JPH064663A - Picture data binarization device - Google Patents

Abstract

PURPOSE:To obtain binary picture data by calculating an optimum threshold coincident with the visual sense characteristic of a human body an/d binarizing a multilevel picture data on the threshold. CONSTITUTION:A binarized error calculation means 2 calculates a binarized error occurred by binarizing multilevel picture data as a square error for the respective thresholds. A degree of complexity calculation means 3 obtains the degree of complexity of a binary picture obtained by binarizing multilevel picture data for the respective thresholds. An optimum threshold calculation means 4 calculates the threshold whose square error becomes minimum and the threshold and whose degree of complexity becomes minimum and calculates a complexity minimum threshold nearest to the minimum threshold of the square error as the optimum threshold from the both values. Multilevel picture data is binarized to binary picture data by the threshold. Thus, the occurrence of noise and the notches of a boundary part can be prevented in a character picture and a graphic picture after a binarization processing, and the binary picture of high quality can be obtained.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image data binarization device for converting multi-valued image data into binary image data, and particularly for binarizing multi-valued character image data.

[0002]

2. Description of the Related Art A conventional technique for binarizing multi-valued image data has been one in which a histogram of signal levels in a multi-valued image is calculated and binarized using the valley of the histogram as a threshold. In addition, as a method of this kind, there is also a method of taking a differential or a Laplacian of a multivalued image to make the valleys of the histogram noticeable and binarizing it. Besides this,
There has also been a method of calculating a squared error between a multivalued image and a binary image as a parameter of the optimality of a binary image generated by binarization, and searching for a threshold value that minimizes the squared error. Also, based on a similar idea, when a human sees a binary image, he uses the perceptual characteristics (the gestalt principle) that he tries to capture as a simple pattern as much as possible, and sets the parameter of the optimality to the complexity of the binary image. There was a thing.

Such a conventional technique is disclosed in the journal of the Institute of Electronics and Communication Engineers, "Automatic threshold selection method based on discrimination and least squares standard", Otsu, '80 / 4, Vol. J63-D, N
o. 4, p349-356, or the Institute of Electronics and Communication Engineers, "A Study on Complexity of Binary Images and Thresholding of Multivalued Images," Taniguchi, Kawaguchi, 1987, Vol. J70-D,
No. 1, p164-173, etc.

[0004]

However, in the conventional method in which the valley bottom of the histogram is used as the threshold value, the threshold value is calculated from the viewpoint of whether or not the generated binary image is visually optimum. There wasn't.

Another method is to use the least square error between the multi-valued image data and the binary image data as the optimum parameter, but this is not based on human visual characteristics. Absent. Then, there are many local minimum values as fluctuations in the value near the least square, and even if the minimum value is simply selected,
It was not always the optimal threshold.

Further, regarding another method in which the parameter of the optimality is set to complexity, the human visual characteristics are based, but when binarization is actually performed by this method, the complexity is minimized. Since the threshold value of 2 is the optimum binary image, there is a problem that the binary image becomes pure white or pure black when the minimum complexity is simply set as the threshold value. To avoid this problem, the valley of the complexity is detected and used as a threshold value. However, when actually binarizing, in the vicinity of the minimum value of the complexity, as in the method of the least square error, many local minima are generated. A value occurred and a simple minimum was not always the optimal threshold.

As described above, according to the conventional method, the optimum threshold value is not determined, so that noise is generated in the generated binary image or jaggedness occurs in the boundary portion of characters or line drawings. There is a problem that the quality of the binary image deteriorates. Therefore, an object of the present invention is to provide an image data binarization apparatus that determines an optimal binarization threshold value that matches human visual characteristics.

[0008]

In order to solve the above-mentioned problems, the present invention is an image data binarizing device which binarizes multi-valued image data with a certain threshold value. A binarization error calculating means for calculating a binarization error that occurs, and 2 obtained by binarizing the multi-valued image data.
A complexity calculating means for calculating the complexity of the value image, an optimum threshold calculating means for calculating a threshold that optimizes the binary image based on the binarization error and the complexity, and a threshold storing means for storing the optimum threshold. And binarizing the multivalued image data into binary image data according to the threshold value stored in the threshold value storage means 2
It is configured to have a quantizing means.

[0009]

In the image data binarizing apparatus configured as described above, the binarization error generated by binarizing the multivalued image data and the binary image obtained by binarizing the multivalued image data A threshold that optimizes the binary image is calculated from both of the degrees of complexity, and the multivalued image data is binarized into the binary image data by the threshold.

[0010]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of an image data binarizing apparatus of the present invention. In FIG. 1, the binarization error calculation means 2 calculates the binarization error at each threshold from the multivalued image data to be binarized obtained from the multivalued image data input means 1. The multi-valued image data is, for example, 8-bit image data obtained by reading a character image, a graphic image, or the like. A method of calculating the binarization error will be described with reference to the flowchart of FIG.

In step 200, it is judged whether or not the input of multi-valued image data is the first time. If it is the first time, go to step 201. If not, go to step 203. In step 201, a density histogram in which the horizontal axis represents the signal level and the vertical axis represents the number of pixels at each signal level is created from the multi-valued image data. In the case of image data of characters and line drawings, this histogram has a bimodal shape having two peaks as shown in FIG.

In step 202, the signal levels at the tops of the two peaks are read in the histogram created in step 201. For example, in FIG. 3, the signal level on the top of the mountain on the low signal level side is C1, and the signal level on the mountain top on the high signal level side is C2.

In step 203, the multi-valued image data is binarized according to the equation 1 with the threshold value θ. gi = C1 {fi <θ} C2 {fi> = θ} (Equation 1) Here, the signal level of a certain pixel in the multi-valued image data is represented by fi (subscript i indicates the pixel position), Equation 1 Let gi be the signal level of the binary image data obtained as a result. θ is
The initial value is 0. As will be described later, this step 20
3 is repeated a plurality of times, and θ is incremented by 1 each time.

In step 204, the mean squared error e of the multivalued image and the binary image of the threshold value θ obtained in step 203.
Calculate (θ). The calculation formula is like Formula 2. Here, N is the total number of pixels.

In step 205, it is confirmed whether the threshold value θ in step 203 is less than the maximum value M of the signal level. M is 255, for example, if the signal is 8 bits. θ is M
If less, go to step 206. If θ is equal to or greater than M, go to step 207.

At step 206, 1 is added to the threshold value θ to obtain a new threshold value θ for binarization at step 203. In step 207, the mean squared error e (θ) at each threshold θ (0 to M) is generated. This e (θ) becomes a binarization error. If this is illustrated by the horizontal axis θ and the vertical axis e (θ), normally, as shown in FIG. 4, the peaks at the positions where θ is 0 and M,
The valley is in the center and the valley is a curve with many local minima.

In FIG. 1, the complexity calculating means 3 calculates the complexity of a binary image generated from multivalued image data. A method of calculating the complexity will be described with reference to the flowchart of FIG. In step 501, the multi-valued image data is binarized according to Expression 1 with the threshold value θ.

[0018] Here, the signal level of a certain pixel in the multi-valued image data is f, and the signal level of the binary image data obtained as a result of Expression 1 is g. The initial value of θ is 0. As will be described later, step 501 is repeated a plurality of times, and θ is incremented by 1 each time.

In step 502, the complexity C (θ) is calculated from the binary image data g at the threshold θ in step 501.
To calculate. The method of calculating complexity depends on the scale.
There are types. The first scale is the number of black and white connected components, the second scale is the length of the black and white boundary line, and the third scale is the number of prime strokes in the DF expression.

The number of connected components will be described. here,
The definition of connectivity is 4 connections. The 4-connection means that a certain pixel and another pixel are adjacently connected to each other via a side. The number of connected components means, for example, as shown in FIG. 6, the number of clusters of the same pixel surrounded by diagonal lines. here,
Since the number of connected components is 4 for both black and white, the number of connected components in FIG. 6 is 3 for black and 1 for a total of 4. Here, a white pixel indicates a signal level 1 and a black pixel indicates a signal level 0.

A complexity calculation method using the number of connected components as a scale will be described with reference to the flowchart of FIG. Step 70
In 1, the pixel of interest is determined. Basically, the scanning is basically performed from the left side to the right side of the image in the main scanning direction and from the upper side to the lower side in the sub scanning direction. That is, the first target pixel is set to the upper left and the subsequent target pixel is set to the right of the previous target pixel. However, the pixel that once becomes the target pixel is skipped. This is because the target pixel may be determined in step 702 as well, which will be described later.

In step 702, one connected pixel group is set as a target pixel. As shown in FIG. 8, the method checks whether four pixels A, B, C, and D shown in FIG. 8 are connected to the pixel of interest X, and if there is a connected pixel, All are considered as target pixels. With respect to this new pixel of interest, it is determined whether pixels on all sides are connected, and the connected pixel is set as the pixel of interest. Repeat this. When there are no connected pixels, go to step 703.

In step 703, the number of times the connected pixels disappear in step 702, that is, the number of connected components α is counted. In step 704, it is confirmed whether all the pixels have become the target pixel. If so, go to step 705,
If not, go to step 701.

In step 705, the complexity C (θ) is calculated from equation (4) from the number of connected components α, and the process ends. C (θ) = α / total number of pixels (Equation 4) A method of calculating the complexity using the length of the boundary line as a scale will be described with reference to the flowchart of FIG. 9.

In step 901, the pixel of interest is determined.
The determination method is, for example, scanning from the left side to the right side on the image in the main scanning direction and from the upper side to the lower side in the sub scanning direction. That is, the first pixel of interest is the upper left, and the subsequent pixel of interest is on the right of the previous pixel of interest. In step 902,
As shown in FIG. 10, it is checked whether the pixel of interest X and the pixel A on the right of the pixel of interest are different types of pixels such as white and black. If the pixel is different, go to step 903,
If the pixels are the same, the process goes to step 904. If there is no adjacent pixel on the right, the processing unconditionally proceeds to step 903.

In step 904, as shown in FIG. 10, it is checked whether the target pixel X and the pixel B below it are different types of pixels such as white and black. If the pixels are different, the process proceeds to step 905, and if the pixels are the same, the process proceeds to step 906. If there is no lower pixel,
Unconditionally, go to step 905.

In steps 903 and 905, the number of times of different pixels in step 902 and step 904 is counted as the boundary line length α. Step 90
At 6, it is checked if all pixels have been scanned. If scanned, go to step 907, otherwise go to step 901.

At step 907, the complexity C (θ) is calculated by the equation 5 from the boundary length α at step 905, and the process ends. C (θ) = α / (total number of pixels * 2) ... Explain. To calculate the number of elementary strokes, the number of pixels is a power of 2.
It must be a power of 2 image. For example, it is an 8 × 8 image as shown in FIG.

In step 1201, the image is divided into four small images A1, A2, A3 and A4 as shown in FIG.
In step 1202, a quadtree is created from the small image in step 1201. The way of making is to make it black if all the pixels of the small image are black, and to make it white if all or some of them are white. In the case of the image of FIG. 11, the quadtree of the first division is as shown in FIG.

In step 1203, the small image is further divided into four.
Check if it can be divided. Step 1201 if possible
If not, go to step 1204. In step 1204, the number of leaf nodes is counted out from the quadtree created in step 1202 to obtain the number of prime images. For example, in the case of the image of FIG. 11, a quadtree as shown in FIG. 15 is created, and the circled marks are leaf nodes, and in this case there are thirteen. Therefore, the number of elementary strokes is 13.

At step 1205, the complexity C (θ) is calculated from the number of prime strokes by the equation (5). C (θ) = number of prime images / total number of pixels (Equation 6) Above, the description of the three types of complexity calculation methods in step 502 of FIG. 5 is completed.

In step 503 of FIG. 5, step 50
It is checked whether the threshold value θ of 1 is less than or equal to the maximum value M of the multilevel signal level (255 for 8 bits). If M or less,
Go to step 504, otherwise step 505
go to. At step 504, the threshold value θ is incremented by 1, and
A new threshold is set in step 501.

In step 505, the complexity C (θ) at the threshold value θ is generated in step 502. When the complexity C (θ) is represented by the horizontal axis θ and the vertical axis C (θ), the curve normally has a valley at both ends and near the center as shown in FIG. Next, the optimum threshold value calculation means 4 in FIG. 1 will be described. In the optimum threshold value calculation means 4, the optimum threshold value θ is calculated from the binarization error e (θ) calculated by the binarization error calculation means 2 and the complexity C (θ) calculated by the complexity degree calculation means 3.
Is to determine. There are two ways to determine the optimal threshold.
There are types.

Each will be described below. The first optimum threshold value determining method will be described with reference to FIG. Step 1701
Then, the threshold value θ at which the complexity C (θ) becomes the minimum is obtained for all the minimums, and θ1, θ2, ..., θn are obtained.

At step 1702, the binarization error e
A threshold value θmin that minimizes (θ) is obtained. Step 1
In step 703, each θi in step 1701 and step 17
The difference of θmin of 02 is taken, θi closest to θmin is searched, and the θ is set as the optimum threshold.

The second optimum threshold value determining method will be described with reference to FIG. In step 1801, thresholds θa and θb corresponding to the first maximum value and the second maximum value of the complexity C (θ) are obtained. In step 1802, in the section from θa to θb,
A threshold value θmin that minimizes the complexity C (θ) is obtained.

At step 1803, the binarization error e
Thresholds θ1, θ2, ..., θn at which (θ) becomes minimum are calculated. In step 1804, the difference between each θi in step 1803 and θmin in step 1802 is calculated to obtain θmi
The θi closest to n is searched, and the θ is set as the optimum threshold.

This is the end of the description of the optimum threshold value calculating method in the optimum threshold value calculating means 4 of FIG. The threshold value storage means 5 of FIG. 1 stores the optimum threshold value calculated by the optimum threshold value calculation means 4.

The binarizing means 6 binarizes the multivalued image data with the threshold value stored in the threshold value storage means 5 to obtain the binary image data 7. All the means described above can be implemented by a program using a CPU, or can be implemented by a part or all of hardware.

The embodiment described above is an example in which the entire target multi-valued image is binarized with one threshold value. However, the multi-valued image is divided into a plurality of areas, and the optimum threshold value is set for each area. Of course, the present embodiment can be applied to the binarizing device for calculation.

[0041]

As described above, according to the present invention, an optimum threshold value is calculated from both the binarization error and the complexity of a binary image, and the multivalued image data is converted into binary image data by the threshold value. Since it is configured to be binarized, it is possible to calculate the threshold value that optimizes the binary image. Therefore, there is an effect that a high-quality binary image can be obtained by preventing generation of noise on the binary image and eliminating jaggedness at the boundary portion of the character line drawing.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of an image data binarizing apparatus of the present invention.

FIG. 2 is a flowchart showing a binarization error calculation method.

FIG. 3 is an explanatory diagram showing a histogram of a multi-valued image.

FIG. 4 is an explanatory diagram showing characteristics of a squared error average e (θ).

FIG. 5 is a flowchart showing a complexity calculation method.

FIG. 6 is an explanatory diagram of the number of connected components.

FIG. 7 is a flowchart showing a method of calculating complexity based on the number of connected components.

FIG. 8 is an explanatory diagram of connected pixels.

FIG. 9 is a flowchart showing a method of calculating complexity based on a boundary line length.

FIG. 10 is an explanatory diagram of a pixel arrangement.

FIG. 11 is an explanatory diagram showing an example of a binary image.

FIG. 12 is a flowchart showing a method of calculating a complexity based on the number of prime strokes of a DF expression.

FIG. 13 is an explanatory diagram showing an image division method.

FIG. 14 is an explanatory diagram showing an example of a quadtree.

FIG. 15 is an explanatory diagram showing an example of a quadtree.

FIG. 16 is an explanatory diagram showing characteristics of complexity C (θ).

FIG. 17 is a flowchart showing an example of an optimum threshold value calculation method.

FIG. 18 is a flowchart showing an example of an optimum threshold value calculation method.

[Explanation of symbols]

 1 Multi-valued image data input means 2 Binarization error calculation means 3 Complexity calculation means 4 Optimal threshold value calculation means 5 Threshold value storage means 6 Binarization means 7 Binary image data

Claims (1)

[Claims] 1. An image data binarizing device for binarizing multivalued image data with a certain threshold, comprising: a binarization error calculating means for calculating a binarization error generated by binarizing the multivalued image data. , A complexity calculating means for calculating the complexity of a binary image obtained by binarizing the multi-valued image data, and an optimum for calculating a threshold value for optimizing the binary image based on the binarization error and the complexity An image data binarization device comprising: a threshold value calculation unit; and a binarization unit that binarizes multi-valued image data into binary image data according to the optimum threshold value.