JPH02191086A - Optimum binarizing method - Google Patents

Abstract

PURPOSE:To improve the rate of recognition by holding plural threshold values into a memory, counting the number of black picture elements in the respective threshold value and the number of the black picture elements which are macro-visualized, obtaining the optimum threshold value from the threshold values, in which the parameter of the fractal dimension of a picture is made minimum, based on the number of these black picture elements and outputting a binary picture. CONSTITUTION:A multilevel picture is read by a multilevel picture read part 1 and held in a multilevel image memory 4. Next, the multilevel picture is read from the multilevel image memory 4 and the black and white binary picture is generated and held in a binary image memory 6. After that, to the whole picture held in the binary image memory 6, the number of the black picture elements and the number of the black picture elements, which are macro-visualized, are counted by a black picture element number count part 7 and a counted result is held in a black picture element number memory 8. Then, in a parameter calculation part 9, the number of the black picture elements and the number of the black picture elements, which are macro-visualized, are respectively read from the black picture element number memory 8 and the parameter is calculated in the fractal dimension of the picture. Based on the threshold value, in which this parameter value is made minimum, the optimum threshold value is obtained and the binary picture is outputted.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to an optimal binarization method in a pattern recognition device such as character recognition.

2. Description of the Related Art In general, images processed in pattern recognition devices such as character recognition are obtained by converting values such as the output of a CCD sensor of a scanner into black and white binary values using a threshold value (threshold level). At this time, in order to enable optimal binarization even for originals with poor printing conditions, it is necessary to generate optimal binary images corresponding to the differences in density of the originals.

Here, various methods have been proposed regarding such a binarization method. For example, “Written by Hideyuki Tari, published by Souken Publishing,
1985 [i' Introduction to Computer Image Processing], No. 67
There are the modal method and the differential histogram method, which are shown in the literature "P. The mode method calculates the histogram of the density values of a given image, and when the distribution has two peaks,
The threshold value is determined at the valley between two peaks.

The differential histogram method uses differential values (
The threshold value is determined using the rate of change in concentration.

In addition, [1972 National Conference of the Institute of Electronics and Communication Engineers, Information Section,
Nobuyuki Otsu, “Threshold Determination Method from Concentration Distribution J, 145”
There is a method for determining the threshold value from the concentration distribution, which is shown in the literature.

This method uses only the 0th and 1st moments of the concentration distribution and determines the optimal threshold based on integration.

Furthermore, there is an "optimal binarization method" disclosed in Japanese Patent Publication No. 60-37952. This involves storing a multilevel video signal in a video buffer, converting the video signal read from the video buffer into a binary signal using a slicing circuit with variable slice levels, and slicing the multilevel video information at different slice levels to create a binary signal. The line width amplification factor (number of black points)/(number of surroundings) for each of the plurality of binarized video signals created by converting it to a digital video signal and slicing it at different slice levels is calculated, and the line width amplification factor and the number of line width amplification factors are calculated. The slice level of the slice circuit is set based on the reference line width amplification factor.

However, the problem to be solved by the present invention is that the mode method cannot be applied to documents with poor printing conditions because the histogram does not have clear valleys.

Further, the differential histogram method does not work effectively when the density value near the boundary between the object and the background changes in a complex manner.

Furthermore, the method of determining a threshold value from the density distribution is not an effective method for processing blurring or blurring of "r-line" as an image used in pattern recognition such as character recognition.

Furthermore, in the optimal binarization method disclosed in the above-mentioned publication, as a result of experiments, it was found that appropriate threshold value determination was unstable depending on the shading of the document.

Means for Solving the Problems The invention according to claim 1 provides an optimal binarization method for a pattern recognition device that converts a multi-level quantized image into a black and white binary image. Binarize the image, keep the binary image at each threshold in memory, count the number of black pixels at each threshold and the number of black pixels when coarse-grained, and calculate the fractal image of the image based on these numbers of black pixels. A parameter called a dimension is calculated, an optimal threshold value is determined from the threshold value at which the parameter value becomes minimum, and a binary image based on this optimal threshold value is output.

In the invention according to claim 2, after an image is binarized using a plurality of threshold values, only when counting the number of black pixels at each threshold value and the number of black pixels when coarse-grained, a binary image is generated at each threshold value. is stored in memory, and a parameter called the fractal dimension of the image is calculated based on the argument value of the number of black pixels at each threshold value and the number of black pixels when coarse-grained, and the optimal value is calculated from the threshold value at which the parameter value becomes the minimum. A threshold value is determined, and the image is binarized again using the optimum threshold value to output a binary image.

Furthermore, in the invention described in claim 3, as in the invention described in claim 2, the image is binarized using a plurality of threshold values, and the number of black pixels at each threshold value and the number of black pixels when coarse-grained are counted. After storing the binary image at each threshold value in memory, the fractal image of the image is calculated based on the count value of the number of black pixels at that threshold value and the number of black pixels when coarse-grained by changing the threshold value from the center value. Calculate the parameter called dimension, always keep the threshold value with the smaller parameter value and the parameter value, and use the threshold value where the parameter value is the minimum as the optimal threshold value, and binarize the image again to output a binary image. I made it J.

According to the invention described in claim 1, a multi-level quantized image is binarized using a plurality of threshold values and stored in a memory. and,
The number of black pixels at each threshold value is counted, and the number of black pixels when coarse-grained is also counted, and based on the counting results, a parameter called the fractal dimension of the image is calculated. The optimal threshold value is determined based on the threshold value at which this parameter value becomes minimum, and a binary image is output. Therefore, the optimal threshold value for binary conversion according to the density of the document is automatically set, and the recognition rate is stabilized. .

In this case, even if the memory storing the binary image is saved to one as in the invention described in claim 2, it is only used when counting the number of black pixels and the number of black pixels when coarse-grained. , there will be no problem.

On the other hand, the threshold value at which the parameter value of the fractal dimension is maximum has the property of being close to the center value among the plurality of threshold values. Therefore, according to the invention as claimed in claim 3, a parameter called the fractal dimension of the image is calculated by changing the threshold value from the central value, and the parameter value and the threshold value, whichever is smaller, are always stored, and the parameter value is By focusing on the threshold value with the minimum value, it is possible to speed up the process of determining the optimal threshold value.

Embodiment An embodiment of the invention set forth in claim 1 will be described with reference to FIGS. 1 to 3. FIG. 1 shows a block diagram for implementing this embodiment. This embodiment relates to a multivalued image reading section 1 to a binary image outputting section 2. A multi-value image reading section 1 reads a multi-value image from a scanner 3, quantizes it into, for example, 16 values, and stores it in a multi-value image memory 4.

Next, a 16-gradation multi-value image (
The density levels 0 to 15) are read by the binarization unit 5, and a black and white binary image in which the threshold value t=15 or more is black and the rest is white is generated, and the numbers in the 15 binary image memories 6,
It is held in the items shown in (1). For the entire image stored in the binary image memory 6, which is No. Count the number of pixels,
The counting result is stored in the black pixel number memory 8.

However, the number of black pixels when coarse-grained refers to the number of adjacent pixels (see Figure 3(a)) for a normal pixel as shown in Figure 3(a).
b)) or 166 pixels (in the case of (c) in the same figure) is treated as one pixel and the coarse-grained pixel is treated as a black pixel when coarse-grained. , the number of black pixels when coarse-grained.

Next, the parameter calculation unit 9 reads the number of black pixels and the number of coarse-grained black pixels from the black pixel number memory 8, calculates a parameter called the fractal dimension of the image, and stores it in the parameter memory IO. The corresponding No. of (1
) is stored in memory.

This is the processing for the threshold value t-15.Next, the threshold value t-14 is set, two images are generated by the threshold value t=14 by the binarization unit 5, and the results are stored in the binary image memory 6.
No. (2) is retained. This No, (
2) Similarly to the above, for the two images stored in the image memory 6, count the number of black pixels and the number of black pixels when coarse-grained, calculate the parameter called fractal dimension, Corresponding No. in parameter memory 10, (2)
stored in memory. Other threshold values t=13, 12. ~,
The same process is repeated for 2 and 1 as well.

When the processing for each of these 15 types of thresholds t is completed, the parameter comparison unit 11 extracts the parameter values for each threshold from each of No. (1) to No. (15) in the parameter memory 10 and compares them. , among which the parameter value (fractal dimension) is the maximum threshold T ma
Find x. In this way, the threshold value is decreased in order from the threshold value at which the parameter value becomes the maximum, and the threshold value T inf at which the parameter value (fragmental dimension) becomes the minimum value is determined. In this way, the optimal threshold value determination unit 12 determines the optimal threshold value T =
Determine T inf. Two images under this optimal threshold are selected from the binary image memory 6, outputted to the binary image output section 2, and further sent to the character recognition section 13 etc. for recognition processing.

Here, a method of calculating the fractal dimension, which is one of the features of this embodiment, will be explained with reference to FIG. First, as shown in the figure, r=l, r=2. Number of black pixels N11 for each case of r=4. Count. Then, the number of black pixels when r=1 is N.

The number of black pixels when coarse-grained to r=2 (in units of 4 pixels) is N(21, and the number of black pixels when coarse-grained to r=4 (in units of 16 pixels) is N141.And, If the fractal dimension is D, then the relationship QOgNtr+ = -DQog r 10C (C is a constant) holds, so the fractal dimension is determined by the least squares method using the three points of r=1.2.4.

Next, an embodiment of the invention according to claim 2 will be described with reference to FIGS. 4 and 5. The same parts as those shown in the previous embodiment are indicated using the same reference numerals. In this example, the 15 binary image memories 6 in the previous example are limited to only one binary image memory 14, and the number of black pixels and the number of black pixels when coarse-grained are counted. In this case, this one binary image memory 14 is used. Therefore, after the optimal threshold value T = T inf is determined, the binarization process is performed again using that threshold value.

First, similarly to the embodiment described above, a multi-value image reading section 1 reads a multi-value image from a scanner 3, quantizes it into, for example, 16 values, and stores it in a multi-value image memory 4. Next, the 16-gradation multivalued image is read from the multivalued image memory 4 by the binarization unit 5, thresholded, and a black and white binary image in which 15 or more is black and the rest is white is generated, and the binary image memory 14 held.

For the entire image held in the binary image memory 14, the black pixel number counting unit 7 counts the total number of black pixels (black pixel number) of the two images and the number of black pixels when coarse-grained. , the counting results are held in the black pixel number memory 8.

Next, the parameter calculation unit 9 reads the number of black pixels and the number of coarse-grained black pixels from the black pixel number memory 8, calculates a parameter called fractal dimension, and calculates the number of black pixels in the parameter memory 10. , (1) Retained in memory.

This is the process for the threshold value t-15.Next, the threshold value t=14 is set, two images are generated by the threshold value t-14 by the binarization unit 5, and the results are stored in the binary image memory 1.
4. 2 held in this image memory 14
Similarly to the above, for individual images, the number of black pixels and the number of coarse-grained black pixels are counted, the parameter called fractal dimension is calculated, and the No. in the parameter memory 10, (
2) Retained in memory. Other threshold values t=13, 12.
The same process is repeated for ~2 and 1 as well.

When the processing for each of these 15 types of thresholds is completed, the parameter comparison unit 11 extracts the parameter value for each threshold from each of Nα (1) to No. (15) in the parameter memory 10 and compares it. Among them, a threshold value T max at which the parameter value (fractal dimension) becomes maximum is determined. Starting from the threshold value T max where the parameter value obtained in this way is the maximum, the threshold value is decreased in order, and the threshold value Tinf where the parameter value (fractal dimension) is the minimum value is determined. In this way, the optimal threshold value determining section 12 determines the optimal threshold value TT inf. Using this optimum threshold T, the image read from the multi-valued image memory 4 is binarized by the binarization section 5 and stored in the binary image memory 14. The binary image held in the binary image memory 14 is output to the binary image output section 2, and further sent to the character recognition section 13 etc. for recognition processing.

Further, an embodiment of the invention according to claim 3 will be explained with reference to FIGS. 6 to 8. In this example, the threshold value (1 to 15=1
The change in fractal dimension with respect to This is intended to speed up the calculation. Therefore, in terms of configuration, compared to FIG. 4 of the above embodiment, 15 parameter memories 10 are
1) It is replaced with the parameter memory 15 only for (2).

First, a multivalued image is read from the scanner 3 and stored in the multivalued image memory 4 as in the case described above.

Then, the binarization unit 5 converts the multivalued image memory 4 into 1
A 6-gradation multivalued image is read in, the threshold value t = 8 (center value), and a binary image is created in which the threshold value t = 8 or higher is black and the rest is white, as shown in Figure 8 (b). The image data is generated and stored in the binary image memory 14.

Then, for the entire image held in the binary image memory 14, the black pixel number and the coarse-grained black pixel number are counted in the black pixel number counting section 7, and the counting results are used as the black pixel number. It is held in memory 8. Next, the parameter calculation unit 9 calculates a parameter ptP, which is a fractal dimension when the threshold value t=8. This parameter pt=
p, is held in No. (1) in the parameter memory 15.

Next, this time, set the threshold value t=7, repeat the same process, and set the parameter pt=p for the threshold value 7.

This parameter Pt=P, is stored in the other Nα(2) in the parameter memory 15.

Then, the magnitude of No. (1) and No. (2) in the parameter memory 15 is compared, and the smaller threshold value is determined. If the new threshold is smaller,
Save the parameter value pt at that threshold, reduce the threshold t by one, and calculate the parameter value pt corresponding to the threshold in the same way as in the previous case until the previous threshold becomes smaller (Pt(
Repeat this until Pmin is no longer present. Furthermore, if the previous threshold t is smaller, the threshold t is set to 9 and the same process is repeated. In this case as well, the smaller parameter value Pt is always saved. For such a threshold value of 9 or more, if the new threshold value is smaller, the threshold value is increased by 1, and the process of calculating the parameter value Pt corresponding to the threshold value is performed in the same way, and the previous threshold value is Repeat this until it becomes smaller.

After such processing, the last remaining threshold t is the minimum value (P+++in) of the parameter value Pt, so
This is set as a threshold value T min, and an optimal threshold value T is determined from this threshold value T min. Once the optimal threshold value T is determined, the image read from the multivalued image memory 4 is binarized using this threshold value T, stored in the binary image memory 14, outputted to the binary image output unit 2, and further It is sent to the character recognition unit 13 or the like.

Further, another embodiment of the present invention will be explained with reference to FIG. This example is a further improvement of the previous example. That is, in order to avoid the trouble of binarizing the multivalued image again after determining the optimal threshold value, and to minimize the memory,
(]) The binary image memory ■6 with the two memory configuration shown in (2) is used, and the property that the threshold value at which the fractal dimension is maximum is near the median value as in the above embodiment is used. This is what I did. That is, as in the embodiment described above, the threshold value is changed and the smaller parameter value and the binary image at that threshold value are always saved.

First, as in the case described above, a multivalued image is read using the scanner 3 and stored in the multivalued image memory 4.

Then, the binarization unit 5 converts the multivalued image memory 4 into 1
A 6-gradation multivalued image is read in, the threshold value t is set to 8 (center value), a binary image is generated in which the threshold value 18 or above is black, and the rest is white, and ( 1). Then, in this binary image memory 16 (
For the entire image held in step 1), the black pixel number counting unit 7 counts the number of black pixels and the number of coarse-grained black pixels, and the counting results are stored in the black pixel number memory 8.

Next, in the parameter calculation section 9, the threshold value 1. The fractal dimension parameter Pt=P when -8.

Calculate. This parameter Pt=P, is held in No. (1) in the parameter memory 15.

Next, this time, the threshold value t=7 is set, and the same process is repeated (however, (2) in the binary image memory 16 is used), and the parameter Pt=P for the threshold value t=7 is calculated, This parameter Pt=P.

is held in the other No. (2) in the parameter memory 15. And parameter memory] No. of 5,
Compare the size of (1) with No. (2) and find the smaller threshold. If the new threshold is smaller, save the parameter value PL at that threshold,
Decrease the threshold value by one, find the parameter value PL corresponding to the threshold value in the same way as in the previous case, and repeat this until the previous threshold value becomes smaller (Pt (until it is no longer Pmin). At this time, 2 The value image is (1) or (2) in the binary image memory 16 according to the threshold value of the larger parameter value.
will be overwritten. Also, if the previous threshold t is smaller,
Set the threshold value to 9 and repeat the same process. In this case as well, the smaller parameter value Pt is always saved.

For a threshold value of 9 or more like this, if the new threshold value is smaller, increase the threshold value by one, perform the same process to obtain the parameter value Pt corresponding to the threshold value, and calculate the value of the previous threshold value. Repeat this until it becomes smaller.

After such processing, the last remaining threshold t is the minimum value (Pmin) of the parameter value Pt, so this is set as the threshold T min, and this threshold Tm1n is set as the optimal threshold T. At this time, since the binary image based on this optimal threshold value T should be stored in either (1) or (2) of the two binary image memories 16, this binary image is output as a binary image. The data is output to the checking section 2, and further sent to the character recognition section 13, etc.

Effects of the Invention Since the present invention is configured as described above, even for a document with poor printing condition, the optimum two-dimensional image can be printed according to the density of the document.
It is possible to automatically set the threshold value for conversion, and it is possible to improve and stabilize the recognition rate. In particular, according to the invention described in claim 2 or 3, the number of black pixels and the By storing the binary image in the memory only when counting the number of black pixels, the memory for the binary image can be saved, and
According to the invention according to claim 3, which focuses on the fact that the parameter called fractal dimension has a property that the threshold value at which the parameter value is maximum is near the center value in the threshold value change characteristic.
We calculated a parameter called fractal dimension by changing the threshold value from the center value, always kept the threshold value with the smaller parameter value, and focused on the threshold value where the parameter value was the minimum. It is also possible to speed up the process of determining the optimal threshold value.

[Brief explanation of the drawing]

1 to 3 show an embodiment of the invention as claimed in claim 1, in which FIG. 1 is a block diagram, FIG. 2 is a flowchart, FIG. 3 is an explanatory diagram of coarse-graining, and FIG. and fifth
The figure shows one embodiment of the invention according to claim 2, and
FIG. 5 is a block diagram, FIG. 5 is a flowchart, FIGS. 6 to 8 show an embodiment of the invention according to claim 3, and FIG. 6 is a characteristic diagram of fractal dimension change with respect to a threshold value.
FIG. 7 is a block diagram, FIG. 8 is a flow chart, and FIG. 9 is a block diagram showing another embodiment of the present invention.

Claims (1)

[Claims] 1. In an optimal binarization method in a pattern recognition device that converts a multilevel quantized image into a black and white binary image, the image is binarized using a plurality of threshold values, and each threshold value The binary image of is always held in memory, the number of black pixels at each threshold value and the number of black pixels when coarse-grained are counted, and a parameter called the fractal dimension of the image is calculated based on these numbers of black pixels. An optimal binarization method characterized in that an optimal threshold value is determined from threshold values at which the parameter value becomes minimum, and a binary image based on this optimal threshold value is output. 2. In the optimal binarization method in a pattern recognition device that converts a multilevel quantized image into a black and white binary image, the image is binarized using multiple thresholds, and the number of black pixels and roughness at each threshold are calculated. Only when counting the number of black pixels when visualized, the binary image at each threshold is stored in memory, and the count value of the number of black pixels at each threshold and the number of black pixels when coarse-grained is calculated. The method is characterized in that a parameter called a fractal dimension of the image is calculated based on the image, an optimal threshold value is determined from the threshold value at which the parameter value becomes minimum, and the image is binarized again using the optimal threshold value to output a binary image. Optimal binarization method. 3. In the optimal binarization method in a pattern recognition device that converts a multilevel quantized image into a black and white binary image, the image is binarized using multiple thresholds, and the number of black pixels and roughness at each threshold are calculated. Only when counting the number of black pixels when visualized, store the binary image at each threshold value in memory, change the threshold from the center value, and calculate the number of black pixels at that threshold and the black when coarse-grained. A parameter called the fractal dimension of the image is calculated based on the count value with the number of pixels, the threshold value with the smaller parameter value and the parameter value are always retained, and the threshold value with the minimum parameter value is set as the optimal threshold value. An optimal binarization method characterized by binarizing again and outputting a binary image.