JPH03153167A - Character area separation system - Google Patents

Abstract

PURPOSE:To extract both a white background character part and a dot character part as a character area by converting an input picture into a digital multi- value density data, separating the area into a dot part, a non-dot part, a photograph part and a non-photograph part respectively and ANDing the non-dot part and the non-photograph part. CONSTITUTION:An inputted picture signal is converted into a digital multi-value density data at an A/D converter 1 and fed respectively to a dot/non-dot separate circuit 2, a photograph/non-photograph separate circuit 3 and a dot- character/non-dot-character separate circuit 4. An overall discrimination circuit 5 ANDs the result of dot/non-dot separation outputted from the dot/non-dot separate circuit 2 and the result of photograph/non-photograph separation outputted from the photograph/non-photograph separate circuit 3 and extracts a picture area being a non-dot and a non-photograph part as a white background character part. Then the circuit 5 discriminates the white background character part and the dot character part separated by the dot-character/non-dot-character separate circuit 4 as the character area and discriminates other areas as the non-character area and outputs an area discrimination signal representing character/non-character corresponding to each area.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a character area separation method for accurately separating character parts from an image containing a mixture of characters, halftone dots, and photographs.

[Conventional technology]

In general, a document image usually contains a mixture of characters consisting of line drawings, halftone dots consisting of dots, continuous tone photographs, and the like. When such a mixed image is reproduced using a digital copying machine or facsimile, for example, in order to improve the image quality of the reproduced image, pre-processing includes sharpening processing by edge emphasis on the text, and It is desirable to adaptively apply dither processing to remove moiré and reproduce gradation for non-text portions other than the above. To do this, first,
It is necessary to accurately separate the input image into text parts and non-text parts.

Conventionally, many methods have been proposed for separating text from an image containing the above-mentioned text, yj points, and photographs. All of these conventional methods basically consist of a block for determining whether it is a photo area or a non-photo area, a block for determining whether it is a halftone area or a non-halftone area, and a block for determining whether it is a halftone area or a non-halftone area. Based on the results, areas that are determined to be non-photograph areas and non-halftone areas are usually configured with blocks that are determined to be character areas.

For example, to give an example of a conventional character region separation method, r
There is a binarization processing method for images containing a mixture of text, photographs, and halftone dot printing (Toshihiko Minamiguchi; Proceedings of the National Conference of the Institute of Image Electronics Engineers, 1989, 22 + pp91-pp94).

This method first separates the image into photographic areas and non-photographic areas by detecting the edges of black and white transition points of pixels in the image, then binarizes the separated non-photographic areas, and then converts the image into two parts. By performing pattern matching on an image based on the distribution state of black pixels and white pixels, character areas (non-halftone areas) and halftone areas are separated.

When separating text areas (non-halftone areas) and halftone areas as described above, in order to further improve the separation accuracy, it is difficult to improve the separation accuracy not only with the method proposed by Minamiguchi above but also with any text area separation method. , - It is necessary to perform area correction on the separation result based on the separation result of surrounding pixels during boat fishing. By performing area correction based on the surrounding pixels, it is possible to further reduce the probability of erroneously separating a halftone dot part into a text part and a text part into a halftone dot part.

[Problem to be solved by the invention]

By the way, the characters in the image include a character part with a white background (hereinafter referred to as "white character part J") and a character part with a background of halftone dots (hereinafter referred to as "halftone character part"). When the above area correction is performed on such an image, since the background of the halftone text part consists of halftone dots, the halftone dot pixels around the character are first determined as halftone dots, and the halftone dots around this character are first determined as halftone dots. Since area correction of character pixels is performed based on the pixel determination results, there is an extremely high probability that character pixels existing on halftone dots will be incorrectly determined as halftone dot areas, making it difficult to accurately extract halftone text. The problem was that I couldn't do it.

The present invention has been made based on the above circumstances, and an object of the present invention is to enable accurate extraction of both white text and halftone text from an image containing a mixture of text, halftone dots, and photographs. The purpose is to provide a region separation method.

[Means to solve the problem]

The character area separation method of the present invention, as shown in the principle in FIG. The areas are separated into a halftone and non-halftone part, a photographic part and a non-photographic part, a halftone text part whose background is halftone dots, and a non-halftone text part whose background is other than that, and the logic of the non-halftone part and non-photographic part is divided into By calculating the product, an image area that is a non-halftone area and a non-photo area is determined to be a white text area with a white background, and both the white background text area and the halftone text area are determined to be character areas. It was designed to be extracted.

[Operation] The input image is converted into digital multivalued density data and separated into halftone dot areas and non-halftone dot areas, photographic areas and non-photographic areas.

To divide the input image into halftone and non-halftone areas, the extreme pixels that give the changing points of the peaks and troughs of density are detected based on the digital multilevel density data, and the density of the extreme pixels in a predetermined small area is calculated. It is determined from the information whether the pixel of interest belongs to a halftone dot candidate area or a non-halftone dot candidate areaj7, and based on the determination results of surrounding pixels, whether the pixel of interest in a predetermined area belongs to a halftone dot area or a non-halftone dot area. All you have to do is decide which one it is.

Also, to separate the input image into photo and non-photo parts,
The input image is classified into five black pixels and white pixels based on digital multilevel density data, and connected black pixels and connected white pixels are respectively detected from the classified black pixel group and the own pixel group. It may be determined whether the pixel of interest is in a photographic area or a non-photographic area based on the number of connected black pixels and connected white pixels existing in the vicinity.

The non-halftone dot area separated as described above includes a text area and a photographic area, which are images other than halftone dots. On the other hand, the non-photograph area includes a text area and a halftone dot area, which are images other than photographs. Therefore, by performing the logical product of the non-halftone dot area and the non-photograph area and extracting only the image area that is the non-knit area and the non-photograph area, only the text area can be extracted from the image.

However, the text extracted as described above is only the so-called white text with a white background, and the so-called silk text with a background of halftone dots is not extracted. This is because the above-mentioned area correction is performed in order to improve the separation accuracy in the separation processing of the area. That is, in the separation process of halftone dot areas and non-halftone dot areas, is each pixel in the image a halftone dot candidate area or a non-halftone dot candidate? After determining whether any part of the nf area is Ja or not, it is further determined whether it is a halftone part or a non-halftone part based on the determination results of a plurality of surrounding pixels. Area correction is performed by For this reason, 1. The halftone text portion whose background is made of knitting dots is corrected as a halftone dot.

This is because characters existing on the halftone dots cannot be separated.

Therefore, in order to separate and extract not only the white text part but also the halftone text part whose background is made up of m points, a separation process for the halftone text part is added in addition to the separation process for the four character parts on a white background. In order to separate this halftone text portion from other halftone text portions, the digital multilevel density data is smoothed using a predetermined weighting coefficient, and then the density changes in a predetermined local area are calculated based on the density changes within the image. , and determine whether the ridge pixel of interest is a character part of a background halftone dot or a character part of a non-halftone dot background based on the density information of the ridge pixel.

Then, the white text portion and halftone text portion extracted as described above are determined and output as a character region.

This/,,':Yo, \input image. Add the juices from the white text and net text inside! <Can be separated.

(Embodiments) Hereinafter, embodiments of the present invention will be described with reference to Figures 6 and 6.

FIG. 2 shows a block diagram of one embodiment of the invention. In the figure, 1 is an A/D conversion circuit, 2 is a halftone/non-halftone separation circuit, and 3
4 is a photo/non-photo separation circuit, 4 is a screen character/non-screen character separation circuit, and 5 is a comprehensive determination circuit.

The input image signal is converted into digital multi-level density data in the A/D conversion circuit 1, and then processed through a halftone/non-halftone separation circuit 2, a photo/non-photo separation circuit 3, and halftone/non-halftone text. The signals are respectively sent to the separation circuit 4.

The halftone dot/non-halftone dot separation circuit 2 separates the input image into a halftone dot portion and a non-halftone dot portion based on the digital multilevel density data, as will be described later. Further, the photo/non-photo separation circuit 3 is
Based on the digital multilevel density data, the input image is separated into photographic parts and non-photographic parts as will be described later. moreover,
The halftone text/non-halftone character separation circuit 4 separates the input image into halftone text portions and non-halftone text portions based on the digital multilevel density data, as will be described later. The separation results of each of these separation circuits are sent to the comprehensive judgment circuit 5.

The comprehensive judgment circuit 5 calculates the logical product ( AND) is performed, and the image area that is a non-halftone dot and non-photograph area is extracted as a blank character area. Then, both this white text area and the halftone text area separated by the halftone text/non-halftone text separation circuit 4 are determined to be character areas, and the other areas are determined to be non-text areas, and each area is Correspondingly, a character/non-character area determination signal is output. Therefore, if each pixel of the input image is subjected to sharpening processing for characters or dithering processing for halftone dots and photographs, according to this area determination signal,
High quality reproduced images can be obtained.

Next, the separation process of the halftone dot portion and the non-halftone dot portion in the halftone dot/non-halftone dot separation circuit 2 will be explained with reference to FIGS. 3 to 6.

FIG. 3 shows a block diagram of a specific example of the halftone/non-halftone dot separation circuit 2. As shown in FIG. The extreme point pixel detection circuit 201 detects pixels (peak pixels) corresponding to the extreme points of the peaks and valleys of halftone dots in the input image, using density information within a predetermined local area. That is, for example, in the 3×3 block shown in FIG. 4, when the density level of the center pixel is the maximum (minimum) compared to the density level of the surrounding eight pixels, the center pixel is defined as the density peak (valley). ) is detected as an extreme point pixel.

The halftone candidate area detection circuit 202 determines whether the pixel of interest belongs to a halftone candidate area or a non-halftone candidate area based on the density information in a predetermined small area of the peak and valley extreme pixels detected by the extreme pixel detection circuit 201. Determine whether For example, the fifth
As shown in Figure (a) or (b), the peak and valley peak pixels within a 4x4 small area are counted. Then, when the total value of the peak and valley peak pixels exceeds a predetermined threshold, as shown in Figure (a)
When processing in block units, all pixels in the small area are detected as halftone dot candidate areas, and in case of processing in pixel units as shown in figure (b), predetermined pixels in the small area are detected as halftone dot candidate areas. A pixel of interest (hatched pixel) is detected as a halftone dot candidate area.

The halftone dot area detection circuit 203 determines, based on the detection result of the halftone dot candidate area detection circuit 202, whether the pixel of interest in a predetermined area belongs to a halftone area or a non-halftone area. For example, as shown in FIG. 6(a), when the determination is made in block units, the number of small areas that are halftone candidate areas among the small area of interest and the surrounding small areas reaches a predetermined threshold. When it is larger than the value, all pixels in the small area of interest are determined to be a halftone area. Furthermore, as shown in FIG. 6(b), when the determination is made pixel by pixel, the number of pixels designated as halftone dot candidate areas among the pixel of interest (/% latching pixel) and its surrounding pixels is If it is larger than the threshold, the pixel of interest is determined to be a halftone dot area. As a result, halftone dots and non-halftone dots are separated for all pixels in the input image.

Next, an example of separation processing of a photograph part and a non-photo part in the photograph/non-photo separation circuit 3 will be explained with reference to FIGS. 7 to 9.

FIG. 7 shows a block diagram of a specific example of the photo/non-photo separation circuit 3. As shown in FIG. The binarization circuit 301 binarizes input digital multi-value density data using a specific threshold value and converts it into a black and white binary image. That is, by this binarization, all pixels on the input image are classified as white pixels or black pixels.

The black pixel pattern matching circuit 302 detects connected black pixels from the binary image, and determines whether a black and white pattern in a 3×3 matrix centering on the pixel of interest is one of the pixel patterns shown in FIG. When matched, the pixel of interest is determined to be a connected black pixel, and an output is generated.

The white pixel pattern matching circuit 303 detects connected white pixels from the binary image, and determines whether a black and white pattern in a 3×3 matrix centered around the pixel of interest is one of the pixel patterns shown in FIG. When matched, the pixel of interest is determined to be a connected white pixel, and an output is generated.

The counting circuit 304 is a black pixel pattern matching circuit 30
Based on the output signal No. 2, connected black pixels existing in a 3×3 matrix centered on the pixel of interest are counted, and an output is generated when the count value becomes, for example, 2 or more. Similarly, the counting circuit 305 counts the connected white pixels existing in a 3×3 matrix centered on the pixel of interest, and generates an output when the count value becomes 2 or more, for example.

Then, the counting circuits 304 and 305 are ANDed, and two pixels are placed in a 3x3 matrix with one pixel as the center.
If more than 10 connected black pixels and connected white pixels exist at the same time, the pixel of interest is determined to be a non-photo candidate pixel; otherwise, it is determined to be a photograph candidate pixel, and the AND circuit 30
6 generates an output corresponding to each case.

The determination circuit 307 uses, for example, a 5×5 pixel centered on the pixel of interest.
For example, if there is one or more non-photo candidate pixels in the matrix, the pixel of interest is determined to be a non-photo pixel,
If there is no such pixel, the pixel of interest is determined to be a photographic pixel. As a result, photographic parts and non-photographic parts are separated for all pixels in the input image.

Next, an example of separation processing between a screen character part and a non-screen character part in the screen character/non-screen character separation circuit 4 will be described with reference to FIG. 10 and FIG. 12.

FIG. 10 shows a block diagram of the screen character/non-screen character separation circuit 4. As shown in FIG. The smoothing processing circuit 401 smoothes input digital multi-level density data using a predetermined weighting factor.

The ridge pixel detection circuit 402 detects the smoothed digital multilevel density data, for example, in a 5×5 pixel (not shown in FIG. 11).
For example, LcL, +≧TH1 and Le-Lzz≧Tl+ or Lc Ls≧TI, and Lc=L**≧TI or Lc Ls≧ TH, and Le-Lzn≧TH or Lc-L, ,≧Tl+, and Lc-1,,4≧TI However, if the − pair of TH:L threshold also holds, then its center image); , 17 is detected as the ridge pixel of the character as shown in FIG.

The determination circuit 403 counts the ridge pixels detected by the ridge pixel detection circuit 402 in units of, for example, 9×9 blocks, and when the counted value exceeds a predetermined threshold value, the determination circuit 403 counts the ridge pixels detected by the ridge pixel detection circuit 402, and when the counted value exceeds a predetermined threshold, the A pixel (center pixel or all pixels in a block) is determined as a character pixel on a halftone dot. In this way, by performing processing after smoothing the input image using a predetermined weight factor, it is possible to separate the halftone text portion whose background is halftone dots from the image.

〔Effect of the invention〕

As is clear from the above, when the present invention is used, it is possible to accurately extract both the white character part and the halftone character part in an image as character areas, so a character separation method with high separation accuracy can be used. Can be offered.

[Brief explanation of the drawing]

Figure 1 is a Kyori diagram of the present invention, Figure 2 is a block diagram of an embodiment of the present invention, and Figure 3 is a halftone dot/
A block diagram of a specific example of a non-halftone dot separation circuit, FIG. 4 is a diagram showing an example of a local N area and a pole pixel, FIG. 5 is an explanatory diagram of detection of halftone dot area candidates, and FIG. An explanatory diagram of detection. Figure 7 is a block diagram of a specific example of a photo/non-photo separation circuit. Figure 8 is a diagram showing a pattern for detecting connected black pixels. Figure 9 is a pattern for detecting connected white pixels. Figure 1θ is a block diagram of a specific example of a screen character/non-screen character separation circuit; Figure 11 is a diagram showing an example of a ridge pixel detection pattern; Figure 12 is a diagram showing an example of a ridge pixel detection pattern;
The figure is an explanatory diagram of ridge pixels. l... A/D conversion circuit, 2... Halftone/non-halftone separation circuit, 3... Photo/non-photo separation circuit, 4... Halftone text/non-halftone text separation circuit, 5 ... Comprehensive judgment circuit.

Claims (4)

[Claims] (1) After converting the input image into digital multi-value density data, based on the digital multi-value density data, the input image is divided into halftone dot areas and non-halftone areas, photographic areas and non-photographic areas, and the background is halftone dots. The halftone text area and the background are separated into the other non-halftone text areas, and by taking the AND of the non-halftone area and the non-photograph area, an image that becomes a non-halftone area and a non-photograph area is created. A character region separation method characterized in that the region is determined to be a white character portion having a white background, and both the white character portion and the halftone character portion are determined to be character regions and extracted. (2) When separating the input image into halftone areas and non-halftone areas, detect the extreme point pixels that give the changing points of the peaks and troughs of density based on the digital multilevel density data, and select the extreme points in a predetermined small area. It is determined whether the pixel of interest belongs to a halftone dot candidate area or a non-halftone dot candidate area based on pixel density information, and based on the determination results of surrounding pixels, the pixel of interest within a predetermined area is determined to be a halftone dot candidate area or a non-halftone dot candidate area. 2. The character area separation method according to claim 1, wherein the character area separation method determines whether the character area is a dot part. (3) When separating the input image into photographic parts and non-photographic parts, the input image is classified into black pixels and white pixels based on digital multilevel density data, and from the classified black pixel group and white pixel group, The system detects connected black pixels and connected white pixels, respectively, and determines whether the pixel of interest is in a photo area or a non-photo area based on the number of connected black pixels and connected white pixels that exist in the vicinity of the pixel of interest. A character area separation method according to claim (1), characterized in that: (4) When separating the input image into character parts whose background is halftone dots and character parts whose background is non-halftone dots, after smoothing the digital multilevel density data with a predetermined weighting coefficient, The ridge pixel in the image is detected based on the density change, and based on the density information of the ridge pixel, it is determined whether the ridge pixel of interest is a background halftone dot character part or a background non-halftone dot character part. The character area separation method according to claim (1), characterized in that: