12 is a bus line 1
This is a memory for temporarily storing the input image data transmitted from the I / O control device 11 via 9 and the binary image data that has been binarized by the CPU 10.

The calculation RAM 13 is a memory for temporarily storing the feature amount obtained by scanning the extraction window and the feature amount used for the character recognition processing.

The binarization program ROM 14 is a read-only memory for storing a program necessary for binarizing the input image.

The cut-out control program ROM 15 scans an extraction window stored in an extraction window ROM 16 to be described later with respect to the binarized binary image data, calculates a feature amount, and determines a character cut-out position. It is a read-only memory that stores the programs necessary for performing.

The extraction window ROM 16 is a read-only memory for storing an extraction window corresponding to a character array of a target character string which is assumed in advance.

The character recognition program ROM 17 is a RA for calculating the input image stored in the image storage RAM 12.
It is a read-only memory for storing a portion of the character recognition result stored in M13 and a program necessary for the character recognition processing performed by the CPU 10 based on a reference pattern stored in advance in the character recognition dictionary ROM 18.

The character recognition dictionary ROM 18 is a read-only memory for storing in advance reference patterns necessary for character recognition processing.

The above I / O control device 11 and image storage R
AM 12, calculation RAM 13, binarization program ROM
14, the cut-out control program ROM 15, the extraction window ROM 16, the character recognition program ROM 17, and the character recognition dictionary ROM 18 are respectively bus lines 19
It is connected to the CPU 10 via.

The feature amount calculation process and the cutout position determination process performed by the CPU 10 in the character recognition process will be described with reference to FIG.

FIG. 4A shows the image data obtained by binarizing the input image data containing the character string whose character array is known.

When blurring 20 and noise 21 as shown in the figure occur in the binarization result of a character string due to an input condition or a binarization failure, a character using conventional labeling or projection length is used. It was difficult to determine the exact character cutout position by the cutout method, and as a result, there were often problems with character recognition.

Therefore, here, in order to cut out a character string from the binary image data as shown in FIG. 4A, the feature amount is calculated by scanning the extraction window 22 shown in FIG. 4B. Determine the cutout position. The extraction window 22 of FIG. 4 (B) is for cutting out a 4-digit character string, and the hatched portion 23 is the character area and the surrounding blank portion 2
4 corresponds to the background area, respectively. The position of the extraction window 22 to be scanned is represented by a point 25 at the upper left corner (this is called a scanning point).

FIG. 4C shows a state in which the above-mentioned extraction window 22 is being scanned in order to determine the cut-out position of the 4-digit character string from the binary image data.

The feature amount is represented by a result obtained by subtracting the number of all black pixels included in the background area 24 from the number of all black pixels included in the character area 23. The calculation of this feature is
This is performed for each of the scanning points 25. In general, the extraction window 22 is scanned in the raster direction, that is, starting from the top left of the image data, from left to right, top to bottom.

FIG. 4D shows a state in which the extraction window 22 is at the position 26 where the feature amount shows the maximum value. By cutting out the character in the character area 23 of the extraction window 22 at this position, the correct character recognition processing can be performed. The position of the extraction window 22 where the feature amount shows the maximum value is called a character cutout position.

FIG. 5 is a flowchart showing a character recognition process in the character recognition device.

The image of the vehicle license plate 6 taken by the camera 1 shown in FIG. 1 is input as image data via the I / O control device 11 (step 30) and stored in the image storage RAM 12. Binarization processing is performed on the stored image data based on the program stored in the binarization processing program ROM 14 (step 3
1), the binarized image data is the image storage RAM 12
Memorized in.

Then, the extraction window 22 stored in the extraction window ROM 16 is scanned with respect to the binary image data, the characteristic amount is calculated for each scanning point 25, and the obtained characteristic amount is used as the calculation RAM 13 (Steps 32 to 36). Here, the variable i is a counter indicating the above-mentioned scanning point 25, and the maximum value of the scanning point 25 is N.
That is, the scanning of the extraction window 22 starts from the scanning point i = 1 at the upper left of the image, scans from the left to the right, from the top to the bottom, and ends at the scanning point i = N at the lower right. The extraction window 22 is moved to the scanning point indicated by the counter i (step 33), and the feature amount is calculated by subtracting the total number of black pixels included in the background region 24 from the total number of black pixels included in the character region 23. , Is stored in the calculation RAM 13 (step 34). This process is repeated until the scanning of the entire binary image data is completed (step 35).

When the scanning is completed and the characteristic amounts at all the scanning points are calculated, the maximum characteristic amount among the characteristic amounts stored in the calculation RAM 13 is calculated based on the program stored in the cutout control program ROM 15. Is found, and the position of the scanning point that caused the feature amount is determined as the character cut-out position (step 37). The result is stored in the calculation RAM 13, and out of the input image data stored in the image storage RAM 12, the image data included in the character area 23 of the extraction window 22 at the character cutting position is cut out. That is, it is extracted for character recognition processing (step 38).

Finally, character recognition processing is performed on the cut out image data based on the program stored in the character recognition program ROM 17 and the reference pattern stored in the character recognition dictionary ROM 18 (step 3).
9). Character recognition processing is performed by normal pattern matching or other methods. Character recognition result is I
Is output via the I / O controller 11 (step 40),
All processing ends.

Second embodiment

In the second embodiment, a cutout position adjusting process is added to the process of the first embodiment.

The electrical configuration of FIG. 3 can be used as it is in the second embodiment.

Depending on the positional relationship between the camera 1 and the license plate 6 to be recognized, the space between the characters on the license plate 6 may become narrow or wide on the image. In the first embodiment, the extraction window 2
The mutual spacing of the character areas 23 in 2 is predetermined. Therefore, as shown in FIG. 6 (A), there may occur a state in which not all the characters fit in all the character areas 23 even at the correct cutout position. In such a case, accurate character recognition processing cannot be performed.

Therefore, after the extraction window 22 is brought to the cutout position by the method of the first embodiment, the extraction window 22 is divided for each character. That is, FIG. 6 (B)
As shown in FIG. 5, the extraction window 50 is provided for each character in the range including the character area 23 of the extraction window 22, and the character area 5 for the character area 23 is provided in the extraction window 50.
Set 1. In the extraction window 50, a portion outside the character area 51 is a background area 52.

The feature amount is calculated for each extraction window 50. The total number of black pixels included in the background area 52 is subtracted from the total number of black pixels included in the character area 51, and the result of the subtraction becomes the feature amount in the extraction window 50. The position of each extraction window 50 is finely adjusted by moving the extraction window 50 vertically and horizontally as shown in FIG. 6C so that the feature amount becomes maximum.

FIG. 7 shows the character recognition processing of the second embodiment.
It is a flowchart corresponding to FIG.

The processing of steps 30 to 37 is the same as each processing in FIG.

In step 37, the extraction window 22
After the character is positioned, the extraction window 50 for each character is extracted using the extraction window 50 divided as described above.
Is finely adjusted (step 60). When the extraction position is determined for each extraction window 50, each extraction window 5
The image data in the character area 51 within 0 is cut out (step 38), and character recognition processing is performed based on the cut out image data (step 39).

Third embodiment

In the third embodiment, a process using a plurality of extraction windows is added to the process of the first embodiment in order to deal with the fact that the size of the characters on the license plate differs depending on the vehicle type.

The electrical configuration of FIG. 3 can be used as it is in the third embodiment.

As shown in FIG. 8, there are two types of vehicle license plates, a license plate 6A for large vehicles and a license plate 6B for medium-sized vehicles. Therefore, the extraction window to be scanned is also licensed. It is necessary to prepare two types according to the size of the plate.

Therefore, a plurality of (in this case, two types) extraction windows (2 in this case) of different sizes are assumed.
2A and 22B) are prepared, each extraction window is scanned, and the feature amount is calculated at each scanning point,
The position of the extraction window that produces the maximum value among all the feature amounts is determined as the cutout position.

9 and 10 are flow charts showing the character recognition processing of the third embodiment and corresponding to FIG.

A counter j for scanning a plurality of types of extraction windows stored in advance in the extraction window ROM 16 is provided. The counter j is set to j = 1 (step 80), the binary image is scanned using the extraction window selected by the counter j (step 81), and the feature amount is calculated at each scanning point (step 32). ~ 36).

This process is performed by incrementing the counter j (step 83) and storing n extraction windows (n = 2 in the case of the above-mentioned license plate).
(Step 82). The cut-out position is determined as the position of the extracted window that produces the maximum value among the calculated feature amounts, that is, the feature amounts of all scanning points N × all extracted windows n.

Fourth Embodiment

The electrical configuration of FIG. 3 can be used as it is in the fourth embodiment.

11 and 12 are flow charts corresponding to FIG. 5 showing the character recognition processing of the fourth embodiment. Here, the extraction window 22 of the first embodiment is used.

The processing of steps 30 to 36 is the same as that shown in FIG.

In step 90, based on the program stored in the cut-out control program ROM 15, the extraction window 22 stored in the calculation RAM 13 is scanned, and the characteristic values obtained in this order from the largest value are ranked m in descending order. The individual is selected as a cutout position candidate.

Next, referring to the image data included in the character area at each candidate position of the input image and the binary image stored in the image storage RAM 12 for each selected candidate, the character string The suitability as the cutout position is calculated (steps 91 to 94). The candidate position having the maximum compatibility is determined as the character cutout position (step 95). Relevance is a value that represents character-likeness,
Specifically, it is represented by the total number of black pixels included in the character area 23, the character feature amount, and the like.

Fifth embodiment

The electrical configuration of FIG. 3 can be used as it is in the fifth embodiment.

FIGS. 13 and 14 are flow charts corresponding to FIG. 5, showing the character recognition processing of the fifth embodiment.

The processing of steps 30 to 36 is the same as that shown in FIG. In addition, among the feature values obtained by scanning the extraction window 22, the upper m
Processing for selecting individual pieces as cut-out position candidates (step 9)
0) is the same as in the fourth embodiment.

Here, the priority is determined in descending order of the calculated value of the feature quantity (step 100).
According to the determined priority order, the input images stored in the image storage RAM 12 are the same as in the fourth embodiment.
The suitability as the cut-out position of the character string is calculated with reference to the image data included in the character area at each candidate position in the binary image (steps 91 to 94). Here, since the priority of the candidate positions has already been determined, the suitability is compared with a predetermined threshold value (step 101). Determined as the cutout position (step 1
02).

Sixth Embodiment

In the sixth embodiment, the processing of the first embodiment is added with the processing of scanning while switching the extraction windows of different sizes depending on the scanning position.

The electrical configuration of FIG. 3 can be used as it is in the fifth embodiment.

As shown in FIG. 1, when the image is taken looking down from the front with respect to the traveling direction of the vehicle, if the viewing angle θ of the camera 1 is set to be relatively narrow as shown in FIG. The size of the image of the license plate 6 hardly changes depending on the location depending on the distance from the object to the recognition object, and the image of the license plate 6 is not distorted. However, when the viewing angle θ is set to be large as shown in FIG. 15 (A), when the vehicle is in the front (in FIG. 15 (B), the vehicle is on the lower side), 5A and when the vehicle is in the rear. At some time (when the vehicle is on the upper side in FIG. 15B), the size of the target character strings (6C and 6D) differs from 5B, and correct character recognition processing cannot be performed.

Therefore, when scanning the extraction window,
Scanning is performed while switching a plurality of extraction windows (22C and 22D) having different sizes depending on the position of the screen, and the character cut-out position is determined from the obtained feature amount.

FIG. 16 is a flow chart corresponding to FIG. 5, showing the character recognition processing of the sixth embodiment.

In step 120, based on the program stored in the cut-out control program ROM 15, the plurality of extraction windows of different sizes stored in the calculation RAM 13 are moved to the scanning position, that is, the position corresponding to the scanning point i. An extraction window of an appropriate size is selected, scanning is performed while switching, and the character cut-out position is determined from the obtained feature quantities.

Seventh Embodiment

The seventh embodiment deals with a white license plate and a green license plate.

FIG. 17 is a diagram corresponding to FIG. 3, showing the electrical construction of the character recognition device in the seventh embodiment. The same components as those shown in FIG. 3 are designated by the same reference numerals to avoid redundant description.

The character area limited program ROM 130 is
It is a read-only memory that stores a program necessary for the process of limiting the character area from the input image data stored in the image storage RAM 12.

The reverse character determination program ROM 131 is
Of the binary image data stored in the image storage RAM 12, it is determined whether the image included in the defined character area is a reverse character, and if the reverse character is input, the image is saved. It is a read-only memory that stores a program necessary for the process of inverting the binary image data of the corresponding portion stored in the RAM 12 for use.

Character area limited program ROM 13 described above
The 0 and the reverse character determination program ROM 131 are connected to the CPU 10 via the bus line 19, respectively.

The seventh embodiment is obtained by adding the process of limiting the character area and the process of inverting the binary image included in the predetermined character area to the processing of the first embodiment.

The number plate of the vehicle is shown in FIG.
As shown in (A), a green character string (normal characters) on a white background
Is written on the white plate 6E, and the green plate is the inverted green plate on which the white character string (reverse character) is written.
There are two types. Therefore, in order to perform correct character recognition processing, it is necessary to reverse the reversed character.

FIG. 18B shows an example of input image data including a character string whose character array is known but which is not known as a normal character or a reverse character. Here, first, the character region (the region in which the characters are written) is generally limited, and the binarization process is performed within the limited region. It is determined based on the binarized image data whether the character is a reverse character. If the character is a reverse character, the character is recognized after being reversed.

The character area is determined by calculating the edge component for each pixel or for each appropriate small area in the image data input through the I / O control device 11 (edge data calculation), This is performed by setting a range in which data (having a large edge likelihood) is collected as a character area. By extracting the edge data, information indicating the existence of the character appears regardless of the black and white (green) inverted character, and therefore the character area can be limited.

19 and 20 are flowcharts corresponding to FIG. 5, showing the character recognition processing of the seventh embodiment.

After inputting the image data (step 30), the edge data is extracted from the input image stored in the image storage RAM 12 based on the program stored in the character area limiting program ROM 130 (step 150). The character area is extracted based on this (step 151).

In step 152, based on the program stored in the binarization program ROM 14, the image data of the limited character area stored in the image storage RAM 12 is binarized and stored in the image storage RAM 12. .

In step 153, the number of black pixels and white pixels included in the binary image of the limited character area stored in the image storage RAM 12 is determined based on the program stored in the reverse character determination program ROM 130. By counting, it is determined whether it is a normal character or a reverse character. When the number of white pixels is larger than the number of black pixels, it is determined as a normal character, and when the number of black pixels is greater than the number of white pixels, it is determined as a reverse character. In the case of reverse characters, the binary image stored in the image storage RAM 12 is reversed in black and white (step 154). In the case of normal characters, the image data is saved as it is.

Among the above processing, the cutout position determination and the character recognition processing according to steps 32 to 40 shown in the first embodiment are performed on the binarized image data in the character area.

Eighth Embodiment

In the eighth embodiment, the one-dimensional extraction window is scanned in the cut out range to determine the cut out position.

FIG. 21 is a diagram corresponding to FIG. 3, showing the electrical construction of the character recognition device in the eighth embodiment. The same components as those shown in FIG. 3 are designated by the same reference numerals to avoid redundant description.

The character line cutout program ROM 160 is
For images stored in the image storage RAM 12,
This is a read-only program that stores a program necessary for the process of cutting out a character line.

The cut-out control program ROM 15A is a read-only program for storing the program necessary for performing the cut-out processing using the one-dimensional extraction window.

The extraction window ROM 16A is a read-only program for storing the one-dimensional extraction window.

The above character line cutout program ROM 16
0, the cutout control program ROM 15A, and the extraction window ROM 16A are connected to the CPU 10 via the bus line 19, respectively.

A character line cutout process according to the eighth embodiment will be described with reference to FIGS. 22, 23 and 24.

Horizontal projection length data 141 is calculated for the binarized image data (step 180). A character line is cut out based on the horizontal projection length data 141 (step 181). Horizontal projection length data 1
A range V1 in which 41 is a predetermined value or more is cut out.

The vertical projection length data 140 is calculated within the range V1 defined by the character line cutout.

A one-dimensional extraction window 170 is used. This one-dimensional extraction window 170 has a character range 171.
And a background range 172 between them. This one-dimensional extraction window 170 is scanned in the horizontal direction within a predetermined range.

At each scanning position, the sum of the vertical projection length data belonging to the background range 172 is subtracted from the sum of the vertical projection length data belonging to the character range 171 in the one-dimensional extraction window 170 to obtain a feature amount (step 18
2, 34).

The position of the one-dimensional extraction window 170 showing the maximum feature amount is the range V2 to be finally cut out (steps 37, 38).

Character recognition processing is performed based on the image data belonging to the areas extracted by the ranges V1 and V2 (steps 39 and 40).

In this embodiment, the one-dimensional extraction window 1
Since the scanning of 70 only needs to be performed once in the horizontal direction, the cutting process becomes quick.

Ninth Embodiment

Like the eighth embodiment, the ninth embodiment scans the one-dimensional extraction window in the cut-out range to determine the cut-out position. However, the processing range is before the binarization process. Therefore, the amount of memory used for the binarization processing can be saved.

The character line cutout process according to the ninth embodiment will be described with reference to FIGS. 25, 26 and 27.

First, image data as shown in FIG. 25A is input (step 30). Here, the entire input image data is not binarized, but a character line is cut out and binarized only on the character line.

In step 200, the image storage RA
With respect to the input image data stored in M12, the horizontal line edge data is extracted by the program stored in the character line cutout program 160, and the character cutout is performed based on the extracted edge data.

In step 31, binarization processing is performed only on the cut-out character line, and vertical projection length data is created for the binarized image (step 20).
1), stored in the calculation RAM 13. FIG. 25 (B)
Indicates that the character area is cut out and vertical projection length data is created.

Similar to the above-described eighth embodiment, the extraction window ROM 1 is applied to the vertical projection length data 170.
One-dimensionalized extraction window 1 stored in 6A
70 is scanned (step 182), and the characteristic amount is calculated (step 34).

Tenth Example

The tenth embodiment shows a more practical character recognition device of the present invention, which is a combination of a plurality of processes selected from the first to ninth embodiments.
It demonstrates using the flowchart of FIGS.

The image of the vehicle license plate 6 taken by the camera 1 shown in FIG. 1 is input as image data through the I / O control device 11 (step 30).

Edge data is extracted from the input image data (step 150), and the character area is limited based on the edge data and the character area limiting program (step 151).

The image data of the portion of the input image corresponding to the character area is binarized based on the binarization program (step 152).

Based on the reverse character determination program, the number of black pixels and white pixels included in the binary image of the limited character area is counted, and it is determined whether it is a normal character or a reverse character (step 153). In the case of a reverse character, the stored binary image is inverted in black and white (step 154), and in the case of a normal character, the image data is saved as it is.

According to the cutout control program, the stored binary image is projected in the vertical direction to create vertical projection length data.

Counter j for scanning a plurality of types of extraction windows stored in advance (the maximum value is n.
Here, n = 2) is set to j = 1 (step 8
0), the extraction window j is selected (step 8).
1). The counter i indicating the scanning point (the maximum value is N) is i = 1
Is set, and the extraction window having a size suitable for the position corresponding to the scanning point i is scanned, and the feature amount is calculated for each scanning point (steps 32 to 36).

Among the feature values obtained by scanning the extraction window, the m uppermost ones are selected as cut-out position candidates in descending order of their value (step 90), and the calculated feature value is prioritized in descending order. Determine ranking (step 10)
0).

According to the determined priority order, referring to the image data included in the character area at each candidate position of the stored input image and binary image, while finely adjusting the extraction window vertically and horizontally, The suitability as the cut-out position of the character string is calculated (steps 91 to 94). The suitability is compared with a predetermined threshold value (step 101), and if the candidate position shows the suitability equal to or higher than the threshold value, it is determined as the cutout position of each character (step 1).
02).

From the stored input image data, the image data included in the character area of the extraction window of the character cut-out position is cut out, and based on the character recognition program and the reference pattern stored in the character recognition dictionary ROM 18,
Character recognition processing is performed on the cut out image data (step 39). The character recognition result is the I / O control device 11
(Step 40), and the whole process ends.

As in the tenth embodiment, any two or more of the first to ninth embodiments can be selected and appropriately combined within a range that does not contradict.

[Brief description of drawings]

FIG. 1 is an arrangement configuration diagram of an embodiment in which a character recognition device is used to read a license plate of a vehicle traveling on a road.

FIG. 2 is a view of a layout configuration diagram of the embodiment shown in FIG. 1 viewed from a side surface of a road.

FIG. 3 is a block diagram showing an electrical configuration of the character recognition device in the first embodiment.

4A is an example of an input image, FIG. 4B is an example of an extraction window, FIG. 4C is a state in which the extraction window is scanned on a binary image, and FIG. 4D is positioned at a cutout position. The extraction windows are shown respectively.

FIG. 5 is a flowchart showing a character recognition process in the first embodiment.

6A is a state in which the character region of the extraction window positioned at the cutout position and the character are misaligned, FIG. 6B is a state in which the extraction window is divided for each character, and FIG. The respective adjustment states are shown.

FIG. 7 shows a character recognition process in the second embodiment, FIG.
It is a flowchart corresponding to.

FIG. 8 is a view comparing a license plate for a large vehicle and a license plate for a medium-sized vehicle.

FIG. 9 shows a character recognition process in the third embodiment, and FIG.
Is a part of the flowchart corresponding to.

FIG. 10 is a part of a flowchart corresponding to FIG. 5, showing a character recognition process in the third embodiment.

FIG. 11 is a part of a flowchart corresponding to FIG. 5, showing a character recognition process in the fourth embodiment.

FIG. 12 is a part of a flowchart corresponding to FIG. 5, showing a character recognition process in the fourth embodiment.

FIG. 13 is a part of a flowchart corresponding to FIG. 5, showing a character recognition process in the fifth embodiment.

FIG. 14 is a part of a flowchart corresponding to FIG. 5, showing a character recognition process in the fifth embodiment.

15A is a view of the arrangement shown in FIG. 1 as seen from the side of the road 4, in which the vision of the camera is set larger than in FIG. 2, and FIG. It is the figure which looked at from the position of the camera shown, and shows the state where the size differs according to the location according to the distance.

FIG. 16 is a flowchart corresponding to FIG. 5, showing a character recognition process in the sixth embodiment.

FIG. 17 is a block diagram showing an electrical configuration of a character recognition device in a seventh embodiment.

FIG. 18A is a comparison between a license plate of normal characters (green on a white background) and a license plate of reverse characters (white on a green background), (B) is an input image, and (C) is a character region cut out. Indicates the status.

FIG. 19 is a part of a flowchart corresponding to FIG. 5, showing a character recognition process in the seventh embodiment.

FIG. 20 is a part of a flowchart corresponding to FIG. 5, showing a character recognition process in the seventh embodiment.

FIG. 21 is a block diagram showing an electrical configuration of a character recognition device in an eighth embodiment and a ninth embodiment.

FIG. 22 shows a state in which a character line is cut out from a binary image using horizontal projection length data and vertical projection length data is extracted in that area, and one-dimensional extraction is performed on the vertical projection length data. An example of Windu is shown.

FIG. 23 is a part of a flowchart corresponding to FIG. 5, showing a character recognition process in the eighth embodiment.

FIG. 24 is a part of a flowchart corresponding to FIG. 5, showing a character recognition process in the eighth embodiment.

FIG. 25A is a state in which a character line is cut out from the input image by using edge data and FIG. 25B is a state in which vertical direction projection length data is extracted in that area, and the vertical direction projection length. An example of a one-dimensional extraction window for scanning over data is shown.

FIG. 26 is a part of a flowchart corresponding to FIG. 5, showing a character recognition process in the ninth embodiment.

FIG. 27 is a part of a flowchart corresponding to FIG. 5, showing a character recognition process in the ninth embodiment.

FIG. 28 shows character recognition processing in the tenth embodiment,
6 is a part of a flowchart corresponding to FIG. 5.

FIG. 29 shows character recognition processing in the tenth embodiment,
6 is a part of a flowchart corresponding to FIG. 5.

FIG. 30 shows character recognition processing in the tenth embodiment,
6 is a part of a flowchart corresponding to FIG. 5.

FIG. 31 shows a character recognition process in the 10th embodiment,
6 is a part of a flowchart corresponding to FIG. 5.

[Explanation of symbols]

 1 TV (video) camera 5 Vehicle 6 License plate 22 Extraction mask 23 Character area 24 Background area 25 Scan point 170 One-dimensional extraction window

Claims (10)

[Claims] 1. Input means for inputting an image containing a target character string whose character arrangement is known, and 2 for binarizing the input image.
The binarizing unit scans the image data binarized by the binarizing unit with an extraction window having a size and arrangement corresponding to a known character array, and calculates a feature amount for each position of the extraction window. And a cut-out position determining unit that determines a character cut-out position based on the feature amount obtained by the feature amount calculating unit, and the extraction window of the position determined by the cut-out position determining unit. A character recognition device comprising: a character recognition means for performing character recognition processing on image data based on a reference pattern stored in advance; and an output means for outputting a result obtained by the character recognition means. 2. The character recognition device according to claim 1, further comprising cutout position adjusting means for finely adjusting the cutout position for each character in the vicinity of the position determined by the cutout position determining means. 3. The character recognition device according to claim 1, wherein a plurality of types of extraction windows are used according to the size of the recognition target. 4. A cut-out position candidate extracting means for extracting a plurality of cut-out position candidates based on the feature amount obtained by the feature amount calculating means, and a cut-out position for each candidate extracted by the cut-out position candidate extracting means. 2. The character recognition device according to claim 1, further comprising: a cutout position candidate selecting unit that calculates the suitability of and selects a candidate having the highest suitability. 5. A cut-out position candidate priority order determination unit that extracts a plurality of cut-out position candidates based on the feature amount obtained by the feature amount calculation unit and prioritizes each candidate.
And whether or not the cutout position is appropriate according to the priority order determined by the cutout position candidate priority order determination means, and when there is one that is determined to be appropriate, the one that is determined to be appropriate is determined as the cutout position. The character recognition device according to claim 1, further comprising: a cutout position candidate determination unit. 6. The character recognition device according to claim 1, further comprising extraction window switching means for preparing a plurality of extraction windows having different sizes and switching the extraction window to be used in accordance with the position of the image. 7. A character area limiting means for extracting edge data from an image input by the input means and limiting a character area based on the extracted edge data, and a binary image of an area limited by the character area limiting means. It is determined whether the binarizing means for digitizing and the binary image binarized by the binarizing means are normal characters or reverse characters, and in the case of reverse characters, the binary image is reversed. The character recognition device according to claim 1, further comprising a reversing means. 8. Input means for inputting an image containing a target character string whose character arrangement is known, and 2 for binarizing the input image.
Created by the binarizing unit, the horizontal projection length data creating unit that projects the image binarized by the binarizing unit in the horizontal direction, and creates the horizontal projection length data, and the horizontal projection length data creating unit. Character line cutout means for determining the cutout position of the character line based on the horizontal projection length data, and known character arrangement for the vertical projection length data of the image cut out by the character line cutout means The one-dimensional extraction window is scanned in the horizontal direction, and the feature amount calculation means calculates the feature amount for each position of each extraction window, and the vertical cut-out position is determined based on the feature amount calculated by the feature amount calculation means. The cut-out position determining means, the character line cut out by the character line cutting-out means, and the vertical direction cut-out position determined by the vertical cut-out position determining means. A character provided with character recognition means for performing character recognition processing based on a prestored reference pattern for image data within the range determined by the above; and output means for outputting a result obtained by the character recognition means. Recognition device. 9. Input means for inputting an image including a target character string whose character arrangement is known, characters for extracting horizontal edge data from the image input by said input means, and determining a cutout position of a character line. Line cut-out position determining means, character line binarizing means for binarizing the character line image cut out by the character line cutting means, and vertical projection length of the image binarized by the character line binarizing means. A one-dimensional extraction window corresponding to a known character array is horizontally scanned with respect to the data, and a feature amount calculation means for calculating a feature amount for each position of each extraction window is calculated by the feature amount calculation means. A cutout position determining means for determining a vertical cutout position based on the characteristic amount, a character line cut out by the character line cutting out means, and a vertical direction cutout position determining means. Character recognition means for performing character recognition processing on the image data within the range determined by the vertical cutout position based on a previously stored reference pattern, and output means for outputting the result obtained by the character recognition means, Character recognition device. 10. An extraction window in which a plurality of known character arrays are unified in the vertical direction with respect to a plurality of character lines, and an input means for inputting an image including a target character string whose character array is known, A character area limiting unit that extracts edge data from the image input by the input unit and limits the character area based on the extracted edge data, and a character area 2 that binarizes the image of the area limited by the character area limiting unit. Valuing means, inverting means for determining whether the binary image binarized by the character area binarizing means is a normal character or a reverse character, and inverting the binary image in the case of a reverse character, The binary image obtained by the reversing means is projected in the vertical direction on the character line, and the projection length data creating means for creating the vertical projection length data is obtained by the character area limiting means. A corresponding one-dimensional extraction window is selected from the position of the selected region, and the selected one-dimensional extraction window is scanned with respect to the vertical projection length data obtained by the projection length data creating means to calculate the feature amount. By the feature amount calculating means, the plurality of cutout position candidates are extracted from the feature amounts obtained by the feature amount calculating means, and the cutout position candidate priority order determining means and the cutout position candidate priority order determining means assign a priority to each candidate. According to the determined priority, the cut-out position of each character determined is finely adjusted in the vicinity thereof to sequentially judge whether or not it is appropriate as the cut-out position, and when there is one that is judged to be appropriate, it is judged to be appropriate. A cut-out position candidate adjustment determining unit that determines an object as a cut-out position, and an image extracted by the cut-out position candidate adjustment determining unit Character recognition means performs character recognition processing based on the reference pattern being beforehand stored, and a character recognition apparatus having an output unit, which outputs the result obtained by the character recognition means.