JP2943303B2 - Image binarization method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a binarization method of an image, and more particularly, to a binarization method of a monochrome character image as an input image of an OCR.

[Prior Art] As a conventional image binarization method, as shown in FIG. 3, a histogram based on the density value of a corner image is taken over the entire area of an image, and it is attempted to distinguish peaks of frequencies appearing in a background and an object. There is a method to set a threshold value in the valley part,
There are several examples of the method of setting the valley threshold.

As a representative of a simple example, there is a method in which a density value of a minimum point of a valley is used as a threshold value. (For example, Daily Diameter Medical 198
4.7.30.P-205) Also, as a complicated example, the peak of the frequency of the background and the object is regarded as a probability distribution, and it is considered as a clustering problem that separates into two classes. There is a method of calculating an inter-variance and setting a threshold value that maximizes the inter-variance as an optimum point. (Method by discriminant analysis, Electrotechnical Research Institute, Research Report No. 818, "Mathematical Research on Feature Extraction in Pattern Recognition", P-157) Since an attempt is made to set a valley, a threshold value with good separability is determined when both the background and the object have an appropriate variance.

[Problems to be solved by the invention]

The conventional image binarization method described above has disadvantages for an OCR (character recognition device) as described below.

First, in OCR, special high-quality paper, which is usually called OCR paper, or general high-quality paper is used.Therefore, the variance in background density is considered to be minimal, and the background should be white and used as the reference level for optical systems. And it is difficult to calculate the density variance of the background. Further, since a sheet on which a character entry frame is printed in advance using a dropout color is often used, the reference level is usually set to a density higher than the density of the dropout color.

Therefore, in OCR, rather than the problem of separation between the background and the character, the problem is how to determine the optimum threshold value in the density distribution range of the character. In other words, if the threshold is too low, the complicated part of the character image will be destroyed, and if it is too high, the thin part of the character image will be lost. Therefore, means for determining this intermediate level as the optimum threshold is necessary. .

In order to solve this problem, the conventional image binarization method does not consider the collapse or lack of characters, and therefore tends to select a point lower than the above-mentioned intermediate level as the optimum threshold value for separating from the background. Further, the conventional latter method is not suitable for OCR for processing slips at high speed because the amount of calculation required to calculate the inter-class variance for each threshold becomes enormous.

[Means for solving the problem]

The image binarization method according to the present invention includes: a base threshold binarizing means for binarizing an image by a base threshold which is a threshold previously determined as a reference value serving as an index of image density; Counting means for counting the number of black spots which are images having, and accumulating means for accumulating the density of the black spots, and a target threshold value based on a ratio between the count value obtained by the counting means and the accumulating means. And a threshold value determined by the threshold value determining means.
And a decision threshold binarizing means for making a value.

[Principle and operation of the solution]

Next, the principle of the image binarization method of the present invention will be described.

As described above, since the OCR is based on the density level at which the dropout color is regarded as the background, the observation result of the density distribution is as shown in FIG. Here, when a point that is equal to or larger than the threshold when a certain threshold is set is called a black point, the number of black points is represented by the following equation, and the relationship between the threshold and the number of black points is as shown in FIG.
FIG. 4 is an integrated form.

(Number of black dots) = ∫ (frequency) (1) In FIG. 5, point A is considered to be a point where the character image appears very darkly and a complicated shape character is broken. Point C is considered to be a point where the character image becomes thin and chipping or blurring occurs. Therefore, the point B at which no character collapse or chipping appears is the target threshold.

By the way, character images need to be normalized because there are various differences in the complexity and thickness of the characters. Then, a base threshold (St) is determined near the above-mentioned reference level (it may be equal to the reference level), and the number of black points is normalized. As shown in FIG. 6, the relationship between the threshold and the black point ratio is as follows.
The variation due to the character type can be regarded as a problem only in the threshold direction.

Here, (black point ratio) = (number of black points) / (number of black points at base threshold) (2) In FIG. 6, points Sa to Sd indicate optimal thresholds in each of cases a to d. Here, the optimum threshold value has a correlation with the area of the black point ratio (the shaded area in FIG. 6) in the threshold region equal to or larger than the base threshold value St, and is considered to be determined by the following equation.

(Optimum threshold) = f (D) (3) where D is the area of the black point ratio, and D = ∫ (black point ratio) = ∫∫ (frequency) / (number of black points at base threshold) (4) is there. Here, when attention is paid to ∫ (frequency) = (density) − (base threshold) (5), D = ∫ ((density) − (base threshold)) / (the number of black spots at the base threshold) (6) The connection between the principle of the method described above and the constituent means described above is as follows.

The above-described base threshold binarization means is a means for determining whether or not the density of each pixel is higher than the base threshold. The black point counting means is means for calculating the number of pixels having a density higher than the base threshold, that is, the number of black points at the base threshold. Further, the concentration accumulating means is means for calculating 分子 ((concentration) − (base threshold value)) which is a numerator on the right side of the equation (6). Finally, the threshold value determining means is a means for calculating D by calculating the expression (6) and determining an optimum threshold value from the relational expression of the expression (3).

〔Example〕

 Next, the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram of a character image binarization system according to one embodiment of the present invention, and illustrates the flow of data. The image on the paper surface captured by the optical scanner is stored in the multi-value image storage unit 1.

As a first step, the density value of each pixel is read from the multi-valued image storage means 1, and the number of pixels determined as a black point by the base threshold binarization means 2 is obtained by the black point coefficient means 3. At the same time, the total of the density values based on the base threshold value for the pixel determined as the black point is obtained by the density accumulation means 4. next,
The threshold value determining means 5 calculates the ratio between the density accumulation value obtained by the density accumulating means 4 and the number of black spots obtained by the black point coefficient means 3, and determines the threshold value from this.

In the second stage, the target binary image is optimally transferred to the binary image storage unit 7 by the decision threshold binarization unit 6 for binarizing the multi-valued image with the threshold value determined by the threshold value determination unit 5. The binarization ends.

In the above description, the method of determining the threshold value from the ratio between the cumulative density value and the number of black spots is not particularly stated, but a first-order approximation is considered as the simplest method. That is, assuming that the ratio between the cumulative density value and the number of black spots is D, and the optimal threshold is S, S = αD + β (α and β are constants) (8). In fact, D is the fifth as described above.
Since it means the area of the hatched portion in the figure, it is easily estimated that S and D are close to the linear correlation. As another method, it is generally considered to approximate to an nth-order approximation, a discontinuous linear approximation, or another function. Good.

Next, FIG. 2 shows the embodiment shown in FIG. 1 in a diagram composed of circuit elements. The components are a multivalued image memory 8,
Base threshold comparing circuit 9, black point counter 10, density adder 1
1, a threshold decision logic circuit 12, a decision threshold comparison circuit 13, and a binary image memory 14. The role of these components is equivalent to the blocks 1 to 7 in FIG. 1, respectively. In the first stage, the operation up to the threshold determination logic circuit 12 is started and the threshold is determined.
At each stage, the decision threshold value comparison circuit 13 is activated to control each circuit so as to generate a target binary image. All of these components can be realized by using those realized by existing integrated circuit technology.

Here, it is necessary to devise the density adder 11 to add the value of (density-base threshold), but there is also a method as follows. That is, the density amount exceeding the reference level determined when the image from the optical system is stored in the multi-valued image memory 8 is treated as the density again, and the base threshold value is set to the reference level (that is, zero). Therefore, the density adder 11 only needs to simply add a new density, and the circuit can be simplified.

Now, in the character image binarization method of the present embodiment, an object has been described as representing characters, but it is stated that the present invention can be applied to general line images such as graphic patterns. Keep it.

Here, FIG. 7 shows an embodiment in which both the binarization method of the present invention and the conventional threshold value determination method are used. Conventional threshold determination means for the image input by the optical scanner 15
A threshold value for separating the background and the object is determined by 16, and the multivalued image whose density has been corrected so that the threshold value becomes zero is transferred to the multivalued image storage device 17. Next, the multi-valued image is binarized by the binarizing means 18 based on the binarizing method shown in FIG.

The binarization method of the present invention determines an optimal threshold value in the density distribution of an image of an object such as a character, and is inconsistent with a conventional method that determines an optimal threshold value for separating an object from a background. Not something. Accordingly, a combined method shown in FIG. 7 in which the conventional method is applied to the separation of the object and the background and the binarization method of the present invention is applied to the separated object is considered. The reference level set by the OCR optical system described above in order to regard the dropout color as the background is the same as that in which the threshold for separating the object and the background is determined.

〔The invention's effect〕

As described above, in the character image binarization method of the present invention, a base threshold value is determined in advance, and the number of pixels that become a black point when the binarization is performed using the base threshold value and the black point density accumulation value are obtained. By performing the ratio calculation, it is possible to determine an optimum threshold value for eliminating the collapse and lack of the character image. Therefore,
As a result, the burden on the recognition unit for character recognition is reduced, and the recognition accuracy is increased.

Further, as described in the embodiment, it is easy to realize by a logic circuit, and there is an effect that the optimum threshold can be determined extremely quickly.

[Brief description of the drawings]

1 and 2 are a block diagram and a circuit diagram, respectively, of an image binarization system according to an embodiment of the present invention. FIG. 3 is a correlation diagram between image density and frequency. FIG. FIG. 5 is a correlation diagram between the threshold value of a character image and the number of black dots, FIG. 6 is a correlation diagram between the threshold values of various character images and the black point ratio, and FIG. 7 is a conventional threshold value determination. FIG. 11 is a block diagram of an image binarization circuit according to another embodiment of the present invention, which is used in combination with a system. 1... Multi-valued image storage means, 2...
3 black point counting means 4 density accumulation means 5 threshold value determination means 6 decision threshold binarization means 7 binary image storage means 8 multi-value image memory 9 ... Base threshold comparison circuit, 10 ... Spot number counter, 11 ... Density adder, 12 ...
… Threshold decision logic circuit, 13… decision threshold comparison circuit, 14…
Binary image memory, 15: Optical scanner, 16: Conventional threshold value determining means, 17: Multi-value image storage device, 18: Binarization means, 19: Binary image storage device.

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06K 9/18-9/44

Claims (1)

(57) [Claims] 1. A base threshold binarizing means for binarizing an image with a base threshold which is a threshold previously determined as a reference value serving as an index of shading of an image, and a black point as a pixel having a density equal to or higher than the base threshold Counting means for counting the number of black dots, and accumulating means for accumulating the density of the black spot, and threshold value determining means for determining a target threshold value from a ratio between the count value obtained by the counting means and the accumulating means. And a decision threshold binarizing means for binarizing the image again with the threshold value determined by the threshold value determining means.