JP2000134609A - Image processing method and method for searching number plate while using the same - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method for exactly recognizing/searching the character image of a number plate or the like at high speed. SOLUTION: The number plate composed of projecting characters is sampled as the image of (SH×m) pixels (SV×n) scans (n>=2), the thinning of (m) pixels × (n) scans is performed to this sampled image and the image is made into the image of SH pixels × SV scans. Concerning such an image processing method, a step for performing the (m) pixels × (n) scans thinning is provided with a step for performing the simple thinning of the (m) pixels or the average processing of (m) pixels, a step for performing simple thinning for extracting one scan per (k) scans on the condition of n=k×j (no thinning on the condition of k=1) and a step for taking out one middle value per (j) scans in scanning direction after the end of these (m) pixels simple thinning or average processing step and (k) scans simple thinning process. In this case, when the number of scan lines related to a luminance fluctuating part corresponding to the upper and lower edge parts of the projecting characters among the (SV×n) scans is defined as k1 and the number of scan lines related to the line width of the projecting characters is defined as k2, concerning the (k), the conditions of k1<=k<=k2 and j>=3 are established.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing method and a license plate searching method using the method.
In a high-speed search for a license plate composed of convex characters, the present invention relates to a method of realizing a high-speed search by avoiding a problem due to a fluctuation in luminance occurring at an edge portion of the convex character.

[0002]

2. Description of the Related Art Generally, in an automatic search for a vehicle number, a license plate of a vehicle is imaged by a camera or the like, and the obtained still image is processed to automatically recognize the license plate of the vehicle. The number is being detected.

FIG. 4 is a diagram for explaining a conventional method of automatically recognizing and searching for a vehicle number. In the conventional high-speed search method, the license plate of the vehicle is changed to (SH ×
(m pixels) × (SV × n scanning) images are collected, and the data amount is reduced by performing simple thinning or average thinning processing on this image so that one pixel per m pixels and one scanning per n scanning. The number recognition / search is performed on the thinned image obtained as described above. In the example shown in FIG. 4, with m = 2 and n = 4, the thinning process is performed on the (SH × 2 pixels) × (SV × 4 scan) number image, and the SH pixel × SV scan image after the thinning is performed. , A number recognition and a matching search are performed. By performing such a thinning process, the search time for each image can be reduced, and a high-speed search can be performed.

[0004]

However, in a high-speed matching search for an image obtained by thinning out an m-pixel × n-scanning block, which is frequently used in the prior art, as described below, the license plate has a convex character. Therefore, there is a problem that it is difficult to obtain an accurate matching result.

Referring to FIG. 5, a description will be given of a luminance level when the license plate is scanned in the vertical direction. FIG.
As shown in (a), a four-character number is arranged on a license plate in a predetermined arrangement. This license plate is a general passenger car number, and is composed of white letters 11 and green letters 12.
FIG. 5B is a cross-sectional view taken along a dotted line TT ′ shown in FIG. As shown in FIG. 5B, the characters 12 on the license plate are formed as convex characters by embossing. Here, since light such as sunlight or an electric light normally shines on the license plate from above (in the diagonally upper left direction in FIG. 5B), the upper and lower edge portions 15 and 16 of the line of the convex characters 12 , The brightness will fluctuate.

FIG. 5 (c) is a diagram showing a change in luminance on the line TT '. As shown in FIG. 5C, the brightness of the upper edge portion 15 of the convex character 12 illuminated by light is much higher than that of the background 11, and conversely, the convex character 12 serving as a shadow
The lower edge portion 16 has lower luminance than the green character 12 portion. Therefore, if the license plate in this state is sampled as an image, the pixel level of the upper and lower edge portions may not fall between the pixel level of the background portion and the pixel level of the character portion. If such a sampled image is subjected to simple thinning or averaging from a block of m pixels × n scans as in the past, the edge portion (fluctuation in brightness) is also targeted, so that an error in number recognition may occur. There is a problem that causes.

In order to avoid this problem, as shown in FIG. 6, a large block of m pixels × n scans corresponding to one pixel after thinning is taken, and a median value is obtained to obtain a fluctuation value. There are ways to avoid it. However, in this method, since the block to be thinned out needs to be relatively large, the amount of calculation required for the calculation of the median value in that block increases, and the search can be speeded up. There is a problem that there is not. In particular, MM
Since the CPU such as X (trade name) is not good at this median value processing, image processing takes a relatively long time. As described above, there is a problem that the above-described method of obtaining a median value in a block of m pixels × n scans and thinning out cannot be simply applied.

In view of such a problem, an object of the present invention is to provide a method for realizing high-speed and accurate recognition / search of a character image such as a license plate.

[0009]

Means for Solving the Problems Among the inventions according to the present application,
2. The image processing method according to claim 1, wherein a sampling image having SV × n (n ≧ 2) scanning lines is thinned out by n scanning to obtain an image having SV scanning lines. In the thinning-out step, n = k × j is set, and a simple thinning is performed to extract one scan per k scans (no thinning is performed when k = 1), and then one median value is extracted to j scans. And

As described above, when performing the thinning-out processing of the n-scanning on the sampled image of the SV × n-scanning, instead of performing the median value processing of the n-scanning from the beginning, in the present invention, n = k × j
First, simple thinning of k scans is performed, and then the middle value processing is performed for j scans.
High-speed processing can be performed as compared with the conventional thinning-out method in which the middle value processing of scanning is performed.

In the method according to the second aspect of the present invention, a sampling image of (SH × m) pixels × (SV × n) scans (where n ≧ 2) is thinned out by m pixels × n scans.
In the image processing method for forming an image of pixels × SV scanning, the step of performing the m pixels × n scanning thinning is a step of performing the simple thinning of the m pixels or the average processing of the m pixels, and the n = k × j, A step of performing simple thinning for extracting one scan per k scans (no thinning when k = 1), j scanning in the scanning direction after the m-pixel simple thinning or averaging process and the k-scan simple thinning process are completed Extracting one median value for each.

As described above, since the normal light is applied to the license plate from the upper side, the fluctuation of the luminance level due to the influence of the convex character is particularly large at the horizontal edge portion. The present invention pays attention to this point, and simply performs simple thinning or averaging processing of m pixels in the pixel direction, and performs median value processing only in the scanning direction to eliminate the scanning line related to the level fluctuation. I have. Therefore, since the scanning line of the luminance level variation portion is eliminated in the image after thinning, the reliability of the image after thinning can be improved. At the same time, the processing can be speeded up by simply thinning out the k scans before performing the median value processing.

In such a thinning processing method, in the (SV × n) scan, the number of scanning lines relating to a luminance variation portion corresponding to the upper and lower edge portions of the convex character is k1, and the line width of the convex character is When the number of such scanning lines is k2, k1 ≦ k ≦ k2 and j ≧ 3
It is desirable that In the simple k-scan simple thinning performed earlier, the simple thinning is performed at an interval (k) that is equal to or more than the scanning line width (k1) related to the brightness variation portion. The number of scanning lines can be reduced to one or less. In the middle value processing of the j-scan performed later, a value corresponding to the center of the j-scans is taken out. Therefore, even when one scan line having a luminance variation is included, j ≧
If it is 3, this fluctuating scanning line can be reliably eliminated. Therefore, the reliability of the image after thinning can be improved.

[0014] In the invention according to the present application, the method according to claim 6 performs thinning of m pixels on a sampled image in which one scanning line is composed of SH × m (m ≧ 2) pixels, and performs one pixel thinning. In the image processing method for forming an image in which the scanning line is composed of SH pixels, the step of performing the m-pixel thinning is a simple thinning (o = 1) in which one pixel is extracted for every o pixels, where m = ox × p.
In the case of (1), there is no thinning out), and then one middle value is extracted for each p pixel. As described above, even when thinning is performed in the pixel direction of an image composed of pixels × scanning lines, the same processing as the thinning processing method in the scanning direction described in claim 1 is performed in the pixel direction, so that m is initially set. The thinning-out processing can be performed at a higher speed than in the case where the middle-value processing for the entire pixel is performed.

In the method according to the present invention, the sampling image of (SH × m) pixels × (SV × n) scanning is thinned out by m pixels × n scanning and the SH pixel × SV scanning is performed. In the image processing method for forming an image, the step of thinning out the m pixels × n scans performed the simple thinning (one without k in the case of k = 1) in which one scan is performed for every k scans, where n = k × j. Later, taking out one median value per j scans, and assuming m = o × p, 1
A step of extracting one median value per p pixels after performing simple thinning for extracting pixels (no thinning when o = 1). Claim 1 described above.
In the methods described in Nos. 1 to 7, only one of the pixel and the scanning is thinned out, but the thinning process of both the pixel and the scanning may be performed in this way.

[0016]

Embodiments of the present invention will be described below with reference to the accompanying drawings. FIG. 1 is a diagram illustrating a license plate search method according to the present invention.
As shown in FIG. 1, the search method of the present invention includes a step 101 of collecting an image of a license plate, a step 102 of simply thinning out or averaging m pixels in the pixel direction,
The method includes a simple scanning thinning-out processing step 103, a j-scan median value processing step 104, and a matching search processing step 105.

First, an image of a license plate to be searched is collected (step 101). The image collected here can be represented as (SH × m pixels) × (SV × n scanning). In this sampled image, since the characters 12 of the license plate are configured as convex characters as shown in FIG. 5, especially the upper and lower edge portions of the lines of the convex characters 12 (reference numerals 15 and 16 in FIG. 5B). Part shown)
Causes a change in the luminance level. Here, the scanning line width at which the level changes at the edge portion of the convex character 12 is k
1, and the scanning line width corresponding to the character 12 is k2.

In the pixel direction of the sampled image, simple thinning or simple averaging of m pixels is performed to obtain one pixel at m pixel intervals, and m pixels are thinned to one pixel (step 102). In order to prevent the character line from disappearing, it is necessary to satisfy m <o2, where o2 is the number of pixels corresponding to the character line in the pixel direction. When this processing is completed, the image becomes (SH pixels) × (SV × n scanning).

Next, assuming that n = k × j (k1 (scanning line width having a level variation at an edge portion) ≦ k <k2 (scanning line width corresponding to a character line)) A simple thinning process for taking out one scan at k scan intervals is performed.
Scanning is reduced to one scan (step 103). When this step is completed, the image becomes (SH pixels) × (SV × j scan).
Since k1 (the number of scanning lines having a level variation) ≦ k, the scanning line having a level variation included in the j scan is 0 or 1, and k <k2 (the number of scanning lines corresponding to a character line). Therefore, the j scan includes at least one or more scanning lines corresponding to the line of the character 12. Here, since the median value processing is performed on j scans in a later step, j is at least 2 or more, and it is desirable that j ≧ 3 in order to calculate an effective median value.

In addition, as shown in FIG.
02 and step 103 may be in front of or behind.
The order of these steps 102 and 103 is determined by considering the manufacturing conditions and the execution conditions, etc., and selecting the one that facilitates the processing is performed first.

Next, for the thinned image (SH pixel) × (SV × j scan) processed in the above steps 102 and 103, a median value calculation process is performed for j scan lines, and j scan is performed for one scan. (Step 104). By this median value processing, the scan line having the level fluctuation is finally removed,
An m-pixel × n-scanning thinned-out image is obtained in which one scanning line corresponding to a character is always left.

For this m pixel × n scanning thinned image,
A license plate matching search process is performed using a predetermined template (step 105) to obtain license plate candidate points.

FIG. 3 is a diagram showing an embodiment in which the above method is applied to perform a thinning process. In the example shown in FIG. 3, in order to facilitate understanding, a case where thinning is performed only in the scanning direction is shown. Finally, four scanning lines are thinned out with m = 1 and n = 4. n = 4 = k × j =
A method of calculating a median value of j = 4 scanning lines as 1 × 4 to generate a thinned image will be described.

As shown in FIG. 3, the sampled image to be subjected to the thinning process is an image of (SH × m pixels) × (SV × n scanning). Since m = 1 for this image, no thinning is performed for m pixels (step 102 in FIG. 1), and since k = 1, simple thinning for k scanning is not performed (step 103 in FIG. 1). Thereafter, median value processing for j = 4 scans is performed (step 104 in FIG. 1), and 1 corresponding to the median value for every 4 scans
The scanning lines are taken out and used as n thinned images. By performing a matching search process with a predetermined template on the thinned image (SH pixels) × (SV scanning) generated in this manner, a high-speed search for a license plate can be realized.

Here, regarding the search method according to the present invention,
The thinning conditions including the above essential conditions are shown below. n: the final thinning interval of the scanning line m: the final thinning interval of the pixel k: the simple thinning interval of the scanning line j: the scanning line width to be subjected to the median value processing of the scanning line k1: the horizontal edge of the convex character in the scanning line direction (up and down) Number of scanning lines at edge) k2: scanning line corresponding to character line o2: pixel corresponding to character line, 1 if the image size after decimation is less than a fraction of (secondary) cache capacity N ≧ 4 2 The number of scanning lines with the minimum line width> final thinning interval K2> n 3 A scanning line having a change in a character edge portion can be finally removed. K ≧ k1 4 A character line does not disappear in a thinned image .... k <k2 , M <o2 5 Sufficient thinning-out and high-speed search can be achieved. > 1 6 Eliminating (0 or 1) scanning lines with level fluctuations and calculating a valid median value ...・ ・ ・ ・ ・ ・ J ≧ 3

With respect to the thinning process used in the above-described method of the present invention, it is possible to suitably use a process by a specially designed hardware or an all-software process by a CPU of a personal computer. Note that processes such as image capturing and image capturing of a license plate are performed using a conventional general image capturing device, image processing device, and the like, and a description thereof will be omitted. Also, the matching search process performed after the image thinning is performed by using the conventional method, and the description is omitted here.

FIG. 7 is a diagram for explaining another embodiment of the present invention. In the example shown in FIG. 7, (SH × m pixels) × (SV
× n scanning) for the pixel thinning
= 2, n = 6 for the scan thinning, and applying the two-scan simple thinning of k = 2 before performing the median value calculation processing between the scanning lines, to obtain the SH pixel × SV scan thinned image. obtain. This is because when the number of scanning lines (k1) having a large pixel level variation at the upper and lower edges of the line of the license plate character is about two scannings, the median value processing of n = 6 scannings may be performed at one time. Although the processing becomes relatively complicated and takes a long time, 2 (= k) scans are thinned out in advance and 3 (=
j) If the median value is calculated from the scan, almost the same result can be obtained at a higher speed.

As shown in FIG. 7, a simple thinning or averaging process for m = 2 pixels is performed (step 10).
2), when simple thinning of k = 2 scans is performed (step 10)
3) In FIG. 7, only the portions indicated by oblique lines or dots are extracted. Thereafter, when a median value calculation process for j = 3 scans is performed on the image (step 104), an SH pixel × SV scan image having a desired accuracy to be searched is generated.

In the method according to the above embodiment, m
In contrast to pixel thinning, which uses only simple thinning or averaging, a relatively complicated process is used for n-scan thinning. Normally, luminance levels fluctuate due to the influence of convex characters on the license plate. This is in view of the fact that it is particularly large at the horizontal edge portion of the convex character. However, the technical scope of the present invention should not be limited to this embodiment. For example, if the luminance level fluctuation of the vertical edge portion of the character is larger according to the characteristics of the sampled image, m pixel thinning is performed. With m = o × p, the median value calculation processing may be performed on the p pixels from the o-pixel simple thinning, and only the simple thinning or the averaging processing may be performed on the n-scan thinning. Or m pixels x
The median value calculation processing may be performed after simple thinning for both n scans.

The simple thinning-out or averaging processing on the edge portion of the line constituting the convex character with less fluctuation of the luminance level (upper / lower edge or left / right edge) includes pipeline processing by hardware and real-time processing by digital signal processor. (Single instruction multi-data) by single instruction multi-processor including, MMX, etc. (trade name)
By performing parallel processing as in the above, it is possible to select processing to perform the entire range, or to perform simple thinning or averaging processing at appropriate pixels and scanning intervals by software.

In each of the embodiments of the present invention described above, image processing of a license plate is described. However, the technical scope of the present invention is not limited to this, and recognition of images other than character images is performed. It is self-evident that the present invention can be applied to such applications.

[0032]

As described above in detail, according to the image processing method of the present invention, m pixels × n
Since the scanning process is performed simply and at high speed, for example, a search for a character image of a license plate can be performed at high speed. If the reason is explained in the case of the embodiment shown in FIGS. 1 and 2, extremely simple thinning processing such as simple thinning or simple averaging is performed in the pixel direction, and simple thinning of k (≧ 1) scanning is performed in the scanning direction. The median value processing between j scan lines is performed for the first time later, and the target of the median value calculation processing, which takes a relatively long time, is reduced in advance. As a result, the total processing time is shortened. Because.

Further, since the influence of the pixel level fluctuation occurring at the edge portion of the convex character constituting the license plate can be eliminated, errors in the search can be reduced. The reason is that in the case of the embodiment shown in FIGS. 1 and 2, scanning line thinning is performed in advance at intervals (k) that are equal to or greater than the number of scanning lines (k1) where pixel level fluctuations occur at the horizontal edge. This is because median value processing is performed between j (≧ 3) scanning lines.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating a method according to an embodiment of the present invention.

FIG. 2 is a flowchart showing a modification of the embodiment shown in FIG. 1;

FIG. 3 is a diagram illustrating an image thinning process according to the embodiment of the present invention.

FIG. 4 is a diagram illustrating a conventional image thinning process.

FIG. 5 is a diagram showing a luminance variation in a license plate scanning direction (vertical direction).

FIG. 6 is a diagram illustrating a conventional median value calculation process.

FIG. 7 is a diagram illustrating an image thinning process according to another embodiment of the present invention.

[Explanation of symbols]

11 Background of license plate (white) 12 Convex character of license plate (green) 15, 16 Brightness level fluctuation portion 101 License plate sampling process 102 Simple thinning-out or m-pixel averaging process 103 Simple thinning-out process at k scanning intervals 104 j Scan median value processing step 105 Matching search step

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 5B029 AA05 BB02 EE00 EE07 5C054 FC01 FC14 FC16 FD07 GB12 HA00 HA26 5H180 AA01 CC04 CC24 DD02 DD10

Claims (8)

[Claims] 1. An image processing method for thinning out n scans on a sampled image having SV × n scan lines (where n ≧ 2) to obtain an image having SV scan lines, wherein said n scan thinning is performed. If the process is n = k × j, k
An image processing method comprising: performing simple thinning for extracting one scan per scan (no thinning when k = 1) and extracting one median value for j scans. 2. An image taken as an image of (SH × m) pixels × (SV × n) scanning (where n ≧ 2) is m pixels ×
In the image processing method of thinning out n scans to make an image of SH pixels × SV scanning, the step of thinning out the m pixels × n scans includes the step of simple thinning out of the m pixels or the average processing of m pixels; = K × j, a step of performing simple thinning to take out one scan per k scans (no thinning out when k = 1), and after the m pixel simple thinning or averaging process and the k scan simple thinning process are completed J about the scanning direction
Retrieving one median value per scan. 3. The image processing method according to claim 1, wherein the image is a character image of a license plate including convex characters, and upper and lower edges of the convex characters in the (SV × n) scan. An image characterized in that, when k1 is the number of scanning lines having a luminance variation corresponding to a portion and k2 is the number of scanning lines related to the line width of the convex character, k1 ≦ k ≦ k2 holds for k. Processing method. 4. The image processing method according to claim 1, wherein j ≧ 3 holds for the j. 5. The image processing method according to claim 1, wherein n = 6,
An image processing method, wherein k = 3 and j = 2. 6. An image processing in which one pixel of a scanning line is SH × m (m ≧ 2) pixels, and m pixels are thinned out to obtain an image in which one scanning line is composed of SH pixels. In the method, the step of thinning out m pixels includes: performing simple thinning for extracting one pixel per o pixel (where no thinning is performed when o = 1) with m = o × p; An image processing method characterized by extracting a value. 7. A sampling image of (SH × m) pixels × (SV × n) scans is subjected to thinning of m pixels × n scans to obtain SH pixels ×
In the image processing method for forming an image of the SV scanning, the step of thinning out the m pixels × n scannings is a simple thinning out of one scanning for every k scannings, where n = k × j (no thinning out when k = 1) Is performed, a step of extracting one median value per j scans, and a step of extracting one pixel for every o pixels with m = o × p (no omission in the case of o = 1) is performed. extracting one median value per p pixels. 8. A search method for a character image or a license plate image in which a thinned image obtained by using the image processing method according to claim 1 is matched with a search template prepared in advance. .