JPH0656615B2 - Binarization threshold decision method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a binarization threshold value determination method for converting a multivalued image including a binary figure such as a character into a binary image, and particularly, a local threshold value determination method.
The present invention relates to a binarization threshold value determination method in binarization.

(Prior Art) As one of the application areas of a character recognition device that reads handwritten or printed characters, it has become necessary to read characters of many documents and mails that are familiar to us in recent years. . In the case of targeting a binary figure such as a character recognition device, if it is captured as a multi-valued image by an image input device such as a scanner from the viewpoint of processing amount, a binary image is obtained using a binarization circuit. Is generally converted to. In the binarization process, it is important to determine the threshold value for binarization in order to quantize the density value of the input image into binary black and white.

As a method of determining the binarized threshold value, the method of empirically fixing the threshold value in advance is the simplest. However, since the input condition often changes depending on the input object, the performance of the image input apparatus, etc., it is necessary to determine the threshold value according to the input image or the partial area of the input image. As a method therefor, there is a method in which the average density in the vicinity of the pixel to be binarized is used as a threshold value. However, if a simple average density value alone does not include an object such as a character or the object and the background The binarization threshold may become unstable depending on the pixel ratio.

In addition, P used when the area of the object is known in advance
Tile method and IEICE Transactions VOL. 63-D, N
O. 42, PP 349 to 356 "Discriminant Analysis and Discriminant Analysis Based on Least Squares Criteria", assuming that the density histogram has a bimodal distribution between the background and an object such as a character. There is a method of deciding the binarization threshold value using the criterion of. However, it is necessary to locally determine the binarization threshold value because the density of the object existing in the image may not be uniform in the mail or general document, or the background of various paper quality may be used. Occurs.
In this case, as shown in the PTL85-73 “Study on Character Extraction Method from Scene Image” by the Institute of Electronics and Communication Engineers (Pattern Recognition and Learning) Study Group, existing 2
A method of linearly interpolating a binarization threshold value obtained by the binarization threshold determination method between each local region is proposed, but the binarization threshold value obtained in each local region is, for example, only for the background. Since it is meaningless and unstable in the local region, noise may occur in the linear interpolation using the binarization threshold value of the local region, and sufficient accuracy cannot be obtained. In addition, when linear interpolation is used, the boundary area between an object such as a character and the background may be blurred.

(Problems to be Solved by the Invention) The object of the present invention is classified according to the binarization threshold value obtained for each local region in order to solve the conventional problems using the local binarization method. The properties of each local region are classified based on the first moment feature (statistical) quantities such as two average density values that represent the properties of the local background and the target, and the properties of neighboring local regions and the first moment features are classified. An object of the present invention is to provide a binarization method capable of performing stable local binarization without increasing the processing calculation amount by adaptively determining the binarization threshold including the amount.

(Means for Solving Problems) The first invention of the present invention is to classify the density value of each pixel in a local area on a multivalued input image into two classes indicating a background and an object, Means for extracting the average density value of, the average density value for each class, and the number of pixels thereof as the feature value of the local region, and means for classifying the local region into monomodal or bimodal according to the feature value. , A bimodal local region group is selected from a plurality of local regions including local regions in the vicinity of the local region, and the background average density value and the target average density value are selected according to the plurality of characteristic values of the bimodal local region group. And means for determining a binarization threshold of the local region, and the one previously determined when the plurality of local regions do not include a preset number or more of the bimodal local regions. There is a binarization threshold for multiple local regions Or means for setting a preset threshold value as a binarization threshold value of the local area.

A second aspect of the present invention is to classify the density value of each pixel in a local area on a multi-valued input image into two classes indicating a background and an object, and to calculate the average density value of the local area and the average for each class. Means for extracting the density value and the number of pixels as the characteristic value of the local area, means for classifying the local area into unimodal or bimodal according to the characteristic value, and a plurality of local areas in the vicinity of the local area A bimodal local region group is selected from the individual local regions, and a threshold value is calculated based on the background average density value and the target average density value calculated according to the plurality of feature values of the bimodal local area group. Means for selecting a unimodal local region group from the plurality of local regions, determine whether the average density value of the unimodal local region belongs to the background or the object according to the threshold value, Including the average density value of the unimodal local region A means for correcting the background average density value and the target average density value, a means for determining the binarization threshold value of the local area from the background average density value and the target average density value, and the plurality of local areas are preset. When the number of the bimodal local regions is not equal to or more than a predetermined number, the one or more binarization thresholds already determined or a preset threshold is set as the binarization threshold of the local region. And means for doing so.

(Operation) In the present invention, the property of the local region is examined according to the first moment feature amount such as two average density values representing the property of the background and the target obtained for each local region. Next, the properties of the regions including the neighboring local regions are examined by using the properties of the respective local regions, and according to the properties of the regions including the neighboring local regions, each local region is adaptively adaptive to each local region from the first-order moment feature amount. By determining the binarization threshold value of, the multivalued image can be easily and accurately converted into a binary image.

(Examples) Hereinafter, specific examples of the present invention will be described with reference to the drawings. FIG. 1 is a table showing an example of features when a local area for a multi-valued image image is binarized.

In FIG. 1 (a), local areas X, Y, and Z indicated by three rectangles in the drawing represent an area (for example, a character portion) in which an object and a background are mixed, an object area, and a background area. Fig. 1
(b) is the density distribution of the local region X shown in FIG. 1 (a) (the horizontal axis represents the density value and the vertical axis represents the frequency expressed in the number of pixels).
Similarly, FIG. 1 (c) shows the concentration distribution of the local region Y shown in FIG. 1 (a), and FIG. 1 (d) shows the concentration distribution of the local region Z shown in FIG. 1 (a). There is. The method of classifying the density distribution in the local area shown in FIGS. 1 (b) to (d) into the shaded objects and the background without shaded lines is the threshold value using the conventional binarization method. S
You can decide L.

As a result, the total average density value μT and the total number of pixels N _T of the local area as the feature amount of the local area, and the average density value μ _{1 of the} area indicating the target (hereinafter referred to as the target class) and the number of pixels N ₁ ,
Also, the average density value μ _{0 of the} area indicating the background (hereinafter referred to as the background class) and the number of pixels N ₀ thereof are extracted. Here, the total number of pixels N _T can be expressed as the sum of the number of target class pixels N ₁ and the background class N ₀ . Note that the above-described feature amount of the local area can be easily calculated because the calculation up to the first moment is sufficient.

When a multi-valued image is converted into a binary image by setting a threshold value for each local area, for example, as shown in FIG. 1 (a), a local area Z consisting only of the background or a local area Y consisting of only the target.
Is divided into two by the threshold value SL, which causes noise in the background or makes the target faint. Further, if the thresholds are extremely different between the local regions, a pseudo boundary line may occur, and some smoothing processing is required between the local regions.

Therefore, in the local binarization method of the present invention, instead of the binarization threshold value extracted for each local region, processing is performed based on the average density value of the target class and background class obtained for each local region, Unimodality for each region (Fig. 1 (c) and 1
(Fig. (D)) or bimodal (Fig. 1 (b)), and the binarization threshold value for each local region is determined by including the features of neighboring local regions. By doing so, the above-mentioned problems are solved. Further, when it is difficult to determine the binarization threshold value including the neighborhood area of the local area, the already extracted binarization threshold value or a preset threshold value can be used.

FIG. 2 is a chart for explaining the principle of the local binarization threshold value determination method in the present invention. In FIG. 2 (a), in order to determine the binarization threshold value of the local area E, the feature quantities of 3 × 3 neighboring local areas A to I are used. As the feature amount of each local area i, the total average density value μ _T (i) (where i indicates the local area names A to I), the target class average density μ ₁ (i), and the background class average density μ ₀ ( i), and the number of pixels N _T (i)
(= N ₀ (i) + N ₁ (i)) · N ₁ (i) · N ₀ (i) is obtained by the conventional technique. In the following description, a 3 × 3 neighborhood area will be considered, but the area is not limited to this.

First, each local region i is classified as unimodal (denoted by symbol s) or bimodal (denoted by symbol b). As a simple method of classifying, for example, the difference μ ₁ between the target class average concentration μ ₁ (i) and the background class average concentration μ ₀ (i) of each local region i
(i) −μ ₀ (i) can be obtained and classified as unimodal or bimodal depending on whether it is larger than a predetermined threshold value. It is also possible to evaluate using the variance value. As a result, the local regions A to I are classified as a single-peaked s and a double-peaked b as shown in FIG. Next, FIG. 2 (b)
The local areas A, D, G, H having bimodal b are selected as shown in FIG. 3, and the four target class average density values and the number of pixels thereof and the four background class average density values and the number of pixels thereof are binary. Determine the activation threshold.

That is, as shown in FIG. 2 (c), the background class average density value of four local regions having the bimodal b Is expressed by the following equation.

However Average density value above And the average density value of the target class of four local regions And are calculated, and the average of them is used as the binarization threshold value SL.
Is determined by the following equation.

Therefore, the embodiment of the first invention is realized by using the above-mentioned binarization threshold value SL as the binarization threshold value of the local area E. Here, in FIG. 1 (a), when there is no bimodal local region including the local region in the vicinity of the local region E, the binarized threshold value of the already determined local region is used. As a simple method, when a preset threshold value or a binarization threshold value is determined, the previously determined binarization threshold value is updated and used for subsequent threshold value determination of the local region. It is also possible to leave all the already extracted binarization thresholds and use one or a plurality of binarization thresholds close to the region where the binarization threshold should be determined.

It should be noted that the binarized threshold value already extracted is used not only in the case of all the unimodal local areas including the local areas in the vicinity of the local area E but also in the case of one or less bimodal areas. You can also

Next, FIG. 2 shows an embodiment of the second invention.
Explanation will be given using (d) and FIG. 2 (c). Figure 2 (c) above
In FIG. 2 (a), the binarized threshold value SL described in the embodiment of the first invention is shown in FIG. 2 (a).
The binarization threshold SL ^* of the updated local region E can be determined by modifying the binarization threshold SL using F · I according to an embodiment of the second invention.

That is, as shown in FIG. 2 (d), the binarization threshold value SL and the total average density values of five unimodal local regions μ _T (B) · μ _T (C) · μ
_T (E) · μ _T (F) · μ _T (I) are compared with each other, and if they are larger than the binarization threshold SL, the unimodal local region is classified into the target class (symbol 1 in the figure). Binarization threshold S
If it is smaller than L, the unimodal local area is classified into the background class (symbol 0 in the figure). Next, the sum of products for the total average densities of a plurality of local regions and the number of pixels determined as the background class (for FIG. 2 (d), Is N _T・ μ _T (B) + N _T (C) ・ μ _T (C) + N _T (E) ・ μ _T (E) + N
_T (F) · μ _T (F), Is N _T (B) + N _T (C) + N _T (E) + N _T (F). ) A background class average density value is calculated by performing calculation and adding it to the sum of products for the background class for only the bimodal local region described in FIG. 2 (c). To calculate. Similarly, average density value of the target class To calculate. These two average density values To the binarization threshold in local region E Can be decided.

When the local area E and the local areas in the vicinity thereof are all unimodal or each of them, one embodiment of the first invention can be used.

FIG. 3 is a logical block diagram showing the configuration of an embodiment of the first invention.

The local binarization feature extraction unit 1 obtains the density distribution of the local area from the multi-valued input image, and determines the threshold value on the density distribution, so that as described in FIG. The average density value / background class average density value / target class average density value and the number of pixels thereof (hereinafter collectively referred to as a feature value) are extracted and output, and sequentially transferred to the local area determination unit 2. The local binarization feature extraction unit 1 can use a known technique. The local area determination unit 2 classifies the local area as unimodal or bimodal by using a threshold value as a difference between the target class average density value and the background class average density value of the local area, and determines the classification. The result is sequentially stored in the local feature storage unit 3 together with the feature value. The bimodal selection unit 4 reads the local region to be binarized and a plurality of neighboring local regions from the local feature storage unit 3, and only the local regions classified as bimodal out of the plurality of local regions are bilocalized. The number of local regions classified as bimodal is stored in the register 5 while being transferred to the peak characteristic class calculating unit 7.

The bimodal class feature calculation unit 7 calculates the background class average density value μ ₀ (b) and the target class average density value μ _{1 of} the plurality of bimodal local regions input as shown in FIG. Using (b) and the number of pixels N ₀ (b) and N ₁ (b), the sum of products value of the background class of the bimodal local region shown in (c) in FIG. (However, Is the sum of the input bimodal local regions)
Sum of products value of target class of bimodal local region Are calculated and stored in the bimodal background feature storage unit 9 and the bimodal target feature storage unit 8. As shown in FIG. 2 (c), the threshold value calculation unit 10 reads out the respective product sum values from the bimodal background feature storage unit 9 and the bimodal target feature storage unit 8 and binarizes them. The value SL is calculated as follows.

However The binarized threshold value SL calculated here is transferred to the binarized threshold value storage unit 11. The contents described above serve as means for determining the binarization threshold value of the local area. The multi-valued input image input to the local binarization feature extraction unit 1 is converted into a binary image by the binarization threshold value transferred to the binarization threshold value storage unit 11.

On the other hand, when the number of bimodal regions stored in the register 5 is transferred to the binarization threshold selection unit 6, the binarization threshold selection unit 6 determines that the number of bimodal regions is equal to or less than a predetermined value. If there is, 2 is used without using the binarization threshold value calculated as described above.
The binarization threshold value determined by the local region already extracted from the binarization threshold value storage unit 11 is selected and is determined as the binarization threshold value of the local region to be binarized at present, and predetermined binarization is performed. It is stored in the storage unit 11.

In the binarization threshold selection unit 6, the already extracted 2
Instead of selecting the binarization threshold value, it is also possible to select a preset binarization threshold value.

FIG. 4 is a block diagram showing the configuration of an embodiment of the second invention.

While the one embodiment of the first invention determines the binarization threshold value only from the local region exhibiting the bimodal characteristic, the one embodiment of the second invention is shown in FIG. The difference is that the binarization threshold value is determined based on the local region exhibiting bimodal characteristics as shown in FIG. That is, in FIG. 4, the local binarization feature extraction unit 1, the local region determination unit 2, the local feature storage unit 3, the bimodality selection unit 4, the register 5, the binarization threshold selection unit 6, the bimodality Class feature calculation unit 7, bimodal target feature storage unit 8, bimodal background feature storage unit 9, threshold value calculation unit 10, binarized threshold value storage unit 11
The respective functions of are similar to those of the first embodiment of the first invention shown in FIG.

Therefore, as described with reference to FIG. 3, the threshold value calculation unit 10 stores the binarized threshold value obtained from only the bimodal local region in the threshold value storage unit 22.

Next, as shown in FIG. 2 (d), the unimodal selection unit 21
A local region showing unimodality is selected from the local regions near the local region to be binarized, and the total average density value and the number of pixels of the unimodal local region are transferred to the background target determining unit 23. The background target determination unit 23 compares the threshold value stored in the threshold value storage unit 22 with the total average density value of the local region, and the unimodal local region indicates the background or indicates the target. If it is a background, the total average density value and the number of pixels are transferred to the unimodal characteristic calculating unit 24 if it is a background, and to the unimodal target characteristic calculating unit 25 if it is a target. The unimodal background feature calculation unit 24 and the unimodal target feature calculation unit 25 have the same function, and the sum of product values of all the average density values input together (for example, in FIG. Indicated S0 is a monomodal local area indicating the background), and is transferred to the background average density calculation unit 26 and the target average density calculation unit 27, respectively.

Further, the background average density calculation unit 26 uses the output values of the bimodal background feature storage unit 26 and the unimodal background calculation unit 24 shown in FIG. 3 to calculate the background average shown in FIG. 2 (e). Concentration value To calculate. Similarly, the target average density calculation unit 27 uses the output values of the bimodal target feature storage unit 26 and the unimodal target calculation unit 25 to calculate the background average density value shown in FIG. 2 (e). To calculate. The background average density value and the target average density value are corrected by the above method.

Finally, the binarization threshold value calculation unit 28 shown in FIG. 4 has the same structure as the threshold value calculation unit 10, and the binarization threshold value calculation unit 28 is connected to the background average density calculation unit 26 and the target average density. The output value of the calculation unit 27 is transferred, and the binarization threshold value of the local area is determined here.

(Effects of the Invention) As described above, by applying the local binarization threshold value determination method shown in the present invention, even when the density of the object in the image is not uniform, only the first-order moment of the local region is obtained. Since it is necessary to use only the circuit for calculating the sum of products, the processing amount is reduced and the binarization threshold value can be stably determined.

[Brief description of drawings]

FIG. 1 is a chart showing an example of features when a local region for a multi-valued image is binarized, and FIG. 2 is a chart explaining the principle of a local binarization threshold value determination method in the present invention.
FIG. 4 is a block diagram showing the configuration of the first embodiment of the present invention, and FIG. 4 is a block diagram showing the configuration of the second embodiment of the present invention. 1 is a local binarization feature extraction unit, 2 is a local region determination unit, 3 is a local feature storage unit, 4 is a bimodal selection unit, 5 is a register, 6
Is a binarization threshold value selection unit, 7 is a bimodal class feature calculation unit, 8 is a bimodal target feature storage unit, 9 is a bimodal background feature storage unit, 10 is a threshold value calculation unit, and 11 is 2 Quantization threshold storage unit, 21 is a single peak selection unit, 22 is a threshold storage unit, 2
3 is a background target determination unit, 24 is a unimodal background feature calculation unit, 2
5 is a unimodal target calculation unit, 26 is a background average density calculation unit, 2
Reference numeral 7 is a target average density calculation unit, and 28 is a binarization threshold value calculation unit.

Claims (2)

[Claims] 1. A density value of each pixel in a local area on a multi-valued input image is classified into two classes indicating a background and an object, and the average density value of the local area and the average density value of each class and its A unit for extracting the number of pixels as a feature value of the local region; a unit for classifying the local region into monomodal or bimodal according to the feature value; and a plurality of local regions including local regions in the vicinity of the local region. Select a bimodal local region group from the region,
Means for calculating a background average density value and a target average density value according to the plurality of feature values of the bimodal local area group to determine a binarization threshold value of the local area; When the predetermined number or more of the bimodal local regions are not included, a binarization threshold value of the one or more local regions which has already been determined or a preset threshold value is set to 2 of the local regions. A method for determining a binarized threshold value, comprising means for setting a binarized threshold value. 2. A density value of each pixel in a local area on a multi-valued input image is classified into two classes indicating a background and an object, and an average density value of the local area and an average density value and pixel for each class. Means for extracting a number as a feature value of the local area, a means for classifying the local area into monomodal or bimodal according to the characteristic value, and a plurality of local areas including local areas in the vicinity of the local area Selecting a bimodal local region group from, means for calculating a threshold value based on the background average density value and the target average density value calculated according to a plurality of the feature values of the bimodal local region group, A group of unimodal local areas is selected from the plurality of local areas, and it is determined whether the average density value of the unimodal local areas belongs to a background or an object according to the threshold value. The background average including the average density value of the peak local area A means for correcting the frequency value and the target average density value, a means for determining a binarization threshold value of the local area from the background average density value and the target average density value, and the plurality of local areas are preset. Means for setting the one or more binarization thresholds already determined when a number of bimodal local regions is not included or a preset threshold as the binarization threshold of the local region A binarization threshold value determination method characterized by having: