JPH07239911A - Binarizing processing method - Google Patents

Abstract

PURPOSE:To set a threshold level by which an object is clearly separated from its background by obtaining a density level that maximizes an inter-class dispersion within a prescribed density level range when density distribution is divided into two classes and setting the density level as a threshold level used to separate the object from the background. CONSTITUTION:A cast iron piece 2 being an object is picked up by a CCD camera 18, and its resulting image is given to an A/D converter section 22, and its output is stored tentatively in a picture memory 24. A character segmentation section 26 specifies a position of each character based on the stored picture signal and a density distribution measurement circuit 28 measures the density distribution of each specified character within a range of a circumscribed rectangle shape. Then, a threshold level calculation circuit 32 calculates a threshold level used to separate the character from its background by using inter-class dispersion. In this calculation, the density level maximizing the inter-class dispersion is obtained within a prescribed density level range stored in a 1st picture memory 34 and a binarization processing section 36 realizes a stable binary image.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a binarization processing method for converting an image obtained by picking up an object into a binary image.

[0002]

2. Description of the Related Art When performing image processing, it is necessary to separate the image into an object and a background as a process before extracting the features of the image. When performing such image separation, a threshold value is given to the density level and the image is binarized. Various methods have been proposed so far for this binarization processing technique. For example, there are a fixed binarization method, a modal method, a peak method, a discriminant analysis method, and a P-tile method.

The fixed binarization method is effective when the object and the background can be separated at a certain fixed density level, and the image is binarized by a fixed threshold value. Also,
The modal method or the peak method is a bimodal peak whose density distribution has one peak near the average density level of the object and one peak near the average density level of the background even if the threshold value fluctuates due to the influence of lighting. This method is effective when it is possible to separate the object from the background, such as the shape distribution.
The modal method sets a threshold at the concentration level corresponding to the bottom of the valley between the two peaks. On the other hand, in the peak method,
Set a threshold based on two peaks.

The discriminant analysis method is an effective method when the peaks in the concentration distribution overlap to some extent and a clear peak cannot be obtained between the object and the background. In this discriminant analysis method, the threshold is set by using the interclass variance. As the evaluation formula for class dispersion, Tatsuyuki Otsu and J. Kittler are famous. Furthermore, P
The tile method is an effective method when the area ratio between the object and the background is known in advance, and the threshold value is set using this area ratio.

[0005]

By the way, for example, a character hot-marked on a slab has a lot of blurring or blurring of printing, and noise is also present on the surface of the slab due to adhesion of scale or the like. In the case of recognizing an image of a character with such blurring or blurring, or an image with noise, each of the above binarization processing methods has the following problems.

In the fixed binarization method, if the preset threshold value is included in the range of the density distribution of the noise portion, the influence of noise remains and a stable binary image cannot be obtained. In the peak method and mode method, the peaks and valleys of the density distribution of the character and the background may overlap due to blurring and blurring, and the peak points and valleys of the peaks may shift from their original positions due to noise. Therefore, a stable binary image cannot be obtained. Furthermore, in the discriminant analysis method, if there is noise, or the character is blurred or blurred, the threshold value tends to be set higher, so part of the character may be missing, A good binary image cannot be obtained. On the other hand, in the P-tile method, a good binary image can be obtained for a character that matches the preset area ratio of the character and the background, but for a character that does not match the set area ratio, However, noise may occur or some characters may be cut off, and this method also has limitations in use.

The present invention has been made based on the above circumstances, and sets a threshold value that can clearly separate an object from the background even if there is noise or the object has blurring or blurring. It is an object of the present invention to provide a binarization processing method capable of performing the above.

[0008]

In order to achieve the above-mentioned object, a binarization processing method according to the invention of claim 1 is a method of circumscribing an object based on an image obtained by imaging the object. After measuring the concentration distribution within a rectangular range, when the concentration distribution is divided into two classes, the concentration level that maximizes the interclass dispersion is determined within a predetermined concentration level range, and the concentration level is the object. It is characterized in that the image is converted into a binary image by setting it as a threshold value for separating the background and the background.

According to a second aspect of the present invention, in the binarization processing method according to the first aspect of the present invention, the density level range is set to each of the target objects when the type of the target object is known. It is characterized in that it is set according to the ratio of the area to the background.

In the image processing apparatus according to the third aspect of the present invention, the clipping means for pinpointing the position of the object based on the image obtained by picking up the object and the position pinpointed by the clipping means. A concentration distribution measuring means for measuring the concentration distribution within the range of the circumscribed rectangle of the object, and a concentration level that maximizes interclass dispersion when the concentration distribution is divided into two classes within a predetermined concentration level range. Threshold calculation means for determining the density level as a threshold for separating the object and background, and binarization processing means for converting the image into a binary image based on the threshold. It is characterized by including.

[0011]

According to the invention described in claim 1, the density level which maximizes the interclass dispersion when the density distribution is divided into two classes is obtained within the predetermined density level range by the above construction,
By setting that density level as a threshold value that separates the object and background, the threshold value is set to a density level that deviates significantly from the density level at which the object and background are actually separated by the influence of noise. , The threshold can be set near the density level where the character and background are actually separated. Therefore, it is possible to set a threshold value that can clearly separate the object from the background even if the image has poor print quality such as noise, blurring, and blurring, and a stable binary image can be obtained. .

According to the invention described in claim 2, the density level range is actually set by setting the density level range according to the ratio of the area of each object and the background when the kind of the object is known. It is possible to accurately select the range of the concentration level in which the character and the background are separated.

According to the third aspect of the present invention, the density level which maximizes the interclass dispersion when the density distribution is divided into two classes is obtained within the predetermined density level range by the above configuration, and the density level is targeted. By providing a threshold value calculation means for setting the threshold value for separating the object and the background, the density level deviated significantly from the density level at which the object and the background are actually separated by the influence of noise etc. It is possible to eliminate the possibility that a threshold value will be set for the threshold value and set the threshold value in the vicinity of the density level where the character and the background are actually separated. Therefore, a stable binary image that is not significantly affected by noise or the like can be obtained, so that the image can be accurately identified based on the binary image.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. 1 and 2 are views for explaining a binarization processing method according to an embodiment of the present invention, and FIG. 3 is a schematic configuration diagram of an image processing apparatus using the binarization processing method.

The image processing apparatus shown in FIG.
A lighting device 14 and a CCD camera 1 as an imaging means.
8, an A / D conversion unit 22, an image memory 24, a character cutout unit 26, a density distribution measurement circuit 28, a threshold value calculation circuit 32, a first image processing memory 34, and a binarization process. It includes a unit 36, a feature extraction unit 38, an image recognition unit 42, and a second image processing memory 44. In the present embodiment, the case where the characters marked on the slab 2 are recognized hot will be considered. Here, such characters often have a blur or blurring of the print, and moreover, noise is often present on the surface of the slab 2 due to adhesion of scale or the like. Also,
It is assumed that the characters to be printed are ten types of numbers from "0" to "9". The slab 2 is conveyed by a conveyer (not shown) such as a conveyor.

In the sensor 12, the cast slab 2 to be conveyed is a CCD.
It detects that the camera has approached the camera 18 and has reached the imaging position, and outputs a detection signal. The lighting device 14 is
It is a light source that irradiates the slab 2 and the printed portion of the slab 2 is imaged under this illumination.

The CCD camera 18 forms an optical image of the printed portion of the slab 2 as a two-dimensional image on the image pickup element by the image pickup lens, performs photoelectric conversion on the image pickup element, and charges corresponding to the optical image. Is output as a one-dimensional electric signal. The one-dimensional electric signal obtained from the image sensor is output to the A / D conversion unit 22 as an image signal. Also, A / D
The conversion unit 22 converts the image signal output from the CCD camera 18 into a digital signal and outputs the digital signal. The image memory 24 temporarily stores the image signal output from the A / D converter 22, and captures the image signal at the timing at which the detection signal from the sensor 12 is captured.

The character slicing section 26 specifies the position of each character based on the image signal stored in the image memory 24. For example, the character clipping unit 26 first collects the number of pixels whose density level is equal to or higher than a predetermined value, and obtains a projection histogram by projecting this in the x direction and the y direction. Then, in the projection histogram projected in the y direction, the number of lines of the printed character can be known by checking the number of peaks, and by checking the coordinates of both ends of each peak, the y coordinate of the range in which each line of the character exists can be obtained. Further, in the projection histogram projected in the x direction, the number of print characters can be known by checking the number of peaks, and by checking the coordinates of both ends of each peak, the x coordinate of the range in which each character exists can be obtained. In this way, the position of each character can be specified.

The density distribution measuring circuit 28 measures the density distribution of each character whose position is specified by the character slicing section 26 within the bounding rectangle. The threshold calculation circuit 32 divides the density distribution within the circumscribed rectangle of each character obtained by the density distribution measurement circuit 28 into two classes, and sets a predetermined density level that maximizes the interclass variance. It is obtained within the concentration level range. Then, the density level thus obtained is set as a threshold value for separating the character and the background. The density level is divided into gradations from 0 to 255. The lower the density level, the blacker the image, and the higher the density level, the whiter the image. Further, in the input image of this embodiment, the background is black and the characters are white.

The first image processing memory 34 stores a density level range used when the threshold calculation circuit 32 obtains a density level that maximizes the interclass dispersion. This density level range specifies the vicinity of the density level where characters and background are actually separated. Usually, since the type of characters to be printed is often known, this density level range is set according to the ratio of the area of the character to the area of the circumscribing rectangle for each character. Here, of the ratio of the area of the character to the area of the circumscribed rectangle, the largest ω _{S max} and the smallest ω _{S min} are stored. For example, in this embodiment, "0" to "9"
Among the numbers up to, the ratio of the area of the number to the area of the circumscribed rectangle is “8”, which is about 55%. The smallest value is "1", and the ratio is about 35%. Therefore, ω
_{S max} is 55/100 and ω _{S min} is 35/100
Becomes However, in general, the values of ω _{S min} and ω _{S max} are
It also varies depending on the thickness of the printed characters. Further, the first image processing memory 34 stores the set value σ of the interclass dispersion.
I remember _max ² . Here, the initial setting value σ _max ² of the interclass dispersion is set to zero.

The binarization processing unit 36 includes a threshold calculation circuit 3
The image signal is binarized for each character based on the threshold value obtained in 2 and converted into a binary image. The feature extraction unit 38 is obtained by the binarization processing unit 36. The feature amount is extracted from the binary image. Further, the second image processing memory 44 stores a feature amount serving as a predetermined reference for each character known to be printed in advance, and the image recognition unit 42 extracts the feature amount from the feature extraction unit 38. Characters are recognized by comparing the generated feature amount with the reference feature amount stored in the second image processing memory 44.

Next, the operation of the image processing apparatus of this embodiment will be described. First, when the slab 2 approaches the CCD camera 18, the sensor 12 detects the slab 2 and outputs a detection signal, and the CCD camera 18 displays the characters printed on the slab 2 under the illumination of the illumination device 14. Take an image. The image signal output from the CCD camera 18 is converted into a digital image signal by the A / D converter 22, and then the image memory 24
Are taken into the image memory 24 at the timing of taking in the detection signal from the sensor 12.

The character slicing unit 26 specifies the position of each character based on the image signal stored in the image memory 24. Then, in the density distribution measuring circuit 28, the density distribution is measured in the range of the circumscribing rectangle for each character whose position is specified by the character cutout unit 26. As an example, it is assumed that the density distribution shown in FIG. 1D is measured for an image including noise. In this case, the background has a density distribution as shown in FIG. 1A, the characters have a density distribution as shown in FIG. 1B, and the noise has a density distribution as shown in FIG. 1C. it is conceivable that. Here, in FIGS. 1A to 1D, the horizontal axis represents the density level and the vertical axis represents the pixel frequency.

Next, in the threshold calculation circuit 32, the threshold for separating the character from the background is calculated by using the interclass dispersion. Now, interclass distribution will be described. First, as shown in FIG. 1 (d), at a certain concentration level k, the concentration level the concentration distribution of k following classes C ₀ and the concentration levels divided into k + 1 or more classes C ₁ and. Performing such classification corresponds to regarding the background as class C ₀ and the character as class C ₁ . In this embodiment, the case of using the evaluation formula of interclass dispersion by Tatsuyuki Otsu will be described as an example. The evaluation formula for the inter-class variance σ _B ² (k) by Tatsuyuki Otsu is σ _B ² (k) = ω ₀ (μ ₀ −μ _T ) ² + ω ₁ (μ ₁ −μ _T ) ²・ ・ ・ ( 1) is given in. Here, ω ₀ is the ratio of the number of pixels included in the class C ₀ to the total number of pixels, ω ₁ is the ratio of the number of pixels included in the class C ₁ to the total number of pixels, and μ ₀ is the class C ₀
, The average concentration level in μC, μ ₁ is the average concentration level in class C ₁ , and μ _T is the overall average concentration level. That is, assuming that the number of pixels of the density level i is n _i and the total number of pixels is N, the ratio of the number of pixels of the density level i P _i = n _i / N is used, and ω ₀ , ω ₁ , μ ₀ , μ ₁ And μ _T is

[Equation 1]Is expressed as This class variance σ_B ²(k) is the overall average
Concentration level μ_TTwo Class C for₀, C₁Rose
It represents the degree of tsuki. Interclass variance σ_B ²(k) is
The smaller, the two classes C₀, C₁Is the overall average
Concentration level μ_TIt is thought to be distributed near the
, Interclass variance σ_B ²The larger (k), the more
Russ C₀, C₁Is the overall average concentration level μ_TAway from
It is thought that it is distributed here.

In this embodiment, the density level that maximizes the interclass variance σ _B ² (k) is determined within the predetermined density level range stored in the first image processing memory 34. That is,
In the threshold value calculation circuit 32, the density level k _C that maximizes the inter-class variance σ _B ² (k) given by the equation (1) is given by ω _{S min} ≦ ω ₁ ≦ ω _{S max.} ) Is solved under the condition. For example, when the interclass variance σ _B ² (k) is calculated for the density distribution in which noise is present in the image as shown in FIG. 1D, the distribution as shown in FIG. 1E is obtained. In this case, in the conventional discriminant analysis method, the threshold value is set to the density level k _T which is deviated in a direction larger than the density level at which the character and the background are actually separated. However, in the binarization processing method of the present embodiment, the density level that maximizes the inter-class variance σ _B ² (k) is obtained in the predetermined density level range δk corresponding to the equation (2).
The threshold value can be set to the density level k _C by eliminating the possibility that the threshold value is set to the density level k _T due to the influence of noise or the like.

Next, the operation of the threshold value calculation circuit 32 will be described with reference to the flowchart shown in FIG. First, μ _T is calculated based on the concentration distribution measured by the concentration distribution measuring circuit 28 (step 2). Then, the density level k for classifying is set from 0 to 255 in order (st
ep 4). In step 6, it is determined whether or not the set density level k is 255 or less. When the density level k is 255 or less, the process proceeds to step 8. First, after calculating the omega ₀ relative density level k = 0 (step 8), the omega ₀
Is used to calculate ω ₁ from ω ₁ = 1−ω ₀ (step 12).
Next, it is determined whether this ω ₁ satisfies the expression (2) (step 14). If ω ₁ does not satisfy equation (2), st
Return to ep 4. On the other hand, if ω ₁ satisfies the equation (2),
Calculate μ ₀ and μ ₁ (step 16). Then, using ω ₀ , ω ₁ , μ ₀ , μ ₁ and μ _T calculated so far,
Calculate the interclass variance σ _B ² (k) by using equation (1) (step
18).

Next, the interclass variance σ _B ² (k) is equal to or greater than the set value σ _max ² (the initial set value is zero) of the _interclass variance stored in the first image processing memory 34. It is determined whether or not there is (step 22). If the interclass variance σ _B ² (k) is smaller than the set value σ _max ² , the process returns to step 4. On the other hand, if the interclass variance σ _B ² (k) is greater than or equal to the set value σ _max ² , k at this time is regarded as the optimal threshold value at this stage (step 26), and the set value σ _{max of the} interclass variance is set. ² is updated to this interclass variance σ _B ² (k). The updated set value σ _max ² = σ _B ² (k) of the inter-class dispersion is stored in the first image processing memory 34 (step 28) and the process returns to step 4. Next, in step 4, the density distribution is divided into two classes at a density level that is one higher than the density level that has just been set. Then, the same processing as described above is performed, and the processing is repeated until the density level k = 255 (step 6), and the flowchart of FIG. 1 ends. Thus, finally,
The maximum value σ _max ² of the _interclass variance σ _B ² (k) is obtained within a predetermined density level range, and the density level k _C that gives this maximum value is selected as the threshold value for separating the character and the background. .

Thereafter, the binarization processing unit 36 binarizes the image information stored in the image memory 24 for each character based on the threshold value obtained by the threshold value calculation circuit 32, Get the value image. Then, the feature extraction unit 38 extracts the feature amount from the binarized image information, and the image recognition unit 42.
In, the character recognition is performed by comparing the feature amount with the corresponding reference feature amount stored in the second image processing memory 44. Then, the recognition result is output to a display device (not shown) or the like.

Next, an example of an image processed by the binarization processing method of this embodiment will be described. FIG. 4A is a diagram showing an original image, and FIG. 4B is a diagram showing a binary image obtained by processing the original image of FIG. 4A by a conventional discriminant analysis method. )
FIG. 8 is a diagram showing a binary image obtained by processing the original image of (a) by the binarization processing method of the present embodiment. Here, three marking characters “3”, “1”, and “8” are printed on the slab, which may be difficult to understand in the original image schematically drawn in FIG. The characters have blurring and blurring. When the binarization processing is performed by the conventional discriminant analysis method, a sufficiently recognizable binary image cannot be obtained as shown in FIG. On the other hand, when the binarization process is performed by the method of the present embodiment, it is possible to set a threshold value that allows characters to be clearly separated from the background even in an image with poor print quality such as blurring and blurring. Figure 4 (c)
As shown in, a stable binary image can be obtained.

In the binarization processing method of this embodiment, when the density distribution is divided into two classes, the density level that maximizes the interclass dispersion is determined within a predetermined density level range, and the density level is expressed as a character. By setting the threshold value to separate the background from the background, it eliminates the possibility that the threshold value is set to a density level that deviates significantly from the density level where the characters and background are actually separated due to the influence of noise etc. However, the threshold value can be set in the vicinity of the density level where the character and the background are actually separated. Therefore, both of the advantages of the conventional discriminant analysis method and the P-tile method are taken in, and even in an image with noise or an image with poor print quality such as blurring or blurring, the character can be clearly separated from the background. The value can be set and a stable binary image can be obtained. Further, in the image processing apparatus that sets a threshold value by using the binarization processing method of the present embodiment and performs image processing using this threshold value, a stable binary image that is not significantly affected by noise or the like is obtained. Therefore, it is possible to accurately identify the image based on the binary image.

The present invention is not limited to the above embodiments, but various modifications can be made within the scope of the gist thereof. In the above embodiment, the case where the evaluation formula by Mr. Tatsuyuki Otsu shown in the formula (1) is used as the interclass dispersion has been described.
The evaluation formula E (k) E (k) = ω ₀ log (σ ₀ / ω ₀ ) + ω ₁ log (σ ₁ / ω ₁ ) by Kittler may be used. Here, ω ₀ and ω are similar to those in Tatsuyuki Otsu's evaluation formula, σ ₀ represents the variance in class C ₀ , and σ ₁ represents the variance in class C ₁ . Also in this case, the concentration level k _C at which the interclass variance E (k) is maximized
_{Is obtained} under the condition of ω _{S min} ≤ω ₁ ≤ω _{S max} .

In the above embodiment, the case of identifying a number has been described. However, for example, a character other than a number, a figure, etc., in which the ratio of the area of the object to the area of the background is known. It is possible to apply the binarization processing method of this embodiment also when binarizing an image of an object.

[0033]

As described above, according to the invention of claim 1, when the concentration distribution is divided into two classes, the concentration level that maximizes the interclass dispersion is determined within a predetermined concentration level range, By setting the density level as a threshold value for separating the object from the background, it is possible to clearly separate the object from the background even if the image has bad print quality such as noise or blurring or blurring. A threshold value can be set, and a binarization processing method capable of obtaining a stable binary image can be provided.

According to the second aspect of the present invention, when the density level range is set according to the ratio of the area between each object and the background when the kind of the object is known, the character level is actually set. It is possible to provide a binarization processing method capable of accurately selecting a range of concentration levels in which the background and the background are separated.

According to the third aspect of the present invention, when the concentration distribution is divided into two classes, the concentration level that maximizes the interclass dispersion is determined within a predetermined concentration level range, and the concentration level is set as the object. By providing the threshold value calculating means for setting the threshold value for separating the background, it is possible to obtain a stable binary image that is not significantly affected by noise and the like. Therefore, image identification based on the binary image is performed. It is possible to provide image processing leaving that can accurately perform the above.

[Brief description of drawings]

FIG. 1 is a diagram for explaining a binarization processing method that is an embodiment of the present invention.

FIG. 2 is a diagram for explaining a binarization processing method that is an embodiment of the present invention.

FIG. 3 is a schematic configuration diagram of an image processing apparatus using the binarization processing method.

4A is a diagram showing an original image, FIG. 4B is a diagram showing a binary image obtained by processing the original image of FIG. 4A by a conventional discriminant analysis method, and FIG. FIG. 7 is a diagram showing a binary image obtained by processing the original image of FIG. 4) by the binarization processing method according to the embodiment of the present invention.

[Explanation of symbols]

 2 Slab 12 Sensor 14 Illuminator 18 CCD Camera 22 A / D Converter 24 Image Memory 26 Character Cutout 28 Concentration Distribution Measuring Circuit 32 Threshold Calculation Circuit 34 First Image Processing Memory 36 Binarization Processing 38 38 Features Extraction unit 42 Image recognition unit 44 Second image processing memory

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification number Reference number within the agency FI Technical indication location H04N 1/403 H04N 1/40 103 A (72) Inventor Keiki Takayama 1 Kimitsu, Kimitsu-shi, Chiba New Kimitsu Works, Nippon Steel Co., Ltd.

Claims (3)

[Claims] 1. A class-to-class variance when the concentration distribution is divided into two classes after measuring the concentration distribution in the range of a circumscribed rectangle of the object based on an image obtained by imaging the object. Is obtained within a predetermined density level range, and the image is converted into a binary image by setting the density level as a threshold value for separating the object from the background. Binarization processing method. 2. The density level range is set according to a ratio of an area between each of the objects and the background when the type of the object is known. Binarization method. 3. A cutout unit for specifying the position of the target object based on an image obtained by imaging the target object, and a density within a circumscribed rectangle of the target object whose position is specified by the cutout unit. A concentration distribution measuring means for measuring the distribution, and a concentration level which maximizes the interclass variance when the concentration distribution is divided into two classes, is obtained within a predetermined concentration level range, and the concentration level is set to the object and the background. An image processing apparatus comprising: a threshold value calculation unit that sets a threshold value for separating the image and a threshold value; and a binarization processing unit that converts the image into a binary image based on the threshold value.