JP2003271973A - Method and program for image processing - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To accurately discriminate an image area even in a complicated lay-out. <P>SOLUTION: This method is provided with a binalization step S201 for binalizing an input image, the first labeling step S202 for imparting the same label to picture elements continued vertically, laterally and diagonally to a black picture element of a binalized picture element, and for patterning a circumscribed rectangle, the second labeling step S203 for imparting the same label to picture elements continued vertically and laterally to a white picture element of a binalized picture element, and for patterning circumscribed rectangle, an inclusion relation detecting step S204 for detecting an inclusive relation of the circumscribed rectangle labeled in the first labeling step with the circumscribed rectangle labeled in the second labeling step, and an attribute detecting step S205 for discriminating attributes for all the picture elements, using the inclusive relation, the circumscribed rectangle in the first labeling step and the circumscribed rectangle in the second labeling step. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing method and an image processing program for performing area identification on an input image.

[0002]

2. Description of the Related Art In recent years, the processing capability of computers has been remarkably improved, and along with this, rich document expression tools have been developed, and documents are being colorized and diversified. In addition, the digitization of document management is progressing, and the digitization of paper documents, which are conventionally stored as paper documents by a scanner, is progressing.

On the other hand, in the prior art, a run length analysis method is used in which the distribution of run lengths for a binary image is examined and a character area, a graphic area, or the like is divided according to the length of a white run or a black run, or an input image. Using a spectrum analysis method for analyzing the Fourier spectrum of the above and identifying it in various regions, or by alternately repeating vertical and horizontal projections as described in JP-A-64-15889.
There is a method using a projective analysis method that divides an area.

On the other hand, due to the improvement in computer processing capability, it has become necessary to improve the functionality of document processing tools as well, and not only extraction of character areas but also figures, photographs, backgrounds, etc. are accurately extracted and scanned. There is a growing demand for a function that allows the document to be reused, such as conversion into a structured document.

[0005]

However, since the run-length analysis method, the projection analysis method, etc., are area identification for each fixed unit area (rectangular unit, etc.) and are not identification for each object, the outline of the pattern is not rectangular. In this case, it is difficult to separate and identify only the picture area. In addition, since there is no difference in the Fourier spectrum in a document in which the background is halftone dots and characters or halftone photographs are formed on it, region identification is performed using the spectrum analysis method for such documents. Is difficult. As described above, it is hard to say that the conventional technique is sufficiently satisfactory as the document area identification technique in view of the current situation that the document is colored and the layout is complicated.

[0006]

The present invention has been made to solve the above problems. That is, according to the present invention, a binarization step of separating an input image into an edge corresponding to a black pixel and a non-edge corresponding to a white pixel, and a black pixel of a binary image binarized in the binarization step are performed. On the other hand, the same label is added to pixels that are consecutive in the vertical, horizontal, and diagonal directions, and the first labeling step that traces the circumscribed rectangle of the pixel groups with the same label and the binarization step is used for binarization. The second labeling step of adding the same label to the pixels that are continuous in the vertical and horizontal directions with respect to the white pixels of the created binary image, and modeling the circumscribed rectangle of the pixel group with the same label; A inclusion relation detecting step of detecting an inclusion relation between the circumscribed rectangle labeled in the first labeling step and the circumscribed rectangle labeled in the second labeling step, a result of the inclusion relation detection step, and a first labeling The first information (width, height, area, pixel density, variance) about the bounding rectangle labeled in the labeling step and the second information (width, height) about the bounding rectangle labeled in the second labeling step. Is an image processing method including an attribute detection step for distinguishing attributes for all pixels, and an area division step for performing area division of the input image based on the detection result of the attribute detection step. It is also an image processing program including these steps.

In the present invention as described above, since the labeling conditions are changed between the first labeling step and the second labeling step, the circumscribed rectangle of the group of black pixels and the circumscribed rectangle of the group of white pixels are changed. The inclusion relation can be accurately grasped, and the area of the input image can be accurately divided.

[0008]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of an apparatus that realizes the image processing method of this embodiment. That is, the present apparatus includes the image input unit 101, the CPU 102, the storage unit 103, and the area identification unit 104.

The image input section 101 is a section for inputting a document image such as a scanner. The CPU 102 is a central processing unit, and performs processing according to the present embodiment according to an image processing program stored in the storage unit 103. The storage unit 103 is composed of a RAM and a ROM, and the CPU 1
An image processing program executed by 02, various parameters used for the processing, input images, and various data are stored. The area identification unit 104 performs area identification for the input image according to the image processing program of the present embodiment stored in the storage unit 103.

Next, the operation of the area identification section 104 will be described. FIG. 2 is a flowchart showing the flow of the area identification processing executed by the area identification unit.

<Step S201> Here, a process of separating the image data input from the image input unit 101 into edges and non-edges is performed. In the present embodiment, if the input image is a color image and there is a pixel between the pixel of interest and its eight neighboring pixels that satisfy the following equations (1) and (2), the pixel is regarded as an edge. Binary output is O
N (black pixel), if not satisfied, it is regarded as a non-edge,
The binary output is turned off (white pixels). The binarized image is stored in the storage unit 103.

((R-R ') ² + (G-G') ² + (B-B ') ² ) ^1/2 > th1 ... (1) Y <Y' ... (2)

Here, the image signals of the pixel of interest are R, G, B,
Let R ′, G ′, and B ′ be neighboring image signals, and let Y and Y ′ be the luminance signals obtained by the equation (3). Y = 0.3R + 0.59G + 0.11B… (3)

<Step S202> Here, with respect to the black pixels of the binary image binarized in step S201, the same label is added to consecutive pixels in the vertical, horizontal, and diagonal directions, and the same label is attached. The circumscribed rectangle of the pixel group that has been created is modeled. In addition, rectangular information is obtained along with this processing. The label information and the rectangular information obtained in this processing step are stored in the storage device 103 and used in the subsequent processing. The rectangle information has the following contents. (A) Width of circumscribed rectangle (B) Height of circumscribed rectangle (C) Area of circumscribed rectangle (D) Left edge coordinate value of circumscribed rectangle (E) Right edge coordinate value of circumscribed rectangle (F) Top edge coordinate value of circumscribed rectangle ( G) Lower end coordinate value of circumscribed rectangle (H) Effective pixel number in circumscribed rectangle (I) Luminance variance: Calculated by referring to an input image corresponding to ON pixel portion (J) Number of cell candidates: ON pixel group (area The number of OFF pixel groups (areas) that can be directly included in the table / frame (K) The number of non-cell candidates: The OFF pixel group (area) that cannot be directly included in the table / frame cells (area) Number of regions) (L) inclusion label: label of OFF pixel group that directly includes ON pixel group (M) attribute At this stage, information of (A) to (H) is acquired, and (I) to
(M) is obtained by a subsequent process such as an inclusion relation detection process. Finally, attribute determination is performed using these.

The labeling operation will now be described with reference to FIG. 3 as an example. First, it is checked whether there are unlabeled black pixels from left to right in the main scanning direction. The process goes to the right end in the main scanning direction, and if it does not exist, the same process is performed from the left end of the next line. As shown in FIG. 3A, the label 1 is added to the first detected black pixel a.

Next, when the unlabeled black pixels are checked in the 8 neighborhoods of the pixel a starting from the black pixel a, the black pixel b corresponds to it. Therefore, this is judged as a continuous black pixel and is the same as the pixel a. Label 1 Next, it is checked whether there are any unlabeled black pixels in the 8 neighboring pixels of the pixels a and b. By repeating this operation, 8 pixels of all pixels to which label 1 is added
Repeat until there are no unlabeled black pixels in the neighborhood.

Next, the process of checking whether there is any unlabeled black pixel from the pixel to the right of the black pixel a is restarted. In FIG. 3A, the black pixel g is detected, and the label 1 is added to this black pixel g. After that, similarly to the case of the black pixel a, the process of adding the label 2 is repeated until there are no unlabeled black pixels in the 8 vicinity of all the pixels to which the label 2 is added. In the case of FIG. 3A, the label 1 is added to the black pixels a to f and the label 2 is added to the black pixels g to 1 as shown in FIG.

Next, extraction and calculation of rectangle information will be described. Where the leftmost coordinate value is Left and the rightmost coordinate value is Righ
Let t be the top coordinate value, the top coordinate value is Bottom, the bottom coordinate value is Bottom, and the number of pixels (the total number of pixels in a pixel group having the same label) is count. First, the above variables are initialized when the first pixel to which the label X is given is specified. When the coordinate value of the first pixel is (X1, Y1) and the brightness corresponding to the first pixel is Y1, the result is as follows. Left = X1 Right = X1 Top = Y1 Bottom = Y1 count = 1

For the second and subsequent pixels to which the label X is given, the following operation is performed. the coordinate value of the nth pixel
(Xn, Yn), and the luminance corresponding to the nth pixel is Yn.

[0020] If Left ≧ Xn, update Left = Xn. If Right ≤ Xn, update to Right = Xn. If Top ≧ Yn, update Top = Yn. If Bottom ≤ Yn, update Bottom = Yn. count = count + 1

This operation is repeated until the label X is applied. The rectangle information to be obtained is obtained from the above variables as follows. Width (hereinafter referred to as W) = Right-Left + 1 Height (hereinafter referred to as H) = Bottom-Top + 1 Area (hereinafter referred to as S) = Width x Height Number of effective pixels in circumscribed rectangle = count Left coordinate value , The right end coordinate value, the upper end coordinate value, and the lower end coordinate value are the finally obtained Left, Right, Top, and Bottom, respectively. These operations are performed for each label.

<Step S203> Here, with respect to the white pixels of the binary image binarized in step S201, the same label is added to consecutive pixels in the vertical and horizontal directions, and the rectangle is simultaneously modeled. At the same time as this processing, rectangular information is obtained. The label information and the rectangular information obtained in this processing step are stored in the storage device 103 and used in the subsequent processing. The rectangle information has the following contents. (A ') Width of circumscribed rectangle (B') Height of circumscribed rectangle (C ') Left edge coordinate value of circumscribed rectangle (D') Right edge coordinate value of circumscribed rectangle (E ') Top edge coordinate value of circumscribed rectangle (F' ) Bottom coordinate value of the circumscribed rectangle (G ') Area of the circumscribed rectangle (H') Number of character inclusions: Number of character candidates directly included in OFF pixel group (area) (I ') Inclusion label: Direct inclusion of OFF pixel group ON pixel group (J ') attribute

At this stage, the information of (A ') to (G') is acquired, and (H ') to (J') are obtained by the subsequent processing such as the inclusion relation detection processing. Finally, attribute determination is performed using these.

The labeling operation will be described with reference to FIG. 4 as an example. First, it is checked whether there are unlabeled white pixels from left to right in the main scanning direction. The process goes to the right end in the main scanning direction, and if it does not exist, the same process is performed from the left end of the next line. In the example shown in FIG. 4A, the label 1 is added to the first detected white pixel a.

Next, when the unlabeled white pixels are checked in the four upper, lower, left, and right neighborhoods of the pixel a starting from the white pixel a, the pixel b corresponds to it. Attach the same label 1. Next, it is checked whether there are any unlabeled white pixels in the four neighboring pixels above, below, left and right of the pixels a and b. This operation is repeated until there are no unlabeled white pixels near the top, bottom, left, and right 4 of all the pixels to which the label 1 is added.

Here, the difference from step S202 is that the white pixel d is not judged to be a white pixel continuous to the pixel c with respect to the white pixel c. Similarly, pixel f for pixel e,
Pixel h for pixel g, pixel h for pixel i, pixel j
The pixel k with respect to the pixel and the pixel m with respect to the pixel l are not determined as continuous white pixels.

Next, the process of checking whether there is any unlabeled white pixel from the pixel to the right of the white pixel a is restarted. In FIG. 4A, the white pixel d is detected, and the white pixel d is detected.
Label 2 is added to.

After that, the label 2 is printed as in the case of the white pixel a.
The process of adding the label 2 is repeated until there are no unlabeled white pixels in the vicinity of the upper, lower, left and right 4 of all the pixels to which the. The above process is repeated until there are no unlabeled white pixels.

In the case of FIG. 4A, as shown in FIG. 4B, the label 1 is added to the continuous pixels starting from the white pixel a, and the label 2 is added to the continuous pixels starting from the white pixel d. , Label 3 is added to consecutive pixels starting from white pixel h.

Since the extraction of the rectangular information is performed in the same manner as in step S202, its explanation is omitted here.

<Step S204> Next, step S2
02 and the inclusion relation between the labeled ON pixel group and the labeled OFF pixel group extracted in step S203 is detected. Here, a target pixel group having a direct inclusion relationship is detected. Rectangle information (inclusion label) obtained in this processing step
Is stored in the storage device 103 and used in the subsequent processing.

The operation of detecting the inclusion relation will be described with reference to FIG. In contrast to FIG. 5A, the above step S202
Then, when the process of step S203 is performed, the result is as shown in FIG. Here, in order to distinguish the label in the process of step S202 from the label in the process of step S203, the label given in the process of step S202 is expressed in the form of and. Further, the rectangular information of each label is as follows (only coordinate value data is shown in this inclusion relation detecting step, since only coordinate value data is used).

Label 1: Left_1 = 1, Right_1 = 11, T
op_1 = 1, Bottom_1 = 9 Label 2: Left_2 = 3, Right_2 = 4, Top_2 = 3, Bott
om_2 = 6 Label 3: Left_3 = 8, Right_3 = 9, Top_3 = 3, Bott
om_3 = 6 Label: Left_ = 2, Right_ = 5, Top_ = 2, B
ottom_ = 7 Label: Left_ = 7, Right_ = 10, Top_ =
2, Bottom_ = 7

Here, the inclusion relation is detected with reference to the one labeled in step S202. First, a binarized OFF pixel adjacent to the pixel to which the label is attached is searched for, and the label attached to it (step S203).
(The one given in step 1) is extracted. In the case of FIG. 5B, the labels are 1 and 2.

Then, based on the label, the rectangular information of the label 1 and the label 2, the inclusion relation of each is determined. Judgment is made by the following method.

The rectangle information of the target label (A) is Left, R
Right, Top, Bottom, rectangle information of the label (B) of the pixel group consisting of pixels adjacent to the pixel of the target label is Lef
t ', Right', Top ', Bottom'.

Left ≧ Left 'and Right ≦ Right' and T
When op ≧ Top ′ and Bottom ≦ Bottom ′, it is determined that the label (A) is included in the label (B). Also, Left ≤ L
eft 'and Right ≥ Right' and Top ≤ Top 'and Botto
When m ≧ Bottom ', it is determined that the label (A) includes the label (B). Since the relationship between the label and the label 1 is Left_ ≧ Left_1, Right_ ≦ Right_1, Top_ ≧ Top_1, Bottom_ ≦ Bottom_1, it is determined that the label is directly included in the label 1. In this case, the rectangle information (12) of the label is "1". In the description, numbers with parentheses (other than mathematical formulas) mean circled numbers, and the same applies hereinafter.

Since the relationship between the label and the label 2 is Left_ ≦ Left_1, Right_ ≧ Right_1, Top_ ≦ Top_1, Bottom_ ≧ Bottom_1, it is judged that the label directly includes the label 2. In this case, the rectangle information of label 2 is ""
Becomes

In step S204, the inclusion relation is detected for each label as described above, and the number of the label including the target label is held as one piece of rectangle information (inclusion label).

In the above example, the inclusion label of the label is "1" and the inclusion label of the label 2 is "".
The inclusion label of label 1 is “none”. As described above, in this inclusion relation detecting step, the binarized ON area and OF
It is important to accurately extract the structure of the F region.

Further, not giving the same label to the pixels which are diagonally continuous in step S203 is effective as the inclusion relation detecting step. Note that, in step S203, similar to step S202, when the same label is given to pixels that are diagonally continuous, the result becomes as shown in FIG. 5C. In this inclusion relation detecting step, the information that label 1 includes the label and the label is included. Can only be extracted,
The exact structure of the ON area and the OFF area cannot be extracted.

<Step S205> Here, attribute determination is performed based on the rectangular information extracted in steps S202, S203, and S204. In addition, in the present embodiment, a character (also a line segment or the like) is present in a background, a frame, or a cell of a table, and a character-like thing existing in the pattern is a part of the pattern. This is because this method is not intended to extract as many characters as possible for character recognition such as OCR, but is intended to identify attributes that are used to structurally decompose a scanned document. is there.

Now, the operation of the attribute discrimination will be described with reference to FIG. 6 as an input image. It is assumed that the areas A to E in FIG. 6 are formed in different colors, and the car picture is also formed in a plurality of colors.

In contrast to FIG. 6, step S20 explained above.
FIG. 7A shows the result of performing the processing from 1 to step S204.
Will be described below. As for the labels in FIG. 7, those labeled in step S202 are represented by circled numbers, and those labeled in step S203 are represented only by numbers, as described in step S204. . FIG. 7B shows the inclusion label obtained in step S204.

First, the background detection will be described. In the present embodiment, the background is set to another object (pattern, character,
It is defined as an area that is not included in any table, frame, etc.). Based on this, the background is extracted from the binarized OFF pixel group in the flowchart shown in FIG.

In the example of FIG. 7A, for the labels 1 to 30 of the binarized OFF pixel group, it is checked whether or not the inclusive label exists, and the background is extracted. From FIG. 7B, FIG.
In (a), since the binarized OFF pixel group (area) with label 1 has no inclusive label, it is determined to be the background.

Next, the attribute detection of the binarized ON pixel group will be described with reference to the flowchart of FIG. First,
If the threshold value th1 that is obtained in advance from the average font size used in the text is used and both the width (W) and height (H) of the circumscribed rectangle size exceed the threshold value th1, the characters in the text of the document cannot be considered. Judge that.

However, it cannot be determined that the character is a character only by the above processing, and the size of the circumscribing rectangle (H or W) and the circumscribing rectangle aspect ratio (H / W) are used. Among the width (W) or height (H) of the circumscribed rectangle obtained by the above process is smaller than th1, it is determined that W or H smaller than the threshold th5 is a noise / segment candidate, and H / W is greater than th6 or th
Those smaller than 7 are judged as line segment candidates, and the others are regarded as character candidates. This is summarized as follows. th5≤W (4) th5≤H (5) th7≤H / W≤th6 (6) W≤th1 (7) H≤th1 (8)

Character candidates satisfying all the conditions of the above formulas (4), (5) and (6) and satisfying at least one condition of the formulas (7) and (8) are the character candidates, and the formulas (7) ( Any one of the formulas (4), (5) and (6) that satisfies one or more conditions of 8)
Noise or line segment candidates are those that do not satisfy one or more conditions.

As described above, the candidates are those characters (including line segments) existing in the background, the frame, or the cells of the table. Therefore, the attribute of the inclusion label is determined. If not, the attributes of these candidates cannot be determined.

At this stage, since the background has been determined by the processing of FIG. 8, among these candidates, the attribute whose inclusion label is background is determined to be the attributes noise, line segment, and character. In the example of FIG. 7A, labels ~ (148), (150) ~
(365), (367) to (402), "Car Photo," is a character candidate. From FIG. 7B, the labels ~ (148), (150) ~ (3
Since the label 1 does not exist in the inclusion labels of 65) and (367) to (402), it remains a character candidate at this stage. These attributes are not determined until the attributes of the inclusion label are determined.

Next, description will be made regarding the attribute determination for the items that do not satisfy the conditions of the expressions (7) and (8). Formula (7)
If the condition (8) is not satisfied, th1 is obtained in advance from the average font size used in the text, so the circumscribed rectangle size such as heading characters, frames / tables, and patterns is relatively small. It's a big one. In order to distinguish them, the image density (= the number of effective pixels in the circumscribed rectangle / the circumscribed rectangle area, hereinafter referred to as ID), the width (W) and the height (H) of the circumscribed rectangle,
Dispersion (V) is used.

First, as shown in FIG. 9, the image density of the circumscribed rectangle
When the (ID) is th2 or less, the attribute is a pattern / frame / table candidate, and when at least one of W and H is th3 or more, the attribute is a pattern / frame / table candidate. When ID is larger than th2 and both W and H are smaller than th3, the luminance variance (V) is obtained based on the label image and the input image stored in the storage unit 103, and when V is th4 or more, Its attribute is a pattern
If it is a frame / table candidate and it is smaller than th4, its attribute is a character candidate. For this character candidate, if the inclusion label has an attribute of background as in the above, the attribute is determined as a character. In the example of FIG. 7A, ..., (14
9) and (366) are determined as picture / frame / table candidates.

Next, the black pixel group determined to be a picture / frame / table candidate in the processing of FIG. 9 is determined based on the characteristics of the white pixel group included in the area. Specifically, the following features are used. The table has a feature that it includes many white pixel groups including characters that can be assumed to be cells of the table.

The frame is characterized in that it contains a small number of white pixel groups and that the white pixel groups often contain a picture, a frame, and a table candidate. The picture has a feature that it has almost no white pixel group including characters that can be assumed to be a table cell.

As this pre-processing, it is determined whether the binarized OFF pixel group not determined to be the background in FIG. 8 is a cell candidate or a non-cell candidate. The determination method is performed by the number of character candidates included in the binarized OFF pixel group.

FIG. 10 is a flow for counting the number of character candidates included in the binarized OFF pixel group. When the binarized ON pixel group is a character candidate, the circumscribed rectangular feature amount of the inclusion label is used. Count up the number of certain character candidates. This is performed for all the binarized ON pixel groups. As a result, the number of character candidates included in each binarized OFF pixel group can be counted.

Based on this, in the flow chart as shown in FIG. 11, whether each binarized OFF pixel group is a cell candidate,
Determine if it is a non-cell candidate. The determination is made based on whether or not the number of character candidates included in the binarized OFF pixel group is the threshold th7 or more.
At the same time, the number of cell candidates and the number of non-cell candidates that are the circumscribed rectangle information of the inclusion label of the binarized OFF pixel group are counted up.

Next, referring to the flow chart shown in FIG.
Attribute determination is performed for frame / table candidates. First, if the number of non-cell candidates is less than th8 and the number of cell candidates is 0, it is assumed to be a frame, so it is determined to be a frame / table. In addition, it is assumed that the ratio of the number of cell candidate numbers in the number of binarized OFF pixel groups included in the group is larger than th9 is a table, so it is determined as a frame / table. Those that do not satisfy the above two conditions are determined to be patterns.

FIG. 13 is a flow of judgment for the area whose attribute is not determined at this stage. First, if the attribute of the inclusion label for the binarization OFF area is a picture, the attribute is set to a picture, and if it is a frame / table, the attribute is set to a cell. If neither is the case, it is undetermined and i is incremented. This is a binarization ON pixel group and a binarization O
This is because the inclusion relationship of FF pixels may be in several layers, and at this stage, character candidates exist as attributes of the inclusion label.

This will be described with reference to FIG. FIG. 14 (a)
Shows a car with letters in it, and FIG. 14 (b) is an enlarged view of the dotted circle. As shown in FIG. 14B, a frame portion (A) whose attribute is determined to be a pattern, a binarization OFF area (B) included in the frame portion, a character candidate (C) included in the binarization OFF area, 2 that the character candidate includes
It is structured as a digitization OFF area (D).

Since the attribute of A is a picture, the attribute of B is decided to be a picture, but the attribute of the inclusion label (C) of D is a character candidate. At this stage, the attribute of D cannot be determined as a pattern. This process is performed on all the binarized OFF groups whose attributes have not been determined.

Next, the attribute for the binarized ON pixel group whose attribute has not been determined is determined. If the attribute of the inclusion label for the binarized ON area is a picture, the attribute is a picture, and if it is a cell, the attribute is a character. If neither is the case, it is undetermined and j is incremented. This process is performed for all binarization ON groups whose attributes are undetermined. This j
Is the same as the above-mentioned i, and therefore its explanation is omitted. Figure 1
In the example of 4, the attribute of C is determined to be a pattern at this stage.

At this stage, if it is not checked whether i = 0 and j = 0, i and j are once reset to 0, the above process is repeated again, and the process is repeated until j becomes 0. In the example of FIG. 14, since the attribute of the inclusion label D is a picture, the attribute of D is also decided as a picture, and all the attributes are decided.

The attribute data of various areas obtained as described above is output to the outside together with the image data.

Next, another example of this embodiment will be described. In another example, steps S203, S204, S20 described above.
5, the following 10 types of information (A ′) to (J ′) are acquired as circumscribed rectangle information for white pixels of the binary image,
Using that, the attribute data of various areas is fixed,
Furthermore, by adding the color information (K ') of the white pixel group (area) to the rectangular information, the background and table / frame cells are white background, table / frame cells, non-white background, table / frame. The cells and attributes of can be subdivided.

The color information is obtained by referring to the input image stored in the storage unit 103 corresponding to the white pixel. A known representative color determination method may be used as the method for obtaining the color.

Since the color information is added to the rectangle information, the correction processing for the image data is performed based on this attribute data.
When performing (spatial filter, background removal, brightness correction, saturation correction, etc.), it is possible to make finer settings than the above-described embodiment. (A ') Width of circumscribed rectangle (B') Height of circumscribed rectangle (C ') Left edge coordinate value of circumscribed rectangle (D') Right edge coordinate value of circumscribed rectangle (E ') Top edge coordinate value of circumscribed rectangle (F' ) Bottom coordinate value of the circumscribed rectangle (G ') Area of the circumscribed rectangle (H') Number of character inclusions: Number of character candidates directly included in OFF pixel group (area) (I ') Inclusion label: Direct inclusion of OFF pixel group ON pixel group (J ') attribute (K') color information

In the above-described embodiment, characters,
It is stated that attributes such as noise / line segment, background, table / frame, picture, and cell of table / frame are output to the outside together with the image data, but the invention is not limited to this. Using the attribute information obtained here, the above attributes may be combined and output as one attribute depending on the purpose of the processing to be performed.

For example, it is possible to make the attributes of the background and the character have the same attribute and output as the character area. In particular, when performing the correction processing on the image data based on the attribute data as described above, it is easier to handle the character and the area directly including the character as a character area.

In step S205 described above,
For some binary ON pixel groups, the luminance distribution is obtained from the input image stored in the storage device 103 by referring to the image data corresponding to the ON pixel. In addition, it is not essential information for all labels. However, it is not necessary to obtain the variance of luminance in step S205, and it is not necessary to calculate in step S202
You may ask for it at the stage. Similarly, the same applies to the color information extraction of the binarization OFF group, and the extraction may be performed in the step S203.

[0072]

As described above, according to the present invention,
By using the circumscribed rectangle information obtained by labeling each of the binarized ON pixel and the OFF pixel and the inclusion relation information of the binarized ON pixel group and the OFF pixel group,
Since the areas such as the table / frame, the photograph, and the background are divided, there is an effect that the division accuracy of the complicated area structure can be improved.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an apparatus that realizes an image processing method of the present embodiment.

FIG. 2 is a flowchart showing a flow of area identification processing executed by an area identification unit.

FIG. 3 is a schematic diagram illustrating an operation of labeling black pixels.

FIG. 4 is a schematic diagram illustrating an operation of labeling white pixels.

FIG. 5 is a schematic diagram illustrating an operation of an inclusion relation.

FIG. 6 is a diagram illustrating an operation of attribute determination.

FIG. 7 is a diagram illustrating a processing result.

FIG. 8 is a flowchart of background extraction.

FIG. 9 is a flowchart of attribute detection of ON pixels.

FIG. 10 is a flowchart of counting the number of character candidates.

FIG. 11 is a flowchart for determining a cell candidate.

FIG. 12 is a flowchart of attribute determination for a picture / frame / table candidate.

FIG. 13 is a flowchart of determination for an area whose attribute is not fixed.

FIG. 14 is a diagram illustrating a determination for a region whose attribute is not fixed.

[Explanation of symbols]

101 ... Image input section, 102 ... CPU, 103 ... Storage section, 104 ... Area identification section

Claims (12)

[Claims] 1. A binarization step of separating an input image into an edge corresponding to a black pixel and a non-edge corresponding to a white pixel, and a black pixel of a binary image binarized in the binarizing step. On the other hand, the same labeling is added to consecutive pixels in the vertical, horizontal, and diagonal directions, and a first labeling step of modeling a circumscribed rectangle of a pixel group with the same label; A second labeling step in which the same label is added to the pixels that are continuous in the vertical and horizontal directions with respect to the white pixels of the binarized image, and the circumscribed rectangle of the pixel group with the same label is modeled, The inclusion relation detection step of detecting an inclusion relation between the circumscribed rectangle labeled in the first labeling step and the circumscribed rectangle labeled in the second labeling step, and a result of the inclusion relation detection step. When, the related labeled circumscribed rectangle in the first labeling step 1
Information and the second information on the circumscribed rectangle labeled in the second labeling step, the attribute detection step of distinguishing attributes for all pixels, and based on the detection result of the attribute detection step. And a region dividing step of dividing the input image into regions. 2. The image according to claim 1, further comprising a tint detection step of obtaining a tint for each label from an input image corresponding to pixels to which the same label is added in the second labeling step. Processing method. 3. The image processing method according to claim 1, wherein the types of attributes distinguished by the attribute detection step include at least characters, figures or photographs, tables or frames, and backgrounds. 4. The attribute type distinguished by the attribute detecting step includes at least a character, a figure or a photograph, a table or a frame, a background, and a table / frame cell. The described image processing method. 5. The image according to claim 4, wherein the types of attributes distinguished by the attribute detection step are sub-classified by using at least one of saturation and lightness of cells of a background and a table / frame. Processing method. 6. The first information is the width, height, area, pixel density, and variance of the circumscribed rectangle, and the second information is
The image processing method according to claim 1, wherein the circumscribed rectangle has a width and a height. 7. A binarization step of converting an input image into a black-and-white binary image and separating it into edges and non-edges, and for black pixels of the binary image binarized in the binarization step. , A first labeling step in which the same label is added to consecutive pixels vertically, horizontally, and diagonally, and a circumscribed rectangle of a pixel group having the same label is modeled, and binarization in the binarization step A second labeling step in which the same label is added to consecutive pixels in the vertical and horizontal directions with respect to the white pixels of the binarized image, and the circumscribed rectangle of the pixel group with the same label is modeled; An inclusion relation detecting step of detecting an inclusion relation between the circumscribed rectangle labeled in the first labeling step and the circumscribed rectangle labeled in the second labeling step; a result of the inclusion relation detecting step; The related labeled circumscribed rectangle in the first labeling step 1
Information and the second information on the circumscribed rectangle labeled in the second labeling step, the attribute detection step of distinguishing attributes for all pixels, and based on the detection result of the attribute detection step. And a region dividing step of dividing the input image into regions. 8. The image according to claim 7, further comprising a tint detection step of obtaining a tint for each label from an input image corresponding to a pixel to which the same label is added in the second labeling step. Processing program. 9. The image processing program according to claim 7, wherein the types of attributes distinguished by the attribute detection step include at least characters, figures or photographs, tables or frames, and backgrounds. 10. The attribute type distinguished by the attribute detection step includes at least a character, a figure or a photograph, a table or a frame, a background, and a table / frame cell. The image processing program described. 11. The image according to claim 10, wherein, in the type of attribute distinguished by the attribute detecting step, cells of a background and a table / frame are finely classified by using at least one of saturation and lightness. Processing program. 12. The first information is the width, height, area, pixel density, and variance of the circumscribing rectangle, and the second information is
The image processing program according to claim 7, wherein the circumscribing rectangle has a width and a height.