JPH03126181A - Area dividing method for document image - Google Patents

Abstract

PURPOSE:To improve attribute discrimination accuracy by repeating segmentation processing with a circumscribing rectangle detecting means, performing classification with an attribute discriminating means, and unifying a circumscribing rectangle to a block with a character unifying means. CONSTITUTION:The circumscribing rectangle detecting means 31, when optimally binarized image data is inputted, makes the information areas of a field separator, a photograph, a chart, and a character part, etc., in the data into the circumscribing rectangles, and detects the feature of the circumscribing rectangle and that of a binary image in the circumscribing rectangle. After that, the circumscribing rectangle discriminating means 33 discriminates the attribute of the field separator, the photograph, the chart, and the character part, etc., in the information area, and furthermore, automatically sets a unified threshold value from the image obtained by extracting only the character part out of the attribute corresponding to individual document image data. Following that, a character part unifying means 37 unifies the circumscribing rectangle of the character part to a character part block. Thereby, it is possible to obtain an area dividing method for document image with easy selection of the threshold value.

Description

Detailed Description of the Invention (Industrial Application Field) The present invention is used for communication equipment such as facsimile and telex, and image database input method, etc. The present invention relates to a method for dividing a document image into regions by extracting information regions such as the following, and classifying the attributes of the regions into 111 categories.

(Prior art) Conventionally, technologies in this field include: ■1.゛j Publication No. 1987-71379, (Japanese Patent Publication No. 10337/1982
Some items were listed as 2nd prize. The configuration will be explained below using figures.

FIG. 2 is a block diagram illustrating an example of the configuration of an image processing apparatus using the conventional document image area division method described in the document (2). This image processing device includes a line memory 1 consisting of a plurality of lines, and cumulative addition circuits 2a and 2b consisting of registers and addition circuits.

2c, determination circuits 3a, 3 for comparison with a predetermined threshold value;
b, 3c, 3d, 3e, 3f, multi-stage shift register 4, gate time 1? ! 5. Differential circuit B6a, 6b consisting of a register circuit and an addition circuit, run-length counter circuit 7
, a run length memory circuit 8, and eight white run length generation number craft circuits. In this type of method, when image data g is input, the image data g is stored in the line memory 1. Thereafter, black pixels are counted in the main scanning direction by the cumulative addition circuit 2a. The cumulative viewing 2
In the process in which the output of M2a passes sequentially through the determination circuit 3a, the differentiation circuit 6a, and the determination circuit 3b, a line exceeding the count threshold is detected, and a state in which a predetermined number of consecutive white lines with the count value equal to or less than the above continues. After the determination, first region cutting is performed.

Next, in the first region extraction 6, the output of the determination circuit 3b is transferred to the gate circuit 5, the shift register 4, the cumulative addition circuit 2b
, the judgment circuit W@3c, the differentiation circuit 6b, and the judgment circuit 3d are sequentially passed through, black pixels are counted in the sub-scanning direction, a column whose count value exceeds the threshold value is detected, and the above count value or the threshold value is detected. The second area extraction is performed by determining a state in which a predetermined number of the following self-interests are continuous.

Furthermore, a third region is cut out using the same process as the second region cutout, and then a fourth region is cut out using the same process as the third region cutout. Therefore, the fourth
For the area detected by area extraction, the character part,
There was a method to determine the attributes of the photo and diagram areas.

(Problems to be Solved by the Invention) However, the method for dividing document images into regions having the above configuration has the following problems.

When dividing a region, it is necessary to arbitrarily set parameters for setting the overall mean value THX. For example, when performing region division processing on individual document image data such as academic journals, it is necessary to set the parameters. I had to fix it.

Furthermore, there is a problem in that it takes time and effort to select appropriate parameters according to individual journals, etc.

The present invention provides a method for dividing document images into regions that solves the problem of the prior art, which is that it takes time and effort to divide regions.

(Means for Solving the Problems) According to the present invention, in order to solve the above problems, an information region including a character part of input image data is cut out, and an attribute determination process of the information region is performed on a document. In the method for dividing an image into regions, binary stroke data that has been optimally scaled to the image data is input to a circumscribing rectangle detecting means, and the circumscribing rectangle detecting means detects the information area in the binary image data. is converted into a circumscribing rectangle, the features of the circumscribing rectangle and the features of the binary image within the circumscribing rectangle are detected, and the attribute determining means determines the information area from the features of the circumscribing rectangle and the features of the binary image within the circumscribing rectangle. The attributes of the character portion are determined by the attribute determining means, and the circumscribed rectangles determined to be the character portion among the attributes by the attribute determining means are integrated into the character portion block using the literature uniting means. Furthermore, the literature department integration means
A surrounding histogram is created for a text image obtained by extracting only the text from value image data, and the circumscribing rectangle of the text is integrated into the text block using the surrounding histogram. . The literature integration means determines from the creation result of the peripheral histogram that there is a white line area in which there are continuous white lines in which the black pixel does not exist in the horizontal direction, and a black line area in which there are continuous black lines in which at least one black pixel exists in the horizontal direction. A pixel line region, a white column region in which there are consecutive white columns in which no black pixels exist in the vertical direction, and a black pixel column region in which there are consecutive black columns in which at least one black pixel exists are extracted, respectively. ,
From the extraction results, horizontal writing/vertical writing is determined, horizontal/vertical integration thresholds are determined based on the determination results, the circumscribed rectangle of the character part is integrated based on the integration threshold, and the result is This is extracted as a partial area.

Further, the determination of vertical writing/horizontal writing is based on the creation result of the peripheral histogram. This is done based on the number of consecutive white column areas.

Then, the integrated threshold in the vertical and horizontal directions is set to 5 from the histogram of the white line area width and the black line area width when horizontal writing is determined, and when it is determined as fir writing, the white column area Each width is determined from a histogram of the width and the width of the ring column region.

The integrated threshold in the vertical and horizontal directions is calculated by scanning from the white line or self-cutting ram area with the largest number in the direction of increasing width based on the histogram of the width of the white line or white column area. After detecting the width of the white line or white column area where the number is 0, extract the white line or self-cutting ram area immediately before that as a value TW, and then calculate the value for the histogram of the width of the black line or ring column area. TB is extracted, and based on the value 1゛W and the value TB, in the case of horizontal writing, the total horizontal distance value THX = TW +
TB, [total value between directions THY=TW+Er (however, E
r: the number of lines for the tilt error), and in the case of vertical writing,
Total horizontal distance 1t! THX=TW+Er (comprehensive longitudinal value T) (Y=TW+TB).

Furthermore, as the integration process, two area coordinates (XSI, YSI) (Xe1, Yet) of the circumscribed rectangle of interest,
(XS2.YS2>(Xe2.Ye2) (However, XS
1. , XS2. Y], YS2; Starting point coordinates of the area in the horizontal and vertical directions, Xe1, X←2, Ye 1. Ye2; End point coordinates of the horizontal and vertical regions), the circumscribed rectangles determined to be the character portion are successively combined into one, based on the following integration conditional expressions (1) to (4), and integrated to the limit. After repeating this, the result is extracted as the character part pro.../.

Integrated conditional expression % expression % (however, THX, horizontal integrated threshold) (2) YS2-Ye l≦TI
X; horizontal integration threshold) (3) XSL-Xe2≦THX (2) YS2≦THY (however, THY: vertical integration threshold) (3) XS l ,
YS2-Ye lX5l-Xe2. When the value of YSl-Yel is negative, the condition is satisfied) (for f'13) Since the present invention has configured the document image area division method as described above, the circumscribing rectangle detection means can The data is optimally binarized to 2ft to avoid losing the characteristics of the document image data! Enter image data. The circumscribing rectangle detecting means converts information areas such as field separators, photographs, charts, and character parts in the two-image image data into a circumscribing rectangle, and converts the characteristics of the circumscribed rectangle and the characteristics of the two quadrants and one image within the circumscribed rectangle. and detect. Furthermore, the attribute determining means determines the field separator of the information area, the photo,
Determine the attributes of figures, tables, text, etc.

Then, the attribute determining means automatically sets a threshold value in place of a parameter for dividing the image into regions from the image obtained by extracting only the character part from among the attributes, according to each document image data, and The circumscribed rectangle of the text section is integrated into the text section block using the literature section integration means.

Therefore, the above-mentioned imposed pressure can be solved.

(Embodiment) FIG. 1 is a functional block diagram of a method for dividing a document image into regions according to an embodiment of the present invention.

This method of dividing the area of a document image is to
Optimal binarization processing means 20 for obtaining a binary image by increasing the value by 1;
It is comprised of a digital processor 7r, and includes an area extraction/discrimination unit 30 for inputting the binary image and extracting an area and determining attributes. The optimal binarization processing means 20 is a CCD (CI) large couple.
The apparatus is equipped with a reading unit 21 which has an image sensor of $ 1), and scans one information medium with the image sensor to detect one injection on the multi-value side. The first output of this reading unit 21 includes a multi-level image memory 22 such as RAM that stores the multi-valued images that have been sold, and a multi-level image memory 22 that stores the multi-valued images. Optimally binarizes the image to create two images with a certain level of image quality without losing their characteristics and properties.
Optimal binarization means 23 for converting into a value image is connected. Furthermore, on the output side of the optimal binarization means 23, there is an RA that stores the binary image converted by the optimal binarization means 23.
A binary image memory 24 such as M is connected.

The area extraction/discrimination processing means 30 that scans the binary image stored in the binary image memory 24 inputs the binary image obtained by optimal binarization, and extracts the information area of the binary image (for example,
A circumscribing rectangle detecting means 31 is provided for converting a black pixel area into a circumscribing rectangle and extracting the circumscribed rectangle.

The output side of the circumscribed rectangle detection means 31 includes a RAM for storing the position of the circumscribed rectangle detected by the circumscribed rectangle detection means 31.
A circumscribed rectangle position memory 32 such as the above is connected to an attribute discrimination means 33 for classifying the circumscribed rectangle into diagrams, photographs, field separators, and others (characters) based on the features of the binary image and the features within the rectangle. .

On the output side of the attribute determination means 33, there are a diagram position memory 34, a photograph position memory 35, a field separator 36, and a literature department integration means 37 that store the positions of circumscribed rectangles for each attribute determination of diagrams, photographs, field separators, and characters. connected sequentially.

Faculty of Literature integration means 37 has attribute 1゛! This means integrates the circumscribed rectangle determined to be a character part by the separate means 33 into a character part block, and extracts the result as a character part area. E: The memory 38 is connected.

The area processing method for the book image configured as described above will be explained with reference to FIGS. 3 to 20.

(I) The operation reading unit 21 of the optimum 21-value conversion processing means 20 scans the information medium and outputs a multi-value image image, and stores it in the multi-value side one-image memory 22.

The reading resolution of the reading unit 21 is moderately thin, and a resolution of 8 lines/mm or more is appropriate. The optimal binarization means 23 reads the multi-value image data in the multi-value side one-image memory 22, and determines the magnitude of the absolute value of the density difference value in the local area in the multi-value image data, and distinguishes between the halftone image part and the non-halftone image part. It is classified into halftone image parts. In this optimal binarization means 23, if it is determined that it is a non-halftone image part, it is simply binarized using a fixed threshold value (simple binarization), whereas if it is determined that it is a halftone image part, it is diluted. Binarization is performed using the method to generate a binary image. These optimal binarization processes may be performed, for example, by the method described in the above-mentioned document (2).

Through the above optimal binary processing, non-halftone image parts such as characters and charts become images with sharp edges of black and white, while halftone image parts such as photographs, etc.
This is the dither method seen in newspapers, etc., and by combining these methods, a binary drawing with a certain level of image quality can be obtained. This binary image is stored in the binary image memory 24.

(II) Optimal operation of circumscribed rectangle detection means 31 2 Binary image memory 2 obtained by suppression means 23
The main scanning direction of the binary image data in 4 is the X-axis direction, and the sub-scanning direction is the Y-axis direction.

The operation of the circumscribed rectangle detection means 31 will be described below with reference to FIGS. 3 and 4.

FIG. 3 is a flowchart showing an example of the operation of the circumscribed rectangle detection means 31 in FIG. 1, and FIG. 4 is a flowchart illustrating the circumscribed rectangle detection means 31. In addition, 350~S in Figure 3
67 indicates a processing step, i in FIGS. 3 and 4.
is the hierarchy, m is the area number cut out in the X direction, and n is YJ
This is an area cut out in the J direction. Further, 40 in FIG. 4 is a background pattern, and 41 is an information area (black pixel area).

First, at 350 to S52 in FIG. 3, i.

Initialize n and m to the initial value 1, respectively. Then ζ, the starting point coordinates of the area (XS (i-1, m), YS (i-
1,n)1. End point coordinates (Xe (i-L m).

For Ye (i-1, n)l, the area in the Y-axis direction is n
The process n1 of cutting out the area into several areas is performed (353).

This cutting process will be explained using a diagram illustrating the cutting process in the Y direction in FIG. 5. Starting point coordinates (XS
(i-1, rtt), YS (i-1, n)),
End point coordinates (Xe (i-1, m), Ye (i-1゜n
A black line in which at least one black pixel exists in the X direction is detected for the information area 41 on the background pattern 40 indicated by >). The Y coordinate value of the black line is extracted as the starting point Y coordinate, and then a white line in which there are no black pixels in the X direction consecutively exceeds the threshold value TH is detected. The Y-coordinate value of the black line (TH+l) before that day's line is extracted as the end point Y-coordinate value. Here, if the halftone density information is quantized to, for example, 4 bits per pixel (16 gradations), the value of the threshold TH is 1 to 4 from the viewpoint of resolution.
It is appropriate to set it within the range of the image.

The above operations are performed in the area (XS (i-1, m), YS (i-1, m),
1, n)), (Xe (i-1, m), Ye(i-1,
n)) for all YFi, marks, and calculate the area of n leap (
XS (i-1, m>, YS (i-1゜n)), (X
e (i-1, m> , Ye (i-1n)) (IB
Then, cut out n (positive integer). At this time, the area (XS
(0, m), YS (0, n)), (Xe (0, m
>, Ye (0, n>) are set in advance as initial values, and the values indicate the entire area of the binary image data.

If a black line is detected before the number of consecutive white lines detected exceeds the threshold TH, the upper and lower areas are determined to be the same area, and processing is repeated until the number of consecutive white lines exceeds the threshold. . Also, the Y coordinate value of the black line is )'e
(i-1,n), then Ye(i 1
．． r+) as the end point)' coordinate value, and the Y coordinate value is Y
When e(i-1, n) is reached, the process ends.

In this way, °ζ, the area (XS (i-1, m).

YS (i-1, n> ), (Xe (i 1, n
+).

With respect to Ye (iL, n>), for the above n regions cut out in the Y direction, the Y coordinates of the n starting points and end points are stored in the circumscribed rectangle position memory 32 (854
). The area cut out in the )' direction, where the value of n when i=1 is 1.2.3 shown in FIG. 4, shows an example of the processing result.

Next, at 353 in FIG. 3, the following processing is performed on the n regions cut out in the Y direction. First, the value of n is initialized to 1 (S55).

Then, the area starting point coordinates (XS (i-1, m>, YS (i
, n)), end point coordinates (Xe (i-L, m).

For Ye (i, n>), perform processing to cut out the region into multiple (m) regions in the X-axis direction <856)
. This cutting-out procedure will be explained using a diagram illustrating the cutting-out procedure in the X-axis direction in FIG. 6.

For the information area 41 on the background pattern 40 indicated by the start point coordinates and end point coordinates, detect a black column in which at least one black pixel exists in the Y direction, and extract the X coordinate value of the black column as the start point X coordinate. do. Next, a white column in which there is no black pixel in the Y direction exceeds the threshold value TH is detected, and the X coordinate value of the black column (TH+1") before that white column is set as the end point X coordinate value. Here, the value of the first threshold TH is set in the range of 1 to 4 pixels.

The above operations are performed in the area (XS (i-1, m), YS (i,
n)), (Xe (i-1, m), Ye (i.

n)) for all X coordinates of m regions (XS
(t, m), YS (i, r+)), (Xe(i, m
>, Ye (i, n>) (where m: a positive integer) is cut out (856). At this time, the number of consecutive white columns detected is 1-1 (if a black column is detected before the direct TH This is repeated until TH exceeds the threshold TH.Also, if the X coordinate value of the black column is equal to Xe(i-], n>, extract Xe(i-1, n> as the end point X coordinate value, When the coordinate value becomes Ye (i-1, n), the process ends.

In this way, the area (XS(i~1.m>).

YS (i, n)), (Xe (i-1, m>, Ye(
With respect to i, n)l, the X coordinates of the starting point and ending point of mrli are stored in the circumscribed rectangle position memory 32 for the m regions cut out in the X direction (S32). i=1 shown in FIG. An example of the processing result is shown in an area cut out in the X direction where the value of m is 1.2 when rt=1.

If the above procedure is performed and the values of m and n in layer i are each only 1.1 (358), area start point coordinates (XS (i, m), YS (i, n)) , the end point coordinates (Xe (t, m), Ye (i, n)) are extracted as a circumscribed rectangular area (559), and the start point and end point coordinates are stored in the circumscribed rectangle position memory 32 (S60). Next, the value of this layer is decremented (561), and it is determined whether this has been done for all n in the i-th layer (S62).

If it has been done for all n, then it is determined whether it has been done for all m in the i-th hierarchy (')363)
, if the process has been performed for all m, all the processes in FIG. 3 are completed (s64).

On the other hand, if the answer is 358, the value of the floor Ni is increased by 56.
7) Then, the process returns to S52. If the answer is NO in S62, the value of n is incremented by 565), and the process returns to S56. Similarly, if the answer is NO in 363, the value of m is incremented by 5.
66), return to 353.

The above is a series of processing by the circumscribed rectangle detection means, and the processing is performed using hierarchical means as shown in FIG. in short,
The area surrounded by the white line and white column above the threshold TH is
Repeat the process until there are no more cuts left! -, the finally cut out area (the entire area of the binary image margin data) is detected as a circumscribed rectangle, and the start point coordinates and end point coordinates indicating the positions of all the circumscribed rectangles are stored in the circumscribed rectangle position memory 32. It is something to do.

(I [I) The movement attribute discriminator 33 of the attribute discriminator 33 reads out the contents of the circumscribed rectangle position memory 32, and for all of the detected circumscribed rectangles, fields are determined from the circumscribed rectangle features and the features within the rectangle. A process is performed to classify the data into four areas: separators, photographs, diagrams, and letters.

An example of this process is shown in FIG. FIG. 7 is a flowchart of the operation of the attribute determining means 33 of FIG. 1. Note that all circumscribed rectangles are (XSj, YSj), (Xej, Yej
) (([!L, j; positive integer).

First, the values of j are initialized by setting each initial value to 1 (S70). Next, the circumscribed rectangle (XSj.

YSj) and (Xej, Yej), a process of extracting the feed separator is performed (S71).

If the circumscribed rectangle is not determined to be the feed separator, photograph and chart candidates are extracted (S72).

Therefore, if the circumscribed rectangle is determined to be a photograph and diagram candidate, processing for determining the photograph and diagram is started (873
). As a result, if it is determined that the input circumscribed rectangle is a photograph, the start and end point coordinates of the circumscribed rectangle are stored in the photograph position memory 35 (S74). On the other hand, the circumscribed rectangle determined to be a diagram is stored in the diagram position memory 34 (S75). Furthermore, all the circumscribed rectangles (XSj
, YSj), (Xej, Yej) (876), and when it is determined that the process has been performed for all circumscribed rectangles, this process ends (S77). S76
If no, increment the value of j by 578), then S
Return to 'l processing. If it is determined as a field separator in the process of S71, the start point and end point coordinates of the circumscribed rectangle are stored in the field separator position memory 36.
79), perform the process of S76. If it is not determined to be a photo or chart candidate in the process of step 72, the coordinates of the starting point and end point of the circumscribed rectangle are stored in the character position memory 37 (580), and then the process of 376 is performed.

The above is a series of operations of the attribute determination means 33, and 871, S72, and S73 will be explained in detail with reference to FIGS. 8 to 15.

FIG. 8 is a flowchart showing an example of the field separator extraction operation in S71, and FIG.
It is a relationship diagram of the X-axis length with respect to the axial length.

As shown in Figure 9, the field separator has an It can be seen that extraction is possible under the condition that the length of the Y axis is small.

According to the above conditions, field separators are extracted using the hand JIIQ shown in FIG.

First, the circumscribed rectangle (XSj, YSj), (Xej.

(S90), and the length of the X axis is X1en.
, find the length Ylen of the Y axis (S91>.

The length X1en of the X axis is determined by Y
The length Ylen of the axis is determined by Yej-YSj+1.

Then, the obtained Ylen and X1en are compared (S
92>, if X1en is larger, it is determined that X1en is larger than the threshold γ (89B). If it is determined that the length of Ylen is smaller than the calculated value, then calculate the value of X1en
94), determines that this circumscribed rectangle is a field separator 595), and ends the process (896>, if no in S92, determines whether Ylen is larger than the threshold γ, as in 593 (S97); if yes in S97 , then 894
Similarly, find the value of Y l er t Xa + β, and
If it is determined that the length of en is smaller than the calculated value (898), then the process of step S95 is performed.

If S93, S94, S97, and S98 are each negative, the process ends (S96>).

FIG. 10 is a photograph showing an example of the operation of 372 in FIG.
Figure 11 is a diagram showing the number of times of reversal from white to black in the vertical and horizontal directions within the circumscribed rectangle with respect to the area of the circumscribed rectangle when the resolution is 8 lines/mm. be.

As shown by the horizontal axis in FIG. 11, characters, etc., and diagrams/photos can be distinguished by the size of the area of the circumscribed rectangle.

Therefore, based on this feature, diagram/photograph candidates are extracted using the means shown in Figure 1.0.

The circumscribed rectangles (XSj, YSj), (Xej, Ye
j) is input (3100), and then the area S of the circumscribed rectangle is determined by the product of the X-axis length X1en and the Y-axis length Ylen determined in S9 of FIG. 8 (Slot). If the obtained area S is smaller than the threshold C (S102>, this circumscribed rectangle is determined to be a character etc. 810B), the process ends (3104). If the area S is larger than the threshold C in 5102, this circumscribed rectangle is The rectangle is determined to be a diagram/photo candidate (
8105), and the process ends (SIO4).

FIG. 12 is a flowchart illustrating an example of the photo/diagram determination operation 873 in FIG.

For the extracted circumscribed rectangle that is determined to be a diagram/photo candidate in the process of S72, a process is performed to determine whether it is a photograph or a diagram.

As shown in FIG. 11, for a circumscribed rectangle whose area S is larger than the threshold value C (circumscribed rectangles whose area S is smaller than the threshold value C are excluded by S72), Y=aX+b (however, a, b; some fixed A photograph or a diagram can be determined depending on whether the number of inversions from white to black is larger or smaller than the boundary line represented by the value (X: area S).

FIG. 13 is a diagram showing the relationship between the area of a circumscribed rectangle and the number of black pixels within the rectangle.

Conventional methods for determining the attributes of characters, diagrams, and photographs often involve determining the black pixel rate based on the area and the number of black pixels within the rectangle, and determining the attributes based on the size of the black pixel rate.

However, when processing the optimally binarized image used in this example, the results shown in FIG. 13 are obtained.
This makes it difficult to set discrimination criteria for distinguishing between photographs and charts. Therefore, by utilizing the features shown in FIG. 11, photographs and charts are distinguished by the operating procedure shown in FIG. 12.

Input the circumscribed rectangle (XSj, YSj>, (Xej, Yej) determined to be a photo/diagram candidate at 372 in FIG.
SLio), first, the count is initialized by setting it to 0 (S111). Also, YSj in the binary image data
In the process of sequentially scanning from XSj to Xe, j while scanning from XSj to Yej, the count is incremented when white is reversed to black (S112). As is clear from the explanatory diagram in Fig. 14, this scans each Y in the X direction in the direction of the arrow, and increments the count when it reverses from white to black (pattern A to pattern Fi). It is. Furthermore, while scanning XSj in the same binary image data to Xe,i,
In the process of sequentially scanning from YSj to Yej, when white is reversed to black, the count is incremented (3113). As is clear from the explanatory diagram of FIG. 15, this scans each X' in the direction of the arrow, and increments the count when it is reversed from white to black (from pattern A to pattern B).

Next, the area S of the circumscribed rectangle obtained at 8101 in FIG.
If the value of SXa is calculated from 8112°5113 and it is determined that the count 1 is larger than the calculated value (S114), it is determined that this circumscribed rectangle is a photograph (S115). ), the process ends (S116).
. If it is determined in 5114 that the count value is small, it is determined that this circumscribed rectangle is a chart (8117), and the process ends (S116).

FIG. 16 is a flowchart of the literature department integration means of FIG.

First, from the 21 direct image data, the attribute determination means 33 determines whether the text portions other than text portions (photographs, diagrams, files, and field separators)
For the image obtained by deleting the image of the circumscribed rectangular area determined as (S120>. From the surrounding hiss I/crumb, there is a white line area where there are continuous white lines with no black pixels at all in the X direction, and a black pixel line where there are continuous black lines where there is at least one black pixel). In addition, in the Y direction, a white column area in which there are consecutive white columns without any black pixels, and a black pixel column area in which there are consecutive black columns in which at least one black pixel exists are extracted ( 8121).

Next, it is determined whether it is vertical writing or horizontal writing based on the size of the number of white line areas to be extracted and the number of white column areas.5122> Based on that determination, horizontal writing is determined. If it is determined that the text is written vertically, the information of the white line area and black line area is used. On the other hand, if vertical writing is determined, the information of the white column area and black column area is used.
Steps are taken to determine the threshold of the integration condition in the vertical direction (S1
23). The circumscribed rectangle determined to be a character part by the attribute determination means 33 is integrated into a character part block based on the conditions of the determined integration threshold, and the result is extracted as a character area (8
124>.

The above is a series of processes of the literature department integrating means 37, and each process will be explained in detail below.

5120.5121 will be explained using FIG. 17.

The image inside the frame is an image obtained by deleting the image of the circumscribed rectangular area that is determined to be other than the text part from the binary image 1-image data. Each Y
The peripheral distribution histogram showing the total number of black pixels in the X direction for each
The total number of black pixels in each direction is shown.

FIG. 17 is a diagram illustrating an example of the result of the above-mentioned marginal distribution and Stodarham f.

Then, from the peripheral histogram, as shown in the figure, a white line area where there are consecutive white lines in which no black pixels are detected in the X direction, and a black line area where there are consecutive black lines where at least one black pixel is detected are determined. , and similarly extract white column areas and black column areas in the Y direction. In the case of FIG. 17, no white column in which there is no black pixel in the Y direction was detected, so there is no white column area.

Next, to explain 5122, if it is determined from the number of extracted white line areas and the number of white column areas that 5121° is (number of white line areas - number of white column areas), then this document is If it is determined that the writing is horizontal or vice versa, it is determined that the writing is vertical.

8123 will be explained using FIG.

This Figure 18 shows an example of processing when horizontal writing is determined in 5122, and is a diagram explaining the integrated thresholds in the vertical and horizontal directions. The figure shows the width (number of lines) of the white line area in the X direction, and the number for each line area width in the Y direction.

Therefore, by scanning in the direction of increasing width from the white line area width that has the largest number (the area where the white line area width is 16),
When a white line area width whose number is 0 is detected, the white line area width immediately before that width is extracted as a value TW. In this figure, it is 19.

Further, similar processing is performed on the histogram of the black line area width to extract the value TB.

As a result, integrated thresholds in the horizontal and vertical directions are determined from the obtained values TW and TB.

The relationship is as follows: Total horizontal value THX=TW+TB Vertical total value THY=TW+Er (However, Er:
(number of lines equal to the slope error).

The above description of the means for determining the integrated threshold in the horizontal direction was based on the case where horizontal writing was determined. However, when it was determined that vertical writing TW and TB are extracted in the same way from the histogram of the width of the black column area, and in the case of vertical writing, the total distance in the horizontal direction is set to THX=TW+Er, and the total distance in the vertical direction is set to THY=TW+TB.

5124 will be explained using FIG. 19.

FIG. 19 is a diagram illustrating the integration of circumscribed rectangles of character parts.

The origin (0, 0> is set as a binary image (at the top left of the elephant data), and for adjacent circumscribed rectangles like pattern A and overlapping circumscribed rectangles like pattern B shown in the same figure, 512
Integration is performed under the following conditions based on the vertical and horizontal integration thresholds found in step 3.

(2) THY; Vertical integration threshold) (3) XS2≦TlX
(However, THX is the 1st detective on the horizontal side) (2) YS2-
Ye 1≦THY (however, THX: horizontal integration threshold) (3) The X and Y coordinates are saved as a new rectangular area as the starting point coordinates and ending point coordinates, respectively.

At this time, it is assumed that the conditions are satisfied if the values on the left side of the above-mentioned integration conditions are negative.

Using the method described above, repeat the process of merging and saving the circumscribed rectangles (including the newly saved rectangle area) one after another until there is nothing more to be integrated under the above conditions. .

In addition, as a storage method, among the memory locations where the information of two circumscribed rectangles is saved, the information of one circumscribed rectangle is saved, and the contents of the memory location where the information of the other circumscribed rectangle is saved are deleted. There is a way to repeat the process.

FIG. 20 is a diagram showing the results of region segmentation performed on an image read from a manuscript of a certain journal at a resolution of 8 lines/mm. 50 is a field separator, 51 is a photograph, 52 is a text portion, and 53 is a diagram.

This embodiment has the following advantages.

(b) When the boundary line Y = a do. The relationship between the area S and the number of inversions is as shown in Fig. 11, and by setting the boundary line Y = a
71, discrimination accuracy can be improved.

(b) Conventionally, to determine the field separator,
Discrimination has been made based on whether the ratio between the length of the X axis and the length of the Y axis is larger or smaller than a certain relationship.

This is similar to the case where β shown in Y=αX+β is shown as 0. However, the relationship between the length of the X axis and the length of the Y axis of the circumscribed rectangle is as shown in Figure 9, and the boundary line (threshold) is Y = αχ + β, where α and β have fixed values. By doing so, the discrimination system can be further improved.

(c) With respect to the binary image obtained by optimal binarization, a circumscribing rectangle detection means is used to determine that there are no black pixels in the horizontal direction (X direction) above an appropriate threshold value TH (appropriately 1 to 4). The extraction process is repeated until the information surrounded by a white line and a white column in which there are no black pixels in the horizontal direction (Y direction) equal to or higher than the threshold TH can no longer be extracted, and the attribute discrimination means detects the information. The circumscribed rectangles are classified into field separators, photographs, diagrams, and text parts based on the circumscribed rectangle features and the features within the rectangle, and the enclosing rectangles that are determined to be text parts by the above-mentioned attribute discrimination means in the character integration means are sequentially classified into text parts. Since it is integrated into blocks, ■ it can be applied to optimally binarized images. It has the following advantages: 1. The accuracy of attribute discrimination can be improved; 2. Character blocks can be extracted with high precision. 2. It can flexibly respond to various target images.

Note that the present invention is not limited to the illustrated embodiment, and various modifications are possible. Examples of such modifications include the following.

(a) In this embodiment, only a horizontally long field separator is described, but a vertically long field separator may also be considered.

(b) In this embodiment, 5103 shown in FIG. 10 is provided, but it may be omitted. This is because the information on the circumscribed rectangle is stored in the circumscribed rectangle position memory 32, and if the circumscribed rectangles that are identified as field separators, photographs, and charts are extracted from it, the remaining information is determined to be the circumscribed rectangle of the character part. Because you can.

(c> In this embodiment, 5112, 511 in FIG.
In 3, when inverting from a white pixel to a black pixel, the increment to Callan 1 is 1-, but when inverting from a black pixel to a white pixel, or from a white pixel to a black pixel, and from a black pixel to a white pixel, it may be incremented by 7. .

(d) Instead of the circumscribed rectangle detection process in FIG. 3, if the distance between the information areas 41 is within the threshold value T'H, it is assumed that they are the same area, and a rectangular area is extracted by the connected area (labeling process Fl). Even if processing is performed using a method of extracting a rectangular area by boundary tracking or a method of extracting a rectangular area by boundary tracking, the same circumscribed rectangle extraction results are obtained, and the same effects as in the embodiment described above can be obtained.

(Effects of the Invention) As described in detail above, according to the present invention, the binary values f and it that can be converted into an appropriate value are collected, and the outer R rectangle detection means is used to detect a value that is equal to or higher than the appropriate threshold value TH. A white line with no black pixels in the horizontal direction (X direction) (1 to 4 is appropriate) and a white column with no black pixels in the horizontal direction (X direction) with a threshold of "I'HL" The extraction process is repeated until it is no longer possible to extract the information surrounded by The circumscribed rectangles that are classified and determined to be character parts by the above-mentioned attribute determination means in the character integration means are sequentially integrated into the character part blocks, so that they can be applied to images that have been optimally binarized. ■It is possible to improve attribute discrimination accuracy.
The following effects can be expected: ■Character blocks can be extracted with high precision; ■It can be flexibly adapted to various target images.

[Brief explanation of the drawing]

FIG. 1 is a functional block diagram of an image processing device showing an embodiment of the present invention, FIG. 2 is a functional block diagram of a conventional image processing device, FIG. 3 is a flowchart of the circumscribed rectangle detection means in FIG. The figure is a diagram illustrating the circumscribed rectangle detection means in FIG. 1,
5 is a diagram for explaining the Y-direction extraction process in FIG. 3, FIG. 6 is a diagram for explaining the X-direction extraction process in FIG. 3, and FIG.
Figure 8 is a flowchart of the attribute determination means in Figure 1, Figure 8 is a flowchart of field separation extraction in Figure 7, Figure 9 is a diagram of the relationship between the X-axis length and the Y-axis length of the circumscribed rectangle, and Figure 10 is a flowchart of the attribute determination means in Figure 1. The diagrams are flowcharts 1 to 1 for extracting photo/diagram candidates in Fig. 7, Fig. 11 is a diagram showing the relationship between the number of inversions and the area of the circumscribed rectangle, and Fig. 12 is the flowchart 1 to judgment for photos/diagrams in Fig. 7. , FIG. 13 is a diagram of the relationship between the number of black pixels and the area of the circumscribed rectangle, FIG. 14 is a diagram explaining 8112 in FIG. 12, FIG. 15 is a diagram explaining 8113 in FIG.
Figure 6 is a flowchart of the literature department integration method in Figure 1.
Figure 7 is a diagram showing an example of the creation result of a peripheral histogram.
19 is a diagram illustrating integration thresholds in the vertical and horizontal directions, FIG. 19 is a diagram illustrating integration of circumscribed rectangles of character parts, and FIG. 20 is a diagram illustrating region division results. 31... Circumscribed rectangle detection means, 33...
Attribute determination means, 34...Chart position memory, 35
. . . Photo position memory, 36 . . . Field separator position memory, 37 . . . Literature department integration means, 38 . . . Character section position memory.

Claims (1)

[Claims] 1. A document image region dividing method that cuts out an information region including a character portion of input image data and performs attribute determination processing on the information region, comprising: optimally binarizing the image data; The binary image data thus obtained is input to a circumscribed rectangle detection means, and the information area in the binary image data is converted into a circumscribed rectangle by the circumscribed rectangle detection means, and the characteristics of the circumscribed rectangle and the binary image within the circumscribed rectangle are determined. and the attribute determining means determines the attributes of the information area from the characteristics of the circumscribed rectangle and the features of the binary image within the circumscribed rectangle, and the attribute determining means determines the character among the attributes. 1. A method for dividing a document surface image into regions, the method comprising: integrating a circumscribed rectangle determined to be a text section into a text section block using a text section integration means. 2. The method for dividing both images of a document into regions according to claim 1, wherein the character part integrating means creates a peripheral histogram for a character part image obtained by extracting only the character part from the binary image data. and integrating a circumscribed rectangle of the character part into the character part block using the surrounding histogram. 3. The document image region dividing method according to claim 2, wherein the character part integrating means determines, from the creation result of the peripheral histogram, a white line region in which white lines in which the black pixels do not exist at all in the horizontal direction are continuous. and a black pixel line area in which a black line in which at least one black pixel exists is continuous, and a white column in which white columns in which no black pixel does not exist in the vertical direction are continuous;
A region and a black pixel column region in which black columns in which at least one black pixel exists are extracted, respectively, horizontal writing or vertical writing is determined based on the extraction results, and horizontal or vertical writing is performed based on the determination result. Find an integration threshold for the character part, perform integration processing on the circumscribed rectangle of the character part based on the integration threshold,
A document image region division method that extracts the result as a character region. 4. In the document image region dividing method according to claim 3, the determination of vertical writing/horizontal writing is based on the creation result of the peripheral histogram, and a white line area in which white lines in which the black pixels do not exist at all in the horizontal direction is continuous. . A method for dividing a document image into regions based on the number of consecutive white column regions in which there are no black pixels in the vertical direction. 5. In the document image region dividing method according to claim 3, the integrated threshold in the vertical and horizontal directions is calculated from a histogram of the white line region width and the black line region width when horizontal writing is determined; If it is determined that the document image is written vertically, the area division method of the document image is determined based on the histogram of the white column area width and the black column area width. 6. The document image area dividing method according to claim 3, wherein the integrated threshold in the vertical and horizontal directions is determined based on a histogram of the white line or white column area widths, and the white line, white column, or area having the largest number of white lines or white columns. After detecting the white line, white column, or region width whose number becomes 0 by scanning in the direction of increasing width, the white line or white column region immediately before that is extracted as a value TW, and then the black line is detected. Or, extract the value TB from the histogram of the black column area width, and extract the value TW and the value TB
Based on this, in the case of horizontal writing, set the horizontal total threshold value THX = TW + TB, the vertical total threshold value THY = TW + Er (however, Er: the number of lines corresponding to the tilt error), and in the case of vertical writing, set the total horizontal value as THX = TW + TB. A document image area division method that sets a threshold value THX=TW+Er and a total vertical threshold value THY=TW+TB. 7. In the document image region dividing method according to claim 3, in the integration process, two region coordinates (XS1, YS1) of the circumscribed rectangle of interest are determined.
) (Xe1, Ye1), (XS2, YS2) (Xe2,
Ye2) (However, XS1, XS2, YS1, YS2; starting point coordinates of horizontal and vertical areas, Xe1, Xe2, Ye
1, Ye2; end point coordinates of the horizontal and vertical regions), based on the following integration conditional expressions (1) to (4), the circumscribed rectangles that have been determined to be the character portion are successively combined into one, and the limit A method for dividing a document image into regions, in which the integration is repeated up to the point where the integration results are extracted as the text block. Integration conditional expression (1) XS2-Xe1≦THX (where THX: horizontal integrated threshold) (2) YS2-Ye1≦THY (however, THY: vertical integrated threshold) (3) XS1-Xe2≦THX ( 4) YS1-Ye2≦THY (However, XS2-Xe1, YS2-Ye1, XS1-X
When the value of e2, YS1-Ye1 is negative, the condition is met)