JP2001175858A - High-speed labeling system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To allocate label values while shortening a retrieval time as compared with the retrieval of all patterns without managing label allocation as to all window patterns as a high-speed labeling system which allocates temporary labels to respective pixels of image information and converts them into real labels. SOLUTION: This system is equipped with an interest-pixel and nearby-pixel extraction part which matches pixels at the same positions of a labeling line including a pixel of interest as an object pixel of label value allocation and a pixel of interest for extracting pixels nearby it and the reference line right above the line and detects plural pixels until a discrepancy is detected and an allocation knowledge matching part which matches labeling knowledge information having a prescribed temporary label giving method for previously given nearby pixel environment, and further equipped with a temporary label constitution part which updates the connection relation between labeling and labels according to the temporary label giving method.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a high-speed labeling system for separating and distinguishing a plurality of regions in an image.

In recent years, as a peripheral device for inputting financial documents, business documents and the like to an information processing system, a demand for a character recognition device (OCR: Optical Charactor Reader) has been increasing.

In the OCR, in order to cut out characters and increase the recognition rate of individual characters, it is important to accurately perform cutout processing, which is a pre-recognition stage. Until now, for a document (such as a form in which the text is written in the same color or density as the black frame or ruled line, instead of the dropout color), the characters are specified within the specified range. If it was written correctly, characters could be recognized automatically at a considerable rate. However, the recognition rate would drop significantly if the frame or ruled line indicating the specified range exceeded or slightly exceeded the specified range. Was.

[0004] As described above, labeling processing is used to perform preprocessing for accurately cutting out only characters from characters touching ruled lines and frames, and it is desired to increase the speed. Note that labeling processing technology is not limited to OCR, but can be broadly defined as a straight line such as a print character recognition device, cutting out characters and symbols in drawing recognition, and separating ruled lines in an image from contact portions of objects, figures, and characters. It is also used when extracting only the figure from the pattern in which the figures are overlapped.

[0005]

2. Description of the Related Art FIG. 11 is an explanatory view of a conventional example.

A conventional raster scanning type labeling method will be described. First, in the first raster scan, temporary labels are assigned to the image parts that are not the background, the connection relation between the temporary labels is detected, and this connection relation is rearranged to convert the temporary labels into actual labels. Create a table,
Subsequently, a process of converting the temporary label image into an actual label image is performed by referring to the conversion table in the second raster scan.

When the raster scanning labeling process is executed, the process shown in FIG. Set the window as shown in
While shifting this window by one pixel in the horizontal direction / vertical direction, a temporary label is assigned to the target pixel while looking at the labels of the pixels included in this window. That is, as shown in FIG. 11A, when the target pixel (pixel to which a temporary label is to be assigned) is “1” (the value of a pixel representing an image of a character or the like), four adjacent pixels The labels of the pixels (the upper left pixel, the pixel immediately above, the upper right pixel, and the left pixel of the target pixel) are identified, and if any of the 4 pixels has a label other than 0, the label B. in FIG. As shown in FIG. 11, the label value is set as the temporary label value of the pixel of interest. The new label value is set as the temporary label value of the target pixel as shown in FIG.

FIGS. 12 to 14 are explanatory diagrams (part 1) to (part 3) of a conventional labeling process.

In the case where the binarized image shown in FIG. 12A is to be processed, in the first raster scan, when the pixel of interest is connected to a pixel to which a temporary label is assigned for an image portion other than the background, By assigning the temporary label to the target pixel and assigning the temporary label to the pixel to which the temporary label is assigned, the temporary label as shown in FIG. , 3 are detected as shown in FIG. 13 (a). Next, by rearranging this connection relationship, the temporary label 1 is not connected to any temporary label, the temporary label 2 is connected to the temporary label 1, the temporary label 3 is connected to the temporary label 1, and the temporary label 4 is connected to the temporary label 1. FIG. 13 shows that no connection is made with any temporary label.
By creating the connection table shown in (b) and further organizing this connection table, the temporary label 1 has the real label 1, the temporary label 2 has the real label 1, and the temporary label 3 has the real label 1,
In FIG. 13, the actual label 2 is assigned to the temporary label 4.
Create the conversion table shown in (c). Note that a real label 0 is assigned to the background. Subsequently, in the second raster scan, the temporary label image is converted into the real label image while referring to this conversion table, so that the real labels as shown in FIG. 14 are allocated.

According to the conventional method of assigning temporary labels by shifting the assignment of temporary labels one pixel at a time using a window as shown in FIG. 11, there is a problem that it takes time to assign temporary labels.

In order to improve this, the same applicant as the present invention has previously described "temporary label allocation processing method and actual label allocation processing method" (Japanese Patent Application No. 5-237552: Japanese Patent Application Laid-Open No. 7-935).
No. 39) has been proposed. The outline of the proposed method is as follows. Control is performed such that a window having a size of 2 × X pixels is moved in the horizontal direction without overlapping, and shifted by one pixel in the vertical direction. Is used as a reference line, and the lower row of the window is set as a target line, and a plurality of label allocation rules that describe a temporary label allocation method to be allocated to each pixel in the lower row are managed in association with the binarization pattern indicated by the pixels in the window. When the window moves to a new position, the binarization pattern of the pixel in the window is specified, a corresponding label allocation rule is searched according to the specified binarization pattern, and according to the searched label allocation rule, A temporary label is assigned to each pixel in the lower row of the window.

FIG. 15 shows an example of labeling according to the proposed method.

In FIG. 15, a window pattern is set in each row represented by a to h, a and b are patterns of 2 rows × 4 images (4 pixels in the upper row and 4 pixels in the lower row), and c to h are 2 × 2
(Upper two pixels and lower two pixels), the pixels constituting each pattern are white for the background, black for the image (character or figure line), and the upper pixel for each reference line. , Each pixel in the lower row is the pixel of interest (all lower images)
It is.

In FIG. 15, the pattern a is an example in which the upper and lower four pixels are all background pixels. In this case, no label is assigned to the lower four pixels, and 0 representing the background is replaced by 0. Assigned. This is an example in which the upper and lower four pixels of the pattern b are all pixels of the image. In this case, the label for the lower four pixels is the same as the label (represented by X) assigned to the upper reference line. X is given. In the case of c and d, the target pixel is a background pixel (a value of 0), and in this case, no label is assigned. In the case of e, Y (new temporary label number) is assigned to the pixel on the right side of the target pixel at the lower stage, and is not assigned on the left side (leave 0). The case of f is a case where the upper and lower two pixels are “01”, and the same label (represented by X) as the label allocated to the upper right pixel is assigned to the lower right pixel of interest. . Or g
In the case of the pattern of (1), a label Z is assigned to a pixel on the lower left side of the target pixel. However, if the immediately preceding pixel is a black pixel, Z is given the same label, and if the immediately preceding two pixels are white pixels, Z is a new temporary label number. In the pattern of h, a label Z is assigned to the pixel on the lower right side of the target pixel as shown in the assignment method.

FIG. 16 is an explanatory diagram of the actual label allocation processing method according to the above proposed method, where は is a circumscribed rectangle of an image (a rectangle circumscribed to a graphic representing an image), and Represents a rectangle. In the conventional method that uses the initial value of the real label as the initial value of the temporary label, the actual label cannot be assigned accurately. To prevent this, in the method of FIG. The temporary label is assigned in advance by making the initial value of the temporary label different from the initial value of the real label. In the example shown in FIG.
The temporary labels L and L + 2 are allocated in the circumscribed rectangle of 1, L + 2.

When converting the temporary label of each circumscribed rectangle shown in (a) into a real label, the temporary label 1 is replaced with the real label 1
And the temporary label (L + 1) as the real label 1 and the temporary label (L
+2) is converted to the real label 2 and the real labels below L are left as they are. As a result, the temporary label L / (L + 1) in the circumscribed rectangle as shown in FIG. 16B is converted into the real label 1 and the temporary label (L + 2) is converted into the real label 2
Is converted to Subsequently, the conversion process is performed on the inside of the circumscribed rectangle, so that the temporary label (L + 2) of the circumscribed rectangle is converted into the real label 2 as shown in FIG. Keeps the value as it is, and accurate real label assignment is realized.

[0017]

It is difficult to create all the window patterns by the above-described method of assigning labels to the window patterns, and when all the patterns are created, which pattern should be used? There is a problem that it takes time to search for.

An object of the present invention is to provide a high-speed labeling method capable of allocating label values with a shorter search time than searching all patterns without managing label allocation for all window patterns. .

[0019]

FIG. 1 is a diagram showing the principle configuration of the present invention. In the figure, 1 is an all white line extracting unit, 2 is a pixel of interest and neighboring pixel extracting unit, 3 is an assigned knowledge collating unit, 4 is an assigned knowledge storing unit, 5 is a temporary label forming unit, 6 is a true label forming unit, 7 Is a storage unit, 7a is an image data storage unit obtained by raster scanning, 7b is a black pixel detection position storage unit, 7a
Reference numeral c denotes a target pixel and neighboring pixel storage unit, 7d denotes an allocation knowledge collation result storage unit, 7e denotes a temporary label data storage unit, and 7f denotes a true label data storage unit.

According to the present invention, the pixels in the vicinity of the target pixel are labeled at once according to a preset rule.

First, with respect to the image data obtained by the raster scanning, when the image data is obtained by the raster scanning, the all white line extracting unit 1 stores the image data in the image data storage unit 7a of the storage unit 7. With respect to this image data, a line is extracted in which each pixel has no pixel (pixel having a bit value representing a line or black) in the image in the row direction (or the column direction). At this time, when all the white lines are extracted, the succeeding lines are also extracted to detect the continuation of the white lines, and the true label forming unit 6 is driven for these all white lines, and the true label (real label) is output. The same as above). When a black pixel is detected during the extraction operation of the all white line extraction unit 1, the black pixel position is stored in the black pixel detection position storage unit 7b of the storage unit 7. The extraction operation of the pixel of interest and the neighboring pixel extraction unit 2 is performed on the line where the black pixel is detected, and the labeling line and the uppermost line are sequentially arranged in the rectangular area of the image including the black pixel from the topmost line in the vertical direction. The corresponding pixel of the reference line is compared and the number of matching pixels is detected to extract a target pixel and a neighboring pixel, and stored in the target pixel and neighboring pixel storage unit 7c of the storage unit 7. Subsequently, the operation of the assignment knowledge collating unit 3 compares the contents of the assignment knowledge storage unit 4 in which the assignment method of the temporary label corresponding to the pattern of the target pixel state is specified in advance. Storage unit 7
Is stored in the provisional label collation result storage unit 7d, a provisional label is allocated to the labeling line in the provisional label configuration unit 5 according to the provisional label allocation method, and is stored in the provisional label data storage unit 7e of the storage unit 7. Subsequently, a true label (real label) is allocated by the true label forming unit 6 using a conventional technique according to the existence of the connection relationship of the temporary label, and stored in the true label data storage unit 7f of the storage unit 7. The contents of the true label data storage 7f can be output as a print or a display.

[0022]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention can be implemented in an information processing apparatus for performing data processing by a program, and will be described below together with a processing flow of each section shown in FIG.

FIG. 2 is a processing flow for extracting the all-white line, which is a flow for realizing the function of the all-white line extracting unit 1 in FIG. First, the binary image data (area 7 in FIG. 1)
is read (S1 in FIG. 2), and first, a variable y representing the attribute of the line (either an all white line or a line including black pixels) is set to 0 (S2). A black pixel on the top (top row) horizontal line is searched (S3 in FIG. 2), and it is determined whether there is a black pixel (S4). If there is no black pixel, an all white line is determined (S5). In this case, it is determined whether the previous line is also an all-white line (S6 in FIG. 2). If the previous line is not an all-white line, it is a new all-white line, and a true label (real label) is generated by the true label component. Is assigned (S7). In this case, y = y2 is set. If the previous line is also an all-white line, y = y3 is set and the process moves to the next vertical line (next line) (S12 in FIG. 2).
Subsequently, it is determined whether or not all lines have been completed (S13). If not completed, the process proceeds to S3 and the same processing is repeated. If completed, the processing of the true label forming unit is executed.

When a black pixel is detected in S4, y = y1 is set, and the position of the black pixel is stored in the storage unit (S8 in FIG. 2). This black pixel position is used as the start position of the process in the process of the next pixel of interest and neighboring pixel extraction unit (S9). The details of the processing of the target pixel and the neighboring pixel extraction unit are shown in FIG. 4 described later.

Subsequent to the process of the pixel of interest and neighboring pixel extraction unit, the process of the assignment knowledge collation unit (S10) and the process of the temporary label configuration unit (S11), which will be described later, are sequentially executed.

FIG. 3 shows a specific example of the all white line extraction processing. FIG. 3 (1) is a binary image obtained by raster scanning. On the other hand, the result shown in FIG. 3 (2) is obtained by the process of extracting all white lines in FIG. That is, all white lines y2 and y3 are located below the line y1 including the black pixels.
Exists, and a line of y1 exists below it. The shaded portion represents the area within the white line.

FIG. 4 is a processing flow for extracting the target pixel and the neighboring pixel, and is a processing flow for realizing the function of the target pixel and the neighboring pixel extracting unit 2 in FIG. First, a horizontal size (xx) and a vertical size (yy) are given as known values in a file in advance as image data to be processed, and these sizes are set at the start of the process (S1 in FIG. 4). . Next, a variable y representing the vertical direction (y direction) is set to 0 to perform processing from the first row (S2 in FIG. 4),
This line is used as a labeling line, and a reference line for the line is also set (S3). This reference line is a line above the topmost labeling line (line of y = 0) to which a label is first assigned, but since this topmost image does not exist, virtually all lines are present. It is set to 00 to 0 having the same value as the background (0) (xx pixels in the horizontal direction).

Next, the head position of the black pixel (stored in S8 in FIG. 2) is set to the initial value of x representing the horizontal position (S4 in FIG. 4), and the initial value of the previous pixel is set to 0 (background). (S5). In this case, as the previous pixel, a pixel whose initial value is virtually zero is set immediately before the first pixel of both the reference line and the labeled line in the horizontal direction (x direction). Here, one of the following two is executed according to the data structure of each pixel of the image. That is, when one pixel is composed of 8 bits or more (multi-valued), one pixel (consisting of 8 bits) is set as a target pixel (S6 in FIG. 4), and one pixel is composed of 1 bit. If so, 8 pixels (composed of 8 bits) are set as the target pixel (S7).

Subsequent to S6 or S7, the same reference / labeling line pattern is obtained (S8 in FIG. 4). In this method, pixels at the same position (x direction) of a reference line and a labeling line arranged above and below are collated with each other to obtain a pattern having the same value. As a result, for example, if the first pixel and the m pixels match and the (m + 1) th pixel does not match, the m pixels of the reference line and the first pixel of the labeling line become neighboring pixels, and the labeling line Excluding the first pixel (previous pixel) of (m−
1) Pixels are pixels of interest.

Subsequently, the processing of the assignment knowledge collating unit described later is performed (S9 in FIG. 4). In the allocation knowledge collation processing, the number of matches due to the correspondence between the pixels of the reference line and the labeling line may increase, and the number is set to n (n ≧ m). Next, the value of x is updated to a pixel adjacent to the pixel processed as described above (S10 in FIG. 4), and it is determined whether or not there is a pixel at a position corresponding to the value of x (S1).
1) In some cases, the last pixel that has been matched in the processing of S8 is set as the next previous pixel (S12), and the processing of S6 is performed.
Alternatively, the process returns to S7 to perform the process of extracting the target pixel and the neighboring pixels for the new line.

If there is no pixel corresponding to the updated x in the process of S11, it means that the pixel to be processed for the line has been completed, and the provisional label forming unit (described later) (Step S13 in FIG. 4). Thereafter, a labeling line is set for the next reference line (S1 in FIG. 4).
4) The value of y is incremented by 1 and updated (S15). Then y
Is determined to be equal to or smaller than the last line (corresponding to the size yy) in the y direction (S16 in FIG. 4). In the following cases, the processing in S4 is performed for the next labeling line, and the same processing is performed thereafter. Repeated. When the image reaches the last line in the y direction of the image, the assignment of a true label (the same as the actual label) by the true label forming unit is performed in the same manner as in the related art (S17).

FIG. 5 shows a specific example of extracting a neighboring pixel and a pixel of interest by initializing the leading pixel of each line.

When labeling is performed from the left end for each line, the temporary previous pixel is set to 0 and added to the left of the leftmost pixel by the processing of S5 in FIG. In the example of FIG. As shown in the figure, the temporary label is “0010002220333044440” on one reference line in the image.
And the pixel of the labeling line following this line is “0001110000101111” as shown in the figure.
0 ”pattern. In this case, the label of the first pixel of the reference line is 0, and the first pixel of the labeling line is 0. However, by the processing of S5 in FIG. 4, the imaginary previous pixel whose initial value is 0 (background) is the first pixel. 5 is added before the pixel, and the processing of S8 in FIG. And the neighboring pixel (0001 of the reference line and 0 at the beginning of the labeling line) coincide with each other.

By setting the previous pixel in this way, it is not necessary to identify whether the pixel is the leftmost end in the process of extracting the pixel of interest and the neighboring pixel in the middle part (position where the value of x is not 0) of the subsequent line. Labeling can be performed in a similar flow.

FIG. 6 shows a specific example of extracting a target pixel and a neighboring pixel for the first line.

In the example of FIG. As shown in (1), the pixels of each line constituting the image area are arranged in order from the top line. When the head line (y = 0) of this image area is used as a labeling line and the pixel of interest and neighboring pixels are extracted, the A.P. (The size in the horizontal direction is the same as the size of the image area.) When the pixel of interest and the neighboring pixels for this top line are extracted, the processing including the processing of setting the top pixel before as shown in FIG. Pixel of interest (3 □ pixels) as shown in
And the neighboring pixels (the four 0 pixels on the reference line and the first 0 pixel on the labeling line) are extracted.

As described above, when the first line is not an all-white line, by setting the all-white line as a reference line, it is not necessary to determine whether the first line is the first line or not. Similarly, labeling can be performed.

FIGS. 7 and 8 show embodiments (part 1) and (part 2) of the allocation knowledge. As the allocation knowledge, there are prepared a large number of patterns that are compared with the pattern of the target pixel and the neighboring pixels obtained by the processing shown in FIG. 4 in the allocation knowledge matching unit (3 in FIG. 1).

In FIGS. 7 and 8, the state of the pixel of interest is set in the first column, and after that, each column of the target pattern example, labeling result and label allocation method is provided. The first (a) shows the case where both the reference line and the labeling line are all white pixels, "0000", "00", "000".
The labeling result of the pattern of “0000” is assigned a pattern of 0 label as shown in the figure. (b) is a case where both the reference line and the labeling line are all black pixels. In this case, the same label (X) as the label of the reference line is attached as shown in the figure. (c) is a case where the reference line is a white pixel (all), the labeling line is a black pixel, and the right end is a white pixel. If there is a black pixel in the immediately preceding pixel, it becomes Y0 (where Y is the immediately preceding pixel). , Otherwise Y assigns a new label).
In the case of (d), the labeling line is a continuous white pixel (the reference line is irrelevant), and the label allocation is all zero.

FIG. 8E shows the case where the reference line is white pixels (all), the left side of the labeling line is white pixels, and the right side is black pixels, and the labeling is 0Y, 00Y, 000Y.
YYY. Here, Y is a new temporary label number. (f)
Indicates a case where the labeling line has the same pattern as the reference line and the left end is a white pixel, and a temporary label as shown in the labeling result in the figure is allocated. (g) shows a case where the labeling line has the same pattern as the reference line and the left end is a black pixel, and a temporary label is assigned as shown in the labeling result in the figure. (h): When the left side of the reference line is a white pixel and the right side is a black pixel, the left side of the labeling line is a black pixel and the right side is irrelevant. The label assignment method is as follows. If the two pixels immediately before the label are white pixels, Z has the same label as the reference line.

With respect to the allocation knowledge shown in FIGS. 7 and 8, the order of each pattern is updated in accordance with the frequency of use, thereby enabling efficient collation.

The target pixel and the neighboring pixel obtained by the processing flow of extracting the target pixel and the neighboring pixel in FIG.
The collation with each pattern of the allocation knowledge shown in FIG. 8 is performed in the allocation knowledge verification unit (3 in FIG. 1). This process is a process of simply comparing patterns, and can be realized by a conventional technique.

FIG. 9 is a processing flow for provisional labeling.
It is executed after the assignment by the assignment knowledge collating unit (3 in FIG. 1).
This implements the function of the temporary label forming unit 5 of FIG. First, the number of target pixels (referred to as Nc) detected in the process of extracting the target pixel and the neighboring pixels is set (S1 in FIG. 9). Next, it is determined whether or not all the pixels of interest are 0 (S2 in FIG. 9).
(A) corresponds to the state of the pixel of interest, and Nc
A value 0 is set as a label for each (S3). S2 above
When it is found that all the pixels are not 0, it is determined whether or not all the pixels have the same L (label number) (S in FIG. 9).
4) If they are the same, the state corresponds to the state of the target pixel in FIG. 7B, and the Nc temporary labels of the target pixel are set to L (number) (S5). Subsequently, the rectangle information of the label L is updated (S6), and y = y + 1. Since the value of the same label continues from the upper line to the target line, the value of the rectangle information representing the size of the label in the row direction is incremented by one. After S6, the process proceeds to S10 described later.

In FIG. 9, if none of S2 and S4 is satisfied, it is next determined whether or not the label is the same as the same position on the reference line (S7 in FIG. 9). If they are the same, this corresponds to the case where (f) or (g) in FIG. 8 is modified, and the labels of the Nc reference lines are copied as labels for the target pixel (S8 in FIG. 9). The rectangle information of the label is updated (S9). Subsequently, it is determined whether or not the labels need to be integrated by checking the connection relationship with other labels (S10 in FIG. 9). If necessary, the labels of the left pixel and the upper left pixel are integrated (S11), and If there is no, it ends. If the determination in S7 is NO, a label K (number) is set for the black pixel (S12), the rectangular information of the label K is updated (S13), and the process proceeds to S10.

FIG. 10 shows a specific operation procedure for provisional labeling.

It is assumed that the reference pixel and the labeling line for the target pixel have eight pixels each having a pattern as shown in FIG. 10 (FIG. 10). When this pattern is collated with the assignment knowledge as shown in FIGS.
(F) of FIG. 8 (the same pattern as the reference line and the left end is a black pixel).
of). As a result, the pixel of interest is searched for while applying (f) (FIG. 10). That is, eight pixels as described above are set as the pixel of interest.
By applying, the pattern of the subsequent reference line and the labeling line are also searched, and it can be seen that the pattern eventually matches up to 25 pixels. Next, a neighborhood label (reference line label) including the previous pixel is extracted (FIG. 10). In the example of the figure, the label of the previous pixel is Z, and A, B, C,
Labels D, E, and F are attached. Subsequently, by forming a temporary label by the label allocation method defined in the above (f) (FIG. 10), the label of the labeling line is allocated by copying from the reference line. Thereafter, rectangular information of labels A to F (coordinates x1, y1 at the upper left of the rectangle)
(Represented by coordinates x2 and y2 at the lower right) (FIG. 10). In this example, for each of the temporary labels A to F, the value of y2 of the rectangular information is incremented by 1 (increased to the lower row).

After the provisional labels are assigned to the respective labeling lines in this manner, the assignment of the true labels (real labels) is performed by the true label forming unit 6 of FIG. 1 using the same technique as the conventional technique. Done.

[0048]

According to the present invention, label values can be effectively allocated because it is not necessary to manage label allocation for all window patterns and the search time can be shortened compared to searching for all patterns. be able to. Further, the capacity and speed of the label rectangle can be reduced. This is particularly effective when dealing with business documents including forms and the like.

According to the measurement results of the actual test by the applicant, the processing time is reduced by 40% or more compared with the conventional labeling.
It decreased by 60%.

[Brief description of the drawings]

FIG. 1 is a diagram showing a principle configuration of the present invention.

FIG. 2 is a diagram showing a processing flow of all white line extraction.

FIG. 3 is a diagram illustrating a specific example of an all white line extraction process.

FIG. 4 is a diagram showing a processing flow for extracting a target pixel and neighboring pixels.

FIG. 5 is a diagram showing a specific example of extraction of a neighboring pixel in which a head pixel of each line is initialized and a target pixel.

FIG. 6 is a diagram illustrating a specific example of extraction of a target pixel and neighboring pixels for a first line.

FIG. 7 is a diagram illustrating an example (part 1) of assignment knowledge;

FIG. 8 is a diagram showing an example (part 2) of allocation knowledge.

FIG. 9 is a diagram showing a processing flow of provisional labeling.

FIG. 10 is a diagram showing a specific operation procedure of provisional labeling.

FIG. 11 is an explanatory diagram of a conventional example.

FIG. 12 is an explanatory diagram (part 1) of a conventional labeling process.

FIG. 13 is an explanatory diagram (part 2) of a conventional labeling process.

FIG. 14 is an explanatory diagram (part 3) of a conventional labeling process.

FIG. 15 is a diagram showing an example of labeling according to the proposed method.

FIG. 16 is an explanatory diagram of an actual label allocation processing method according to a proposed method.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 All white line extraction part 2 Target pixel and neighboring pixel extraction part 3 Assigned knowledge collation part 4 Assigned knowledge storage part 5 Temporary label construction part 6 True label construction part 7 Storage part 7a Image data storage part 7b Black pixel detection position storage part 7c Target pixel and neighboring pixel storage unit 7d Assigned knowledge collation result storage unit 7e Temporary label data storage unit 7f True label data storage unit

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Koichi Kanamoto 4-1-1, Kamiodanaka, Nakahara-ku, Kawasaki-shi, Kanagawa Prefecture Inside Fujitsu Limited (72) Inventor Katsushi Ohara 4-chome, Kamiodanaka, Nakahara-ku, Kawasaki-shi, Kanagawa Prefecture No. 1 Fujitsu Limited (72) Inventor Watichi Chiba 4-1-1 Kamiodanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture Fujitsu Limited (72) Inventor Toshiyuki Huitada Kamiodanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture 4-1-1 1-1 Fujitsu Limited (72) Inventor Kazunori Yamamoto 4-1-1 Kamikadanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture Fujitsu Limited (72) Inventor Hiroshi Katsumata, Nakahara-ku, Kawasaki City, Kanagawa Prefecture 4-1-1 Odanaka F-term in Fujitsu Limited (Reference) 5B029 BB03 EE12 5B050 BA16 EA02 EA06 5L096 AA07 CA09 GA34

Claims (7)

[Claims] 1. A high-speed labeling method in which a temporary label is assigned to each pixel of image information obtained by raster-scanning an image and converted to a real label, and is a pixel to which a label value is assigned. To extract the pixel of interest and its neighboring pixels, compare the pixels at the same position on the labeling line containing the pixel of interest with the reference line in the row above it, and detect multiple pixels until a mismatch is detected , A neighboring pixel extracting unit, an assignment knowledge collating unit for collating with a labeling knowledge information defining a tentative label assigning method based on a given neighboring pixel environment, and updating a labeling and label connection relationship according to the tentative label assigning method A high-speed labeling method, comprising: 2. The high-speed labeling method according to claim 1, wherein the application order is updated in descending order as appropriate for the knowledge information to be applied having a high use frequency. 3. The high-speed labeling method according to claim 1, further comprising means for storing a pattern of a pixel immediately before a pixel of interest on an arbitrary labeling line and a reference line. 4. The high-speed labeling method according to claim 3, wherein an initial of a previous pixel of an arbitrary labeling line and a first pixel of a reference line is set as a white pixel to be distinguished from neighboring pixels of the previous line. method. 5. The high-speed labeling method according to claim 3, wherein the same processing as the lines other than the head can be performed by setting the reference line of the head labeling line to the line of all white pixels. 6. An apparatus according to claim 1, further comprising an all-white-line extracting unit for detecting the presence or absence of a black pixel on an arbitrary line and detecting an all-white pixel line. A high-speed labeling method characterized in that a temporary label area is reduced as much as possible by updating a temporary label to a true label. 7. The method according to claim 6, wherein all the white pixel lines are 2
A high-speed labeling method characterized in that temporary label update processing is not performed if the number of lines has been continued.