JPH10285399A - Binarization method for image - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the identification performance of a character part by preventing a background part close to the character part from being binarized. SOLUTION: A CPU 24 divides received image data into a plurality of partial areas and generates a density distribution histogram for each partial area. Then a weight is provided to a partial area around a noted partial area, and a weight is multiplied with each density distribution histogram of the partial areas and the products are summed to obtain a weighted density distribution histogram of the noted partial area. Next, based on the weighted density distribution histogram, a binarization threshold level of the noted partial area is set, and the density is binarized with respect to each pixel in the noted partial area.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for binarizing an image, and more particularly to a method for binarizing an image in which only a desired character portion or the like can be satisfactorily extracted from an image taken outdoors or the like.

[0002]

2. Description of the Related Art Compared with recognizing characters written on white paper, recognizing characters in an image captured by a camera or the like outdoors is based on the fact that the brightness of the entire image changes depending on weather and time. In addition to this, the brightness of the background is not uniform, which is difficult. As such preprocessing for character recognition, it is necessary to binarize the image density. As one of the methods, a so-called dynamic binarization method for setting a threshold value for each image is known. However, simply setting a unique threshold value for the entire image cannot cope with a case where the brightness varies in each part of one image due to the influence of a shadow, a background, or the like. Therefore, in order to further remove the fluctuation of brightness and the influence of the background, a method of dividing an image into several partial regions and setting an appropriate binarization threshold for each partial region is known. As a dynamic binarization method for determining a threshold value of each partial area, a discriminant analysis method (“automatic threshold value selection method based on discrimination and least square criterion”, IEICE Transactions, J63-D (4), p349
To p356, 1980).

However, in the above discriminant analysis method, for example, as shown in FIG. 8A, with a large partial area E, a very bright background portion B1 has a character portion L representing "7".
, The partial area E including the background portion B1
Since the binarization threshold of is increased, as shown in FIG.
A part of the character portion L darker than the background portion B1 may be missing during binarization. Therefore, if the partial area E is reduced as shown in FIG.
1 can be prevented, but this time, for example, if there is a background portion B2 having a slightly darker intermediate density near the character portion L, the background portion B2 Since the binarization threshold value is low in the partial region E including the character portion L, as shown in FIG.
(Hereinafter, referred to as a binary shape), the shape of the character portion L is deformed, and the identification becomes difficult.

Accordingly, for example, Japanese Patent Application Laid-Open No. 61-194580.
In (2), the binarization threshold of the partial area of interest is determined as a weighted average of the binarization thresholds of the surrounding partial areas, and the binarization threshold of the high-contrast partial area including the character portion is set around the threshold. A binarization method has been proposed in which a background portion adjacent to a character portion is prevented from being binarized into a shape continuous with the character portion by propagating the character portion.

[0005]

However, with the above-described binarization method, it is not possible to sufficiently prevent the background portion close to the character portion from being formed into a binary shape, and the noise-like background generated near the character portion has not yet been achieved. There has been a demand for a method capable of removing the part more reliably.

Therefore, the present invention solves such a problem, and more reliably prevents a background portion close to a character portion or the like from being formed into a binary shape, thereby improving the distinguishability of the character portion or the like. It is an object of the present invention to provide an image binarization method that can be improved.

[0007]

In order to achieve the above object, an image binarization method according to the present invention comprises the steps of: (1) dividing an obtained image into a plurality of partial areas; Creating a density distribution histogram for each region;
(3) assigning a weight value to the partial area of interest and at least a predetermined number of partial areas before and after the partial area; and (4) multiplying the density distribution histogram of each partial area by a weight value and adding the weighted histogram. Obtaining a weighted density distribution histogram of the area; (5) setting a binarization threshold value of the partial area of interest based on the weighted density distribution histogram; and (6) setting the binarization threshold value based on the binarization threshold value. Performing a binarization of the density for each pixel in the partial area to be performed, and repeating the steps (3) to (6) for all the partial areas in the image.

According to the method of the present invention, a weighted density distribution histogram multiplied by a weight value is obtained for a partial area of interest, and a binarization threshold is set based on the weighted density distribution histogram. Therefore, if there is a partial area of interest in the vicinity of a partial area where a character portion or the like exists and presents a density distribution histogram with high bimodality, the weighted density distribution histogram of the partial area of interest also has high bimodal density. And the binarization threshold value set based on the weighted density distribution histogram shifts to the higher density side. As a result, even if the partial area of interest has an intermediate density background portion, it is removed in the binarization process, and as a result, the background portion close to the character portion or the like is prevented from being binarized, and Reliable identification of parts and the like becomes possible.

In order to more reliably prevent the background portion from being binarized, a step is further provided to propagate the binarization threshold value in another partial region having a high evaluation value to the partial region of interest. be able to. Also, in order to prevent the binarization threshold from suddenly changing at the boundary of each partial area,
A step of setting a binarization threshold for each pixel by interpolation or the like may be further provided.

[0010]

FIG. 1 shows the configuration of an apparatus for carrying out the method of the present invention. In the figure, a TV camera 1
The image captured by is input to the A / D converter 21 of the character identification device 2 including the image binarization unit, converted into digital data, and stored in the subsequent frame memory 22. The image data stored in the frame memory 22 is taken in by the CPU 24 via the input interface 23, and is binarized by an image binarization unit realized as software by a process described later. The output interface 25 is identified by a character identification unit also realized by software.
Is displayed on the display device 3 through

The image stored in the frame memory 22 is, for example, as shown in FIG. 9 (A) which has already been described in the prior art, and in addition to the character portion L of "7", a very bright background portion B1 and the like. An intermediate density background portion B2 slightly darker than the character portion L exists near the character portion L. In order to reliably identify the character portion L, it is necessary to prevent the background portion B2 adjacent to the character portion L from being formed into a binary shape.

Therefore, the image binarization unit performs the binarization processing of the image data according to the procedure shown in the flowchart of FIG. That is, in step 101, as shown in FIG. 9A, the image is divided into a plurality of rectangular partial regions E including a plurality of pixels. Then, in a subsequent step 102, a density distribution histogram is created for each partial area E. For example, the density distribution histograms of the partial areas m, m + 1, and m-1 in FIG. 9A are as shown in FIGS. 3A, 3B, and 3C, respectively. In this state, when the binarization threshold th1 is obtained from the density distribution histogram of the partial region m by a discriminant analysis method described later, the intermediate density portion indicated by oblique lines in the drawing becomes larger than the binarization threshold th1. As shown in FIG. 9B described above, the background portion B2 becomes the “H” level and is formed into a binary shape. The partial region E may have, for example, a honeycomb shape (hexagonal shape) other than the above-described square shape.

Therefore, in the present embodiment, in step 103 of FIG. 2, a weight value is assigned to the partial area m of interest and the predetermined number of partial areas including the partial areas m + 1 and m-1 before and after this. In step 104, the density distribution histogram of each partial area is multiplied by a weight value, and these are added.
A weighted density distribution histogram of the partial area m as shown in FIG. 4 is obtained. Then, in step 105, the threshold value th2 of the weighted density distribution histogram is calculated by the discriminant analysis method, and this is adopted as the binarization threshold value of the partial region m. The processes in steps 102 to 105 are all sequentially performed on each partial region in the image. FIG. 5 shows step 1
An example of weighting in 03 is shown. In the figure, with respect to the focused partial area (m, n), a weight value of 0.05 to 0.8 is assigned to the surrounding 24 partial areas that spread two-dimensionally. Note that these weight values and the number of surrounding partial regions are experimentally determined.

The processing for obtaining the weighted density distribution histogram in step 104 and the processing for obtaining the binarization threshold th2 in step 105 are described in order to facilitate understanding.
Partial area m of interest and partial areas m + 1 and m before and after this
An example in the case of performing one-dimensionally with respect to −1 will be described below. In this example, a weight value of 1 is assigned to the partial area m, and a weight value of 0.5 is assigned to the partial areas m + 1 and m−1. The weighted density distribution histogram in this case is shown in FIG.
FIGS. 3B and 3C show the density distribution histogram of FIG.
The density distribution histogram shown by the broken line in FIG.
As shown by the solid line. The discriminant analysis method for obtaining the binarization threshold th2 is based on the bimodal histogram shown in FIG.
When a class is divided into two classes (threshold or less and th + 1 or more) at a certain threshold th, a threshold th that maximizes the inter-class variance is obtained. The threshold value is th2. As is apparent from FIG. 4, the binarization threshold th2 is obtained by weighting the density distribution histogram of the partial region m + 1 in which the high density character portion shows a large contrast with respect to the background.
The binarization threshold value th1 obtained from the density distribution histogram of only the partial area m is largely shifted to the high density side.

Therefore, as shown in FIG. 3A, if each pixel in the partial area m is binarized with reference to the threshold value th2 instead of the threshold value th1, the intermediate density portion of the hatched portion in the drawing is binarized. Since it is smaller than the threshold value th2 and becomes the "L" level, as shown in FIG. 6, the background portion B2 in the partial region m is removed without being formed into a binary shape. The same occurs in the partial areas p, q, and r that are in contact with each of the high-contrast partial areas including the character part. In these partial areas p, q, and r, the background portion B2 is formed into a binary shape. Removed without. Thus, the character portion L
Is removed, and the background portion B2 of intermediate density is removed and the two portions are separated from each other.
There is no problem that the image is deformed and cannot be identified.

In this embodiment, steps 106 and 1 in FIG.
By providing 07, the character portion L can be further reliably identified. That is, in step 106, an evaluation value ηm, n of the likelihood of the binarization threshold thm, n calculated by using the discriminant analysis method in each partial area (m, n) is obtained. , An appropriate binarization threshold thm, n having a large evaluation value ηm, n (that is, a binarization threshold of a partial region including a character portion L and having a high contrast) is replaced with a peripheral portion having a binarization threshold having a small evaluation value. Region (that is, a partial region including the middle density background portion B2)
To be propagated. This propagation is performed specifically as follows. First, a discriminant analysis method as a premise will be described. Assuming that the total number of pixels in the partial region is N, the density level is L, and the frequency (number of pixels) of the density i is h (i),
The average concentrations μ1 and μ2 of each class equal to or smaller than the threshold th and equal to or larger than th + 1 are calculated by the following equations (1) to (3).

[0017]

(Equation 1)

If the distribution ratios ω1 and ω2 of each class are expressed by the following equations (4) and (5), the inter-class variance is expressed by the following equation (6).
The threshold th that maximizes this inter-class variance is defined as a binarization threshold thm, n.

[0019]

(Equation 2)

In step 106, the binarization threshold th
The evaluation value ηm, n of m, n is calculated by the following equation (7). The evaluation value ηm, n takes a value between 0 and 1, indicating that the closer to 1, the higher the degree of separation.

[0021]

(Equation 3)

Here, σ 2 is the variance of the density, and the following equation (8)
It is represented by

[0023]

(Equation 4)

In step 107, the binarization threshold thm,
Propagation of n is performed as follows. That is, first, the weight value W of each partial region is determined by the following equation (9).

[0025]

(Equation 5)

Here, thη is a constant, for example, 0.7.

Next, when the weight value W (m, n) of the focused partial area (m, n) is equal to or smaller than a constant thW (for example, 0.95), the binarization threshold of each surrounding partial area is set. And a new threshold value th'm, n and a weight value W 'of the partial area (m, n) based on the following equations (10) and (11).
(M, n) is calculated. If the weight value W (m, n) of the partial area (m, n) is larger than the constant thW, the binarization threshold thm, n is left as it is.

[0028]

(Equation 6)

By repeating the above processing P times for all the partial areas, the weight value W '(m, n) of the area where the weight value W (m, n) is 1 or less gradually approaches 1,
The evaluation value ηm, n is large, that is, the contrast is high,
The binarization thresholds of the partial areas including the character portion L propagate to the periphery, and the binary thresholds of these peripheral partial areas are raised. As a result, in the vicinity of the character portion L, the background portion B2 having the intermediate density is more reliably removed. Note that the constant thη,
thW is experimentally set to a value of 1 or less close to 1. Also,
The number of repetitions P for propagation of the binarization threshold is also experimentally set to a value of 1 or more. The actual value of P is, for example, about 40 times.

If the same binarization threshold thm, n is applied to all pixels in the partial region, the binary threshold may suddenly change at the boundary of each partial region. Therefore, in order to prevent this, in step 108 of FIG. 2, the binarization threshold for each pixel in each partial region is determined by interpolation. This is, as shown in FIG. 7, where the center pixel of the partial area (m, n) is p (m, n), and the binarization threshold th
m, n, p (m, n), p (m + 1, n), p
The binarization threshold thp of the target pixel p inside the rectangle having the vertices of (m, n + 1) and p (m + 1, n + 1) is obtained by the following equation (12).

[0031]

(Equation 7)

Here, a and b are ratios when the x and y coordinate values of the target pixel p internally divide the sides of the rectangle (FIG. 7).

In step 109 of FIG.
Based on the binarization threshold thp determined in step 8, the binarization of the density is performed for each pixel p in the partial area. As described above, in the binarization method of the present embodiment, the binarization threshold of each partial region is determined based on the weighted density distribution histogram, and the binarization threshold having a high evaluation value is transferred to the surrounding partial regions. Propagation and interpolation of each pixel in the partial region so that the binarization threshold value changes smoothly, thereby forming a binary shape of the intermediate density background portion B2 especially near the character portion L. Is reliably prevented, and subsequent character identification can be performed satisfactorily.

In the above embodiment, step 1
Even if steps 06 to 108 are omitted, the effect of preventing the intermediate density background portion from being binarized can be sufficiently obtained. Further, the character portion L in the above embodiment includes not only original characters but also pictures and symbols. Further, the partial region does not necessarily need to include a plurality of pixels, and the partial region may be each pixel (that is, one pixel is included). In this case, the histogram is obtained by the following method. For example, if the brightness of the attention area (= pixel) is 10
If the weight is 0 and the weight is 1, the brightness is 100 in the histogram.
Is increased by one. At this time, if the brightness of the peripheral area (= pixel) is 80 and the weight value is 0.5, the frequency of the brightness 80 in the histogram is increased by 0.5.

A program for realizing the above processing by the CPU can be provided in a state recorded on a medium. Examples of a medium storing the program include a flexible disk, a CD-ROM, and a memory card. The program recorded on the medium is installed in a memory provided in the character identification device, whereby the program is executed to realize an image binarization unit and a character identification unit.

[0036]

As described above, according to the image binarization method of the present invention, it is possible to more reliably prevent a background portion close to a character portion or the like from being formed into a binary shape. And the like can be significantly improved.

[Brief description of the drawings]

FIG. 1 is a block diagram of an apparatus for implementing a method of the present invention.

FIG. 2 is a flowchart showing a procedure of binarization in a CPU.

FIG. 3 is a density distribution histogram of each partial region.

FIG. 4 is a weighted density distribution histogram of a partial area of interest.

FIG. 5 is a diagram illustrating an example of assigning a weight value to each partial region.

FIG. 6 is a conceptual front view of a binarized image showing the effect of the present invention.

FIG. 7 is a conceptual front view of an image plane for explaining interpolation of a binarization threshold.

FIG. 8 is a conceptual front view of an image plane.

FIG. 9 is a conceptual front view of an image plane.

[Explanation of symbols]

1 ... TV camera, 2 ... character identification device, 24 ... CPU,
3. Display device.

Claims (1)

[Claims] 1. A step of dividing an obtained image into a plurality of partial areas; a step of creating a density distribution histogram for each of the partial areas; Assigning a weight value to a predetermined number of partial areas before and after this; and (4) multiplying the density distribution histogram of each partial area by the weight value and adding the result to obtain a weighted density distribution histogram of the focused partial area. Obtaining; (5) setting a binarization threshold value of the partial area of interest based on the weighted density distribution histogram; and (6) setting a binarization threshold value of the partial area of interest based on the binary threshold value. Binarizing the density of each pixel, and repeating the steps (3) to (6) for all partial regions in the image. Law.