JPH0696275A - Image processor - Google Patents

Abstract

PURPOSE:To shorten the processing time at the time of dividing an area, to form a storage area in a small scale, and to improve the accuracy for dividing a complicated area constitution. CONSTITUTION:This device is provided with an image thinning-out part 106 for taking OR of vertical (m) dots and horizontal (n) dots against an input image from an input part 101 and thinning out newly mXn picture elements to one picture element, a labeling part 107 for adding a label by one line each to a black picture element of its thinned-out image, putting the same label to picture elements continued vertically, to the left and the right, and obliquely, and simultaneously, forming a rectangle, and a separator, etc., detecting part 108 for discriminating a rectangle corresponding to a text, a rectangle corresponding to a graphic or a photograph, a rectangle corresponding to a table, a rectangle corresponding to a separator, etc., by using the number of picture elements to width height, and an area of its rectangle, that is, picture element density, etc.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an electronic device such as an OCR (optical character recognition) device, a copying machine or a facsimile, and particularly divides an input image into a character area and an area such as a figure or a table. The present invention relates to an image processing device.

[0002]

2. Description of the Related Art In a conventional image processing apparatus, as a region dividing means, a device using a run length analysis method for examining a distribution of run lengths and dividing a character region, a graphic region or the like according to the length of a white run and a black run, A method that uses a spectrum analysis method that analyzes a Fourier spectrum of an input image and divides it into various regions, and alternately projects vertical and horizontal projections (histograms) as described in JP-A-64-15889. There is a method using a projective analysis method in which an area is repeatedly divided into pieces based on information of a peripheral portion.

[0003]

However, in the above-mentioned prior art, a great deal of time is spent on the arithmetic processing,
In addition, since the pixels of the image are processed, the storage area becomes large in scale, and there is a problem that the division accuracy for an image having a complicated area configuration is low.

In view of the above points, an object of the present invention is to shorten the processing time when dividing an area, reduce the size of a storage area,
Another object of the present invention is to provide an image processing device that improves the division accuracy of a complicated area configuration.

[0005]

To achieve the above object, the present invention provides an image input means for inputting an image and a logical sum of vertical m dots and horizontal n dots for an input image from the image input means. A new image thinning means for thinning out mXn pixels to one pixel and a label is added to the black pixels of the image thinned out by the image thinning means, one line at a time, and the pixels are continuous vertically, horizontally and diagonally. Are labeled with the same label, and at the same time, a labeling unit that traces a rectangle is used, and the width, height, area, and the number of pixels with respect to the area of the rectangle labeled by the labeling unit, that is, the pixel density, is used to correspond to the text portion. A rectangle corresponding to a rectangle, a figure or a photograph, a rectangle corresponding to a table, a rectangle corresponding to a separator, and the like, and a separator or the like detection means, and based on the detection result of the separator or the like detection means. And performing segmentation of the input image Te.

Further, the present invention is preferably a set direction detecting means for estimating whether the text is a vertical writing text or a horizontal writing text by comparing the width and height of a rectangle corresponding to the text portion divided by the detecting means such as the separator. , A headline detection means for detecting headlines using the set direction and character size determined by the set direction detection means, and a rectangle merger for performing rectangle merger using a virtual rectangle obtained by expanding the size of the original rectangle It can be characterized by further comprising means.

[0007]

In the present invention, the input image is converted into rectangular data by the area dividing means having the image thinning section, the labeling section, etc., and divided into areas such as characters, figures and tables.

[0008]

Embodiments of the present invention will now be described in detail with reference to the drawings.

FIG. 1 shows a schematic configuration of an image processing apparatus according to an embodiment of the present invention. In the figure, 101 is an input unit for image data. Reference numeral 102 denotes an arithmetic processing unit (CPU) that controls the apparatus and performs arithmetic processing of each unit. Reference numeral 103 denotes a storage unit that stores a control program of a flowchart described below and various data. An output unit 104 outputs the calculation processing result, the image processing result, and the image data. 10
An area dividing unit 5 performs the entire area dividing process.

Next, the internal structure of the area dividing section 105 will be described. First, an image thinning unit 106 thins an input image, and newly thins out mXn pixels to one pixel by taking a logical sum of vertical m dots and horizontal n dots with respect to the input image.
Reference numeral 107 denotes a labeling unit that labels the thinned pixels and at the same time creates initial rectangular data. Labels are added to the black pixels of the image one line at a time, and the same is applied to pixels that are continuous vertically, horizontally, and diagonally. Label and shape the rectangle at the same time. Reference numeral 108 denotes a separator or the like detection unit that detects a separator, a table, a figure, etc., and uses the width, height, area, and the number of pixels per area, that is, the pixel density, to identify a rectangle or figure corresponding to a character portion (text). Alternatively, a rectangle corresponding to a photograph, a rectangle corresponding to a table, a rectangle corresponding to a separator, etc. are distinguished. Reference numeral 109 denotes a set direction detecting unit that detects a set direction of a sentence such as vertical writing or horizontal writing, and estimates a vertical writing sentence or a horizontal writing sentence by comparing the width and height of a rectangle corresponding to the character portion (text). . Reference numeral 110 denotes a headline detection unit for detecting a headline, which detects the headline using the set direction and the character size. Reference numeral 111 denotes a rectangle merging unit, which merges rectangles by using a virtual rectangle obtained by expanding the size of the original rectangle.

The flowchart of FIG. 2 shows an image processing procedure in the area dividing unit 105 of FIG. 1, and a control program for executing this processing is stored in the storage unit 103.

Step S 201 First, an original image is input from the image input unit 101.

Step S202 Next, in the image thinning unit 106, the logical sum of vertical m dots and horizontal n dots is added to the original image to newly generate m.
Xn pixels are thinned out to one pixel. Here, if even one dot has a black pixel in the mXn image of the original image, the thinned image is black.

Step S203 Next, in the labeling unit 107, labels are added to the black pixels of the thinned image line by line, and the vertical, horizontal, and horizontal lines are added.
The diagonally continuous pixels are labeled with the same label, and at the same time, the rectangle is modeled.

Taking FIG. 3 as an example, the first detected pixel A is labeled 1, and the coordinates (Xa, Xa,
Ya) is the start and end points of the rectangle, the number of pixels is 1, and the same label 1 as the pixel is added to the rectangular label for distinguishing the rectangle, and the above data is stored in the storage unit 103 as rectangular data as shown in FIG. Remember.

Next, a label 2 is attached to a pixel B which has no continuous pixel in the left direction (there is no continuous pixel from the top because it is the first row), and the coordinates (Xb, Yb) of this pixel B are given.
Is the starting point and the ending point of the rectangle, the number of pixels is 1, and the same label 2 as the pixel is added to the rectangular label for distinguishing the rectangle, and the above data is also rectangular data as shown in FIG.
Store in 03.

When the labeling of the first line is completed as described above, the process moves to the second line.

The first pixel C of the second row is the pixel A of label 1.
Since it is continuous from the top, pixel label 1 is added, and the number of pixels is added to the rectangular data of rectangular label 1 by 1 to make a total of 2 pixels, the rectangular label remains 1 and the rectangular coordinate is only the end point. Is updated from (Xa, Ya) to (Xa, Yc) (the coordinates of the starting point do not change).

Since the next pixel D is continuous with the pixel C from the left, the label 1 is added, and the number of pixels is added to the rectangular data of the rectangular label 1 by a total of 3 pixels, and the rectangular label remains unchanged 1. As it is, the rectangular coordinates are only for the end point (Xa, Y
Update from (c) to (Xd, Yc) (the Y coordinate of the end point does not change). At this time, the pixel D is diagonally continuous with the pixel B and is continuous from the pixel C. Therefore, the label of the pixel B is changed from label 2 to label 1, and the rectangular label 1
The number of pixels of the rectangular label 2 is added to the rectangular data of 4 to make a total of 4 pixels, the rectangular label remains 1, and the rectangular coordinates include only the end points so as to include all the pixels A, B, C and D. Update from (Xd, Yc) to (Xb, Yd). The rectangular data of the rectangular label 2 is invalidated by setting the rectangular label to 0.

When the second line is completed as described above, the process moves to the third line.

Since the first pixel E in the third row is diagonally continuous with the pixel C, the pixel label 1 is added, and the number of pixels is increased by 1 to the rectangular data of the rectangular label 1, resulting in a total of 5 pixels.
The rectangular label remains unchanged at 1, and the rectangular coordinates are updated from the starting point (Xa, Ya) to (Xe, Ya) and the ending point (Xb, Yd) to (Xb, Ye). That is, in the rectangular data of FIG. 4 on the storage unit 103 at this time, the rectangular label is 1, the starting point coordinates are (Xe, Ya), the ending point coordinates are (Xb, Ye), the pixel label is 1, and the number of pixels is 5. Become.

In the same manner, labeling and rectangular tracing are performed on all thinned pixels.

Step S204 After the above-described labeling and tracing of rectangles, in the separator or the like detection unit 108, next, a rectangle corresponding to a character portion (text), a rectangle corresponding to a figure or a photograph, a table, etc., and a rectangle corresponding to a separator. Etc., the width W, height H of the above rectangle,
The area S and the number of pixels with respect to the area, that is, the pixel density D (which can be easily obtained by calculation using these rectangular data) are used for the discrimination. The details of this processing will be described below with reference to the flowchart of FIG.

In the separator, the width W is less than the threshold Tw1 and the height H is more than twice the threshold Tw of the width W (step S50).
3) or the width W is larger than the threshold Tw1 and the height H
Is 3 times the threshold Tw of the width W or more (step S504), the rectangular label on the storage unit 103 is unified as -3 as a vertically long separator, and the pixel label forming the rectangle is the current label number as it is, apart from the rectangular label. Is held (step S514).

Also, the case where the width and the height are interchanged is determined in the same manner as described above (steps S505 and S506), and if applicable, the rectangular label is changed to -3 as the horizontally long separator.

Next, as shown in FIG. 6, step S50.
When the pixel density D is equal to or less than the threshold Td1 (step S507), the rectangular label is changed to -3 and the rectangular label is changed to -3 (step S514).

When step S507 is negative but the area S is larger than the threshold value Ts1 (step S5).
08), the pixel density D is less than the threshold value Td2 (step S5).
If the pixel density D is equal to or larger than the threshold value Td2, the rectangle label is changed to -5 (step S516).

When the area S is less than or equal to the threshold Ts1 and greater than or equal to the threshold Ts2 (step S510), the pixel density D is greater than or equal to the threshold Td3 (step S511), and the width W and height H are both greater than or equal to the threshold Tw4. Further, when the pixel density D is equal to or larger than the threshold value Td5 (step S513), the rectangle is regarded as a figure or a photograph and the rectangle label is changed to -5 (step S516).

Further, when the area S is less than or equal to the threshold Ts1 and greater than or equal to the threshold Ts2 (step S510), the pixel density D
Rectangles whose threshold is less than the threshold Td4 (step S512) are regarded as a table, and the rectangle label is changed to -4 (step S51).
5).

As described above, a rectangle corresponding to a figure, a photograph, a table, etc., a rectangle corresponding to a separator, etc. are detected, and the remaining rectangle is used as the text (character portion) and the rectangle label is made the same as the pixel label. Put (Step S51
7). FIG. 6 shows the relationship between the above-mentioned threshold value in the area S and the pixel density D and the division of the separator and the like.

Step S205 Next, the set direction detecting unit 109 detects the set direction of sentences such as vertical writing and horizontal writing. This processing will be described with reference to the flowchart of FIG.

In the case of a horizontally written sentence, the rectangle remaining as the body tends to become a horizontally long rectangle with pixels thinned out in the horizontal direction, and in the case of a vertically written sentence, the rectangle remaining as the body tends to become vertically long. Therefore, in the set direction detecting unit 109, the average value of the width W and the height H of the rectangle of the text is calculated (step S701), and when the average width W is larger than the average height h, it is regarded as a chapter with many horizontal writing, The average height h is set to the character size of one character (steps S702 and S703). On the contrary, when the average height h is larger than the average width W, it is regarded as a sentence with many vertical writings, and the average width w is set to the character size of one character (steps S702 and S710).

Next, as shown in FIG. 8, a rectangular histogram is created in the direction opposite to the set direction (steps S704, S).
711), from the shape of the marginal distribution, places where the threshold value is equal to or less than the threshold value Tk are set as paragraph breaks (steps S705 and S712), and a rectangular histogram is taken for each paragraph in the same direction as the set direction (steps S708 and S715). From the shape, the continuous length of black pixels is set as the character size of the character in the paragraph, and the continuous length of white pixels is detected as the line pitch (steps S709 and S7116). This detection process is repeated until the paragraph ends (steps S706 and S713).

Step S206 Next, the caption detection unit 110 detects captions from the set direction and the character size. As shown in the flowchart of FIG. 9, taking the case of vertical writing as an example, a rectangle in which the width W of the rectangle is larger than the threshold value Tm1 times the character size w by 1 times or more is detected from the body rectangles as a candidate for a heading rectangle (step S).
905). Further, when the height H of the rectangle is smaller than the threshold value Tm2 times the character size w (step S906), it is considered that the horizontally-written characters are continuous and is set as a heading (step S908).

Further, since some of the rectangles used as headings based on the size of the rectangles as described above include characters in which the text of the text is combined, the distance between the text rectangle and the heading rectangle is the threshold value T1. If it is closer (step S907), the heading is corrected to the text (step S90).
9).

Further, among the body rectangles remaining by the above processing, the body and heading type rectangles having no rectangle in the character size w are set as isolated rectangles (steps S910 to S913).

Step S207 Next, in the rectangle merging unit 111, the rectangles corresponding to the main text that have been irrelevant and irrelevant to each other and the rectangles of the headings are merged.

Considering an example of FIG. 10, considering the combination of the rectangle A1 and the rectangle B, a virtual rectangle A is obtained by expanding the rectangle A1 by Px in the X direction and Py in the Y direction.
Considering 1 ', it is searched whether or not there is a rectangle surrounding or in contact with A1'. If they are in contact with each other like rectangle B, rectangles A1 and B are merged to newly merge rectangle A2. Is created and stored by updating the rectangular data of the rectangle A1, and the data of the rectangle B is invalidated. However, at this time, Px and Py are values obtained from the character size and the line pitch, and when a rectangle formed by merging rectangles comes into contact with a figure, table, separator, etc., the merging itself is invalidated. The rectangle is left as it is.

In this way, the headlines are first merged, and then the body rectangles are merged. At this time, when merging texts with each other, avoid contact not only with figures, tables, and separators but also with headings. If they do contact headings, the merger itself is invalidated and the original rectangle remains unchanged. I will keep it.

Step S208 Finally, the rectangular data of various areas obtained as described above is output from the output unit 104 to the outside together with the image data.

<Other Embodiments> Step S2 in FIG. 2 described above.
When inputting the image of 01, it is judged whether it is multi-valued,
By converting the image into a binary image, it is possible to perform area division processing even if the input image is a multivalued image such as a color image.

Further, in the image thinning process of step S202, if the number of pixels of the input image is sufficiently small so as not to impair the processing speed, the image thinning process can be omitted.

Further, in the detection of the separator or the like in step S204, by further distinguishing the rectangular labels by the vertical / horizontal difference of the separator, the difference of the threshold value when obtaining figures, tables, etc., it is possible to further classify the area attributes. It is also possible to do so.

Similarly, in the step of detecting the separator or the like in step S204, if only an image having a specific attribute such as only a sentence is input from the beginning, the detecting process of the separator or the like can be omitted.

Further, in the detection of the set direction in step S205, instead of obtaining the average width and the average height, the maximum value of each can be obtained for substitution.

Similarly, in the detection of the set direction in step S205, if the set direction is confirmed again for each paragraph, it is possible to accurately perform processing on a document in which vertical writing sentences and horizontal writing sentences are mixed.

Similarly, in the case of detecting the set direction in step S205, if only the text of a fixed set direction is input from the beginning, this detection process can be replaced by setting the set direction to a fixed value.

Further, in the heading detection in step S206, if only an image without a heading is input from the beginning, this detection processing can be omitted.

In the merge of rectangles in step S207, as shown in FIG. 11, when a rectangle such as a separator is contacted, it is confirmed whether or not it is also in contact with an actually thinned pixel. If it is not, you can deal with skewed images by performing a merge.

Similarly, in the rectangle merging in step S207, if the image has a small number of pixels such that the image thinning process can be omitted, the merging process can be omitted.

Further, at the final output of step S208, it is possible to refer to the rectangular data of various areas and output only the image data of the necessary area, which further reduces the storage area, The processing time can be further shortened.

[0052]

As described above, according to the present invention,
After thinning the input image, it is converted to rectangular data using labeling and divided into character areas, figures or photographs, tables, separators, headings, etc., so there is no change in existing electronic parts, circuits, etc. In addition, the processing time can be shortened, the storage area can be downsized, and the division accuracy of the complicated area configuration can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of an image processing apparatus according to an embodiment of the present invention.

FIG. 2 is a flowchart showing an entire image processing procedure of area division according to an embodiment of the present invention.

FIG. 3 is an explanatory diagram showing an example of labeling processing according to an embodiment of the present invention.

FIG. 4 is an explanatory diagram showing a structure of rectangular data on a storage unit according to an embodiment of the present invention.

FIG. 5 is a flowchart showing a processing procedure of a detector such as a separator according to an embodiment of the present invention.

FIG. 6 is an explanatory diagram showing division by attribute density D and area S of a rectangle (region) according to an embodiment of the present invention.

FIG. 7 is a flowchart showing a processing procedure of a set direction detecting unit according to an embodiment of the present invention.

FIG. 8 is an explanatory diagram showing an example of paragraph detection processing according to an embodiment of the present invention.

FIG. 9 is a flowchart showing a processing procedure of a heading type detection unit according to an embodiment of the present invention.

FIG. 10 is an explanatory diagram showing an example of rectangle merging processing according to an embodiment of the present invention.

FIG. 11 is an explanatory diagram showing an example of separator rectangle merging processing according to another embodiment of the present invention.

[Explanation of symbols]

 101 input unit 102 arithmetic processing unit (CPU) 103 storage unit 104 output unit 105 area dividing unit 106 image thinning unit 107 labeling unit 108 separator detection unit 109 set direction detection unit 110 heading detection unit 111 rectangular merged unit

Claims (2)

[Claims] 1. An image input means for inputting an image, and m dots in the vertical direction and n pixels in the horizontal direction with respect to the input image from the image input means
Image thinning-out means for newly thinning out mXn pixels to one pixel by taking the logical sum of dots, and labels for each line are added to the black pixels of the images thinned out by the image thinning-out means, and vertical, horizontal, diagonally The same label is applied to consecutive pixels, and at the same time, labeling means for tracing a rectangle is used, and the width, height, area, and the number of pixels per area, that is, the pixel density, of the rectangle labeled by the labeling means are used. A rectangle corresponding to a text part, a rectangle corresponding to a figure or a photograph, a rectangle corresponding to a table, a rectangle corresponding to a separator, and the like, and a separator or the like detection means, and based on the detection result of the separator or the like detection means. The image processing apparatus is characterized in that the input image is divided into areas. 2. A set direction detecting means for estimating a vertically written sentence or a horizontally written sentence by comparing the width and height of a rectangle corresponding to the sentence portion divided by the detecting means such as the separator, and the set direction detecting means. It further has headline detection means for detecting headlines using the determined set direction and character size, and rectangle merging means for merging rectangles using a virtual rectangle that is an extension of the original rectangle size. The image processing apparatus according to claim 1, which is characterized in that.