JP2670372B2 - Binarization method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Purpose of the Invention] (Field of Industrial Application) The present invention relates to a binarization method for converting a grayscale image into a binary image.

(Prior Art) In a character recognition device that recognizes and reads characters included in a grayscale image (an image represented by grayscale digital data) captured by an image input device such as an ITV camera, a grayscale image is converted into a binary image. It has been general to convert the image into a figure and cut out a figure having an area (the number of pixels in an independent area) of a predetermined value or more on the binarized image to be a character recognition target. This specific example will be described below with reference to FIG.

Binarization Processing First, a grayscale image captured by an ITV camera or the like is converted into a binary image using a data conversion function. At this time, if binarization is performed at a low density level (that is, binarization is performed using the low density level as a threshold value), as shown by reference numeral 41 in FIG. ) Is also extracted to obtain a binarized image. On the other hand, if the binarization is performed at a high density level, as shown by reference numeral 42 in FIG. 4, although the noise is removed, a binarized image in which the characters are faint or a part thereof disappears is obtained. Since character recognition is not possible with such a binarized image 42, conventionally, the binarized image 41 in which characters are clearly highlighted by binarizing at a low density level is used.

Labeling Processing The binarized image 41 obtained by binarizing at a low density level contains noise as described above. Therefore, as a first step for removing this noise without affecting the extracted characters, a labeling process,
That is, an independent label value (number) is provided for each region where pixels having true values (pixels having a value of “1”) in the binarized image 41 are connected.
Is performed on each pixel to obtain a labeling image indicated by reference numeral 43 in FIG. In the labeling image 43 of FIG. 4, reference characters L2, L3 ... L
Reference numeral 30 denotes an independent image area (here, a character and noise area) having label values of 2, 3,...

Histogram Processing As the second step of noise removal, a histogram of the labeling image 43 obtained in step 1, that is, a histogram as shown by reference numeral 44 in FIG. 4 showing the number of pixels (area) for each label value of the labeling image is created. Here, an image in a small region (a region where the number of pixels is 10 or less) lower than a certain constant value, for example, 10 is regarded as noise as described below.

Data conversion process As the final step of noise removal, the image of the area where the histogram value (number of pixels) is a certain value or more (here, 10) on the histogram 44 obtained in (here, as is clear from the figure, the label "1" for the image of the area where the value is 6)
As described above, the labeling image 43 is subjected to data conversion (binarization). As a result, as indicated by reference numeral 45 in FIG. 4, a binarized image in which the image (noise) in the area having a histogram value lower than the constant value (10) is removed is obtained. This binary image 45 is used for character recognition and the like.

(Problems to be Solved by the Invention) As described above, in the conventional binarization method, noise is also extracted by binarizing the grayscale image at a low density level in order to completely extract the target image (figure). Even if it is done, if the noise (the area of the connected pixels constituting the noise) is small, it is possible to remove the noise by each of the processes of labeling, histogram, and data conversion described above. However, when a large noise 52 remains like the binarized image 51 shown in FIG. 5 in the above binarization processing, the number of pixels of the noise 52 becomes a certain value (threshold value) or more, The noise 52 cannot be removed, and as shown in FIG.
It remains in the binarized image 53 finally obtained. As a result, the noise 52 remaining in the binarized image 53 is erroneously determined as a part of a character in the case of character recognition, for example.

The present invention has been made in view of the above circumstances, and its object is to generate a large noise (noise) when a grayscale image is binarized at a low density level in order to surely binarize and extract a target image (graphic). (Area) can be reliably removed even if extracted 2
It is to provide a valuation method.

[Configuration of the Invention] (Means for Solving the Problems) The present invention is intended for a binarized image (first binarized image) obtained by binarizing a grayscale image at a low density level. A labeling image is obtained by performing a labeling process, and thereafter, when a histogram of this labeling image is created, a binarized image (second image) obtained by binarizing the grayscale image at a high density level is used. (Binarized image) is used as a mask, and a histogram is created only for an image portion corresponding to a high-density grayscale image portion of the labeling image, and a histogram value of a predetermined value or more on this histogram is set. The label value of the area having the (number of pixels) is detected, and only the image of the area having the detected label value is extracted from the labeling image and binarized to obtain the binarized image (the third image). It is characterized in that to obtain a binary image).

(Operation) In the above configuration, the first binarized image is obtained by binarizing the grayscale image at a low density level. For this reason, the first binary image not only extracts the target image (graphic) such as a character almost completely, but also extracts and includes noise. Therefore, when the labeling process is performed on the first binarized image described above to obtain the labeling image and the histogram of the labeling image is created, the histogram data of the noise portion is also included in the histogram. Therefore, when the histogram of the labeling image is created, the second binarized image obtained by binarizing the grayscale image at the high density level as described above is used as a mask for the labeling image, and the entire labeling image is used. Instead, the histogram is created by targeting only the portion where the second binarized image is effective (the portion of "1"), that is, the image portion corresponding to the grayscale image portion of the high density level.
It is possible to exclude the noise portion from the histogram data (even if it does, a low histogram value). Therefore, by detecting the label value of the area having the histogram value of the predetermined value or more on such a histogram and extracting only the image of the area having the detected label value from the labeling image and binarizing it. It is possible to obtain a binarized image from which noise has been removed.

(Embodiment) FIG. 1 is a block diagram showing an embodiment of a binarizing apparatus to which the present invention is applied. In the figure, 11 is an input image memory for storing a grayscale image of a grayscale (multivalued) digital data input by an image input device (not shown), and 12 is a grayscale image stored in the input image memory 11. Is a data converter for binarizing at a constant density level.
In this embodiment, the data converter 12 performs a first binarization function of binarizing the grayscale number image at a first density level (here, a low density level), and a second density level (here, a high density level). It has a second binarizing function for binarizing at a density level). A binarized image memory 13 for storing an image binarized at a low density level by the data converter 12;
Reference numeral 14 denotes a binarized image memory for storing an image binarized at a high density level by the data converter 12. Reference numeral 15 is a labeling device that obtains a labeling image by performing a labeling process on the binarized image in the binarized image memory 13, and 16 is a labeling image memory for storing the labeling image obtained by the labeling device 15. .

17 is for the labeling image in the labeling image memory 16
A histogram device for masking the binarized image in the binarized image memory 14 to obtain a histogram for each labeling value, and a histogram table 18 for storing the histogram of the labeling image obtained by the histogram device 17. 19
Detects a label value having a number of pixels (histogram value) greater than or equal to a predetermined value on the histogram table 18, extracts only the image having the detected label value from the labeling image, binarizes the final binary value. A data converter 20 for obtaining a binarized image is a binarized image memory for storing the binarized image obtained by the data converter 19. The first
Each element of the figure, that is, the input image memory 11, the data converter 12,
The binarized image memory 13, the binarized image memory 14, the labeling device 15, the labeling image memory 16, the histogram device 17, the histogram table 18, the data converter 19, and the binarized image memory 20 are interconnected by an image data bus. It is also possible to configure a binarization device by the system.

FIG. 2 is a diagram for explaining the flow of the binarization processing in the binarization device of FIG. 1, and FIG. 3 is a diagram for explaining the operation of the histogram device 17 shown in FIG.

Next, the operation of the configuration shown in FIG. 1 will be described with reference to FIGS. 2 and 3 as appropriate, taking as an example the binarization processing of a grayscale image necessary for character extraction for character recognition and the like.

Grayscale image input processing First, using an image input device such as an ITV camera (not shown)
Binary target grayscale image converted to grayscale digital data
21 (see FIG. 2) is input and stored in the input image memory 11.

Low level bell binarization process The data converter 12 binarizes the grayscale image 21 stored in the input image memory 11 at a predetermined first threshold level (low density level). 22 (see FIG. 2) is obtained. Since the binarized image 22 is binarized at a low density level, the character (j in the figure) is completely extracted, but noise is also extracted. Binary image
22 is stored in the binarized image memory 13.

Binarization Processing by High Level Bell Further, the data converter 12 binarizes the grayscale image 21 at a predetermined second threshold level (high density level).
A binarized image 23 (see FIGS. 2 and 3) is obtained. This 2
Since the binarized image 23 is binarized at a high density level, the noise is removed, but the characters are faint or a part thereof disappears. The binarized image 23 is stored in the binarized image memory 14. Note that the binarization processes and can be performed in parallel.

Labeling Processing The binarized image (binary image binarized at a low density level) 22 stored in the binarized image memory 14 is labeled with a labeling unit 15.
Supplied to The labeling device 15 targets the binarized image 22 with an independent label value (number) for each region in which pixels having a true value (pixels with a value of “1”) are connected in the image 22. A labeling process is performed on each pixel to obtain a labeling image 24 (see FIGS. 2 and 3). The labeling image 24 is stored in the labeling image memory 16.
It should be noted that in the labeling images 24 shown in FIGS. 2 and 3, the symbols L2, L3 ... L20 represent independent image areas (here, character and noise areas) having label values 2, 3 ... 20, respectively.

Histogram processing The labeling image 24 stored in the labeling image 24 is supplied to the histogram unit 17. The binarized image (binarized image binarized at a high density level) 23 in the binarized image memory 14 is also supplied to the histogram device 17.
The histogram unit 17 converts the binarized image 23 into the labeling image 24.
3 is masked, and as shown in FIG. 3, the portion of the labeling image 24 corresponding to the effective portion (the portion of "1") of the mask image (binarized image 23), that is, the high density portion of the grayscale image (here, For only the portion of the labeling image 24 corresponding to the character portion), a histogram representing the number of pixels (area) for each label value is obtained. As a result, in this embodiment, the histogram of only the character portion (each character portion having label values of 5 and 8) is obtained.
25 (see Figures 2 and 3) is required. This histogram 25 does not include the histogram data of the noise part (not only the noise part of the small area such as label value 3 or 4 but also the noise part of the large area such as label value 7 or 20) However, the histogram value is small). The histogram 25 is stored in the histogram table 18.

Data Conversion Processing When the histogram 25 is obtained in the histogram table 18 by the histogram unit 17, the labeling image memory
A data converter 19 for converting the labeling image in 16 into a binary image is activated. The data converter 19 detects a label value (here, 5 and 8) having a histogram value (number of pixels, area) equal to or larger than a predetermined value from the histogram 25 on the histogram table 18, and detects this. Only the image portion in the labeling image 24 having the label value is extracted and binarized. As a result, the binarized image 26 in which only the character is reliably extracted, that is, the label value of the character j (
And each character part of 8) is extracted, and the label value is 3 or 4.
The binarized image 26 from which the noise portion of the large region such as 7 or 20 as well as the noise portion of the small region such as is removed is obtained. The binarized image 26 is stored in the binarized image memory 20 and is used for recognition processing by the character recognition device.

Although the binarization processing of the grayscale image necessary for character cutting for character recognition and the like has been described above, the present invention is required to binarize a correct binarized image without noise. It can be applied to all processes.

As described above in detail, according to the present invention, the grayscale image is binarized at a low density level in order to surely binarize and extract the target image (graphic), and the histogram of the labeling image is displayed. At the time of creation, a binary image obtained by binarizing a grayscale image at a high density level is used as a mask image,
By creating a histogram only for the labeling image portion corresponding to the effective portion of the mask image, the histogram value of the noise portion can be set to zero or an extremely small value. By detecting the label value of the area having the histogram value equal to or more than the value and extracting only the image of the area having the detected label value from the labeling image and binarizing it, the small area can be retained while the target image is left correctly. It is possible to obtain a binarized image from which noise of a large area is removed as well as the noise.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of a binarizing apparatus to which the present invention is applied, and FIG. 2 is a diagram for explaining the flow of binarizing processing in the binarizing apparatus of FIG. Figure 3 is the first
Diagram for explaining the operation of the histogram unit 17 shown in the figure,
FIG. 4 is a diagram for explaining the flow of conventional binarization processing,
FIG. 5 is a diagram for explaining the problem of the conventional binarization processing. 11 …… Input image memory, 12,19 …… Data converter, 13,1
4,20 …… Binary image memory, 15 …… Labeling device, 16 ・ ・ ・
… Labeling image memory, 17… Histogram device, 18…
… Histogram table.

Claims (1)

(57) [Claims] 1. A grayscale image represented by grayscale digital data is binarized at a first density level to obtain a first binarized image, and the grayscale image is secondly higher than the first density level.
First binarization to obtain a second binarized image
Binarizing means, a labeling means for performing a labeling process for giving an independent label value to each independent region in the first binary image, and a labeling image for obtaining a labeling image; and the second binary image. With a mask as a mask, a histogram means for obtaining a histogram showing the number of pixels for each label value of the labeling image, and a label value having a pixel number of a predetermined value or more on the histogram is detected, and only the image having the detected label value is detected. And a second binarizing means for extracting the target image from the labeling image and binarizing it to obtain a target third binarized image.