JPS61194580A - Binary coding method for light and shade image - Google Patents

Abstract

PURPOSE:To code into binary number automatically and stably, characters, graphs and objects in light and shade images regardless of size by compensating inappropriate threshold values sought in low contrast part areas in backgrounds with the threshold and weight values obtained. CONSTITUTION:The address controller 103 shall provide the threshold value/ weight value deciding part 104 with address for reading image patterns of designated area parts from the frame memory 102. The deciding part 104 shall assign in sequence area part numbers to the address controller 103 and obtain threshold and weight values for each area part which is given the interpolating processor 105. The processor 105 shall compensate the threshold value based on the threshold and weight values obtained from the weight deciding section 104 and furnishes same to the interpolating processor 106. Processor 106 shall interpolate the threshold value for each area part compensated by the processor 105 and obtain the threshold value for the entire image pattern, which is output to the frame memory 107. The comparator 108 depending on the size comparison of the picture element level corresponding to memory 102, 107, binarizes the picture element patterns and outputs to the frame memory 107 to attain automatic and stable binary coding.

Description

DETAILED DESCRIPTION OF THE INVENTION <Field of Industrial Application> The present invention relates to a method for binarizing grayscale images, which is widely used in the field of image processing such as character recognition.

<Conventional technology> Conventionally, character recognition devices (OCR), etc., target characters placed under uniform illumination conditions with known brightness, so a constant binarization threshold is applied over the entire image. Therefore, binarization was possible. Furthermore, in apparatuses for recognizing objects such as mechanical parts, since binarization is performed using a fixed threshold value, there are restrictions on the colors and illumination conditions of the object and its background. On the other hand, in image processing other than recognition, there is a method (differential binarization method) in which a grayscale image is first differentiated and then binarized using a certain threshold value, which is different from OCR and other methods that handle objects that are close to binary. Alternatively, the threshold value is dynamically changed for each part of the image, and the image is divided into regions, and the local threshold value of the central point of the partial region is determined based on the average value of the brightness or density of the points within the partial region. The threshold value for any point in the image is determined by interpolating four local threshold values around that point.

<Problems to be solved by the invention> However, when it comes to the conventional differential binarization method described above, the entire image is binarized using a fixed threshold value, so the threshold value is set so as not to damage objects such as characters and objects. There was a problem with setting a lot of noise in the background. On the other hand, when using the conventional dynamic binarization method as described above as the first stage of character recognition processing, 1) areas with low contrast, such as character lines that are thicker than the size of the partial area or background areas that do not contain characters, etc. In this case, an inappropriate threshold is given due to the background noise rather than the contrast between the characters and the background.

2) There are problems such as the need to dynamically determine the size of the partial area according to the size of characters, etc. Therefore, with the conventional method, noise occurs in the background part of binary and tanned images, making it difficult to cut out and recognize characters, and the number of partial areas needs to be reset depending on the nature of the grayscale image being handled. There were drawbacks such as: In addition, in the conventional dynamic binarization method, by increasing the size of the partial region, background noise can be suppressed to some extent, but on the other hand, the binarization process becomes stable due to lighting and shadows in the scene. There was a problem with sexuality. Therefore, an object of the present invention is to provide a method for binarizing grayscale images that eliminates the drawbacks of the conventional method.

In order to solve the above-mentioned problems, the present invention provides a method for detecting and recognizing characters and figures in a three-dimensional scene under uneven lighting conditions. , is a dynamic binarization method for differential images that reliably separates characters, figures, etc. from the background regardless of their size, and also reduces background noise as much as possible.

<Operation> In this way, an inappropriate threshold value determined for a low contrast partial area such as a text background is corrected by the threshold value and weight value determined for the surrounding partial area, thereby correcting the inappropriate threshold value. can be determined.

<Example> Fig. 2 shows an example of the procedure for recognizing characters in a scene using the method for binarizing gray scale images of the present invention, in which 1 is the binarization process, 2 is the character extraction process, and 3 is the character recognition process. This is recognition processing. First, in the binarization process 1, a grayscale image i9 is input, and a black and white binary image is output. In character extraction processing 2, the binary image/mother tongue is scanned vertically and horizontally, and -
Cut out each character using its circumscribing rectangle. Next, in character recognition processing 3, the extracted characters are recognized one by one and the results are output.

FIG. 3 shows the detailed processing procedure of binarization processing 1, and 11
12 is a threshold value determination process, 13 is a weight value determination process, 14 is a first interpolation process, 15 is a second interpolation process, and 16 is a binary determination process.

First, the grayscale image G(i, j) is divided into a partial area (the number of the partial area is (m, n)) of a predetermined size (for example, 16 x 16 pixels) in area division processing 11.
divided into This division process is not necessarily the same as shown in Figure 4 (,).
There is no need for equal division as shown in Figure 4(b),
As shown in (c), the regions may overlap each other and may be separated from each other.

In the threshold value determination process 12, a threshold value is determined for each divided partial area based on the appearance frequency of the density value of the pixel within the partial area. The method for determining the threshold value from the frequency of appearance of concentration values is as follows:
Various methods can be considered, such as a simple method that uses the average density value, median value, average value of maximum and minimum values as a threshold, and a method that treats threshold processing as a two-class problem that separates black and white classes. , a method (discriminant analysis method) in which the concentration value with the best degree of separation (maximum inter-class variance) is used as a threshold.

In the weight value determination process 13; divided partial area weight values W
Determine mn. There are, for example, the following four types of weight values Wmn.

[Method a]: wrrln = σmn [Method b]: Wmn = σmn [Method c): Wmn = σ2mn/Amn [Method d]: wmn = σmn/Amn When using variance or standard deviation as a weight value (Method a) ,
b) is a system that determines that the higher the contrast of a partial region, the more likely it is to contain information important for image processing, and gives a higher weight value to the threshold given for that partial region. , is effective when the lighting conditions are relatively good. In addition, when using (normalized) the variance or standard deviation by the average brightness value as the weight value (method C2
d) For example, when character strings with the same contrast are placed under uneven lighting conditions, even if the illumination intensity differs, if the original contrast of the character strings is the same, the weight will also be This is to ensure that the values are the same, and is effective when the brightness of certain parts of characters, etc. is not uniform.

The first interpolation process 14 is a process for correcting the threshold value θmn based on the threshold value θmn obtained in the threshold value determination process 12 and the weight value wmn obtained in the weight value determination process 13. When the threshold after interpolation is θ″mn and the threshold before the first interpolation is θmn, there are two methods given by the following equations.

[Method 1] [Method 2] Here, ωmn−〇: W□ゎ〈φ1 (constant) Wmn: otherwise However, if the weight value is Wmn≧φ2 (φ2 is a constant), this process is not performed.

By repeating this process, the threshold value of the partial area given a large weight value is propagated to the surrounding partial areas, and the threshold value of the small area that does not include characters can suppress the occurrence of voice distortion≦background noise.

As shown in FIG.
The threshold value T (*ej) for (i w j) is the threshold value given to each vertex as θl and θ2.

θ3, θ4, the base and height of the rectangle are L 1 h L
2. When the horizontal and vertical distances from the upper right vertex of the rectangle to the pixel P (i, j) are α and β, they are given by the following equation.

T(i,j)-θl・(Ll-α)(L2-β)/L1
'L2+θ2・(Ll-α)・β/L1'-L2+θ3
・α・(L2-β)/Ll-L2+θ4・α・β/L
1/L2 In the binary determination process 16, the binary i9tan B(i, j ).

B(i,j)=0:G(i,j)'<T(i,j)1:
G(i, j)≧T(i, j) FIG. 1 shows an example of the hardware configuration for realizing the binarization process 1, in which 101 is a common control unit, 102 is a frame memory (1), 103 is an address control unit, 104 is a threshold 7 weight value determining unit, 105 is an interpolation processing unit (1), 106
is an interpolation processing unit (II), and 107 is a frame memory (2).
, 108 is a comparison circuit. In the figure, each section operates according to a timing signal given from a common control section 101. First, the frame memory (1) 102 is a memory for one screen of grayscale images that stores grayscale images, and an image i4 is inputted via a television camera or the like (not shown).

The address control unit 103 includes a frame memo IJ (1)
From 102, an address for reading out the image/guttan of the specified partial area is given to the threshold value 7 weight determining unit 104.

Threshold 7 Weight value determination unit 104 sequentially gives partial area numbers to address control unit 103 , determines a threshold value and weight value for each partial area, and provides these to interpolation processing unit (I) 105 . The interpolation processing unit (I) 105 corrects the threshold value from the threshold value and weight value obtained from the threshold value/cover value determination unit 104, and provides the corrected value to the interpolation processing unit (II) i 06. The interpolation processing unit (If) 106 interpolates the threshold values for each partial region corrected by the interpolation processing unit (I) to obtain a threshold value for the entire image/gut, and outputs this to the frame memory (2) 107. The comparison circuit 108 includes a frame memory (1) 10
The grayscale image pattern is binarized by comparing the levels of corresponding pixels in frame memory (2) 107 and frame memory (2) 107, and the result is output to frame memory (2) 107.

<Effects of the Invention> As described above, in the present invention, an inappropriate threshold value determined for a low-contrast partial area such as a character background can be replaced by a threshold value and a weight value determined for the surrounding partial area. An appropriate threshold value is determined by correcting the following.

Therefore, according to the present invention, as a preprocessing for region extraction processing for recognizing characters, figures, objects, etc. in a scene, characters, figures, objects, etc. in a gray scale image are automatically and stably detected regardless of their size. can be binarized into

[Brief explanation of the drawing]

Fig. 1 shows an example of the configuration of the binarization processing software, which shows an embodiment of the method for binarizing grayscale images according to the present invention, Fig. 2 shows the procedure for recognizing characters in a scene, and Fig. 3 shows the processing procedure for recognizing characters in a scene. The detailed procedure of the value processing, FIGS. 4(a), 4(b), and 4(e) are examples of partial region division, and FIG. 5 is an explanatory diagram of the second interpolation processing. 1...Binarization processing, 2...Character cutout processing, 3.
...Character recognition processing, 11...Region division processing, 12...
・Threshold value determination process, 13... Weight value determination process, 14...
・First interpolation processing, 15...Second interpolation processing, 16.
...Binary judgment processing, '101...Common control unit, 10
2...Frame memory (1), 103...Address control unit, 104...Threshold value 7 weight determination unit, 105...
Interpolation processing unit (I), 106... interpolation processing unit (II),
107... Frame memory (2), 108... Comparison circuit.

Claims (5)

[Claims] (1) Determine a local threshold for each partial region of the grayscale image based on the frequency of appearance of pixel density within the partial region, and perform interpolation processing on the obtained local threshold. In the binarization method, a threshold value for the entire grayscale image is obtained by calculating the threshold value for the entire grayscale image, and a binary image is obtained based on the magnitude of the threshold value for the entire grayscale image and the value of the grayscale image. Means for processing a grayscale image to obtain a weight value for each partial region; a means for determining a local threshold for the center within the partial region; A first interpolation means determined by a weighted average based on weight values for partial regions adjacent to the region, and a local threshold corrected by the processing by the first interpolation means or its repeated processing, and all of the grayscale images are A method for binarizing a grayscale image, comprising a second interpolation means for determining a threshold value for a pixel. (2) The weight value is given by the standard deviation of the brightness of the grayscale image within the partial area.
1) Binarization method for grayscale images described in section 1). (3) The weight value is S_d/A, where A_v is the average brightness value of the grayscale image in the partial area and S_d is the standard deviation.
Claim No. (1) characterized in that it is given by _v.
2. Binarization method for grayscale images described in Section 2. (4) The first interpolation means is m-th in the horizontal direction and n-th in the vertical direction.
The weight value for the th partial region (m, n) is set as W_m_n
, when the threshold values before and after correction by the first interpolation means are θ_m_n and θ′_m_n, respectively, θ′_
m_n=Σ^m^+^1_p_=__m_-_1Σ^n^
+^1_q_=_n_-_1W_p_q・θ_p_q/
Σ^m^+^1_p_=__m_-_1Σ^n^+^1_
Claim (1) characterized in that the process q_=_n_-_1W_p_q is performed once or repeatedly.
2. The method for binarizing a grayscale image as described in Section 1, Section (2), or Section (3). (5) The first interpolation means is m-th in the horizontal direction and n-th in the vertical direction.
The weight for the th partial region (m, n) is W_m_n,
When φ1 and φ2 are predetermined parameters such as φ1≦φ2, if W_m_n≦φ1, replace W_m_n=0 and perform interpolation processing on the partial area, and φ2<W_
If m_n, interpolation processing is not performed on the partial area.
), or (3), or (4), the method for binarizing a grayscale image.