JP2000339456A - Picture processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To speed up the picture of a complicated form in the labeling processing of a binary picture. SOLUTION: A labeling processing part 10 scans the binary picture of a binary picture memory 20 and starts labeling at a new value (n) from a white pixel with the white pixel as start when the white pixel to which a label is not given is detected. When adjacent pixels are checked in order in a line direction from the start and the pixel is the white pixel to which the label is not given, the label (n) is given to the pixel. When the upper/lower adjacent pixels of the pixel are the white pixels to which the labels are not given, they are accumulated in a stack 12. When the labeling of the pixel connected to the start is terminated, the head pixel of the stack 12 is taken out and a same processing is repeated with the pixel as start. The processing is equivalent to the depth priority search algorithm of a multiple tree and the white pixels connected to the white pixel which is detected at first are detected without omission and they can be labeled.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an image processing method, and more particularly to an image labeling process.

[0002]

2. Description of the Related Art An important field of image processing is image recognition based on images. In recognition of an object from an image, a portion of a desired object is cut out from the image, a feature amount of the object portion as an image is obtained, and recognition is performed from the feature amount. In general, a technique of binarizing and labeling an image is often used to cut out a target portion from the image.

[0003] Binarization is a process in which the pixel value of each pixel of a grayscale image is compared with a threshold value and each pixel is classified according to whether it is larger or smaller than the threshold value. By appropriately selecting a threshold value and performing binarization, a target portion can be extracted. Labeling extracts a connected component, that is, a group of pixels having a pixel value of interest (either white or black) and connected to each other in a binarized image, and assigns a label for identification to each connected component. This is the processing to be performed. For labeling,
A case where the connectivity between the upper and lower neighboring pixels of the own pixel is considered (four connectivity), and a case where the connectivity with surrounding eight pixels which is obtained by adding four neighboring pixels in the oblique direction is considered (eight connectivity). And are generally known. By this labeling process, 2
Each pixel having the pixel value of interest in the valued image can be divided into a block (connected component) of connected pixels. Each connected component obtained in this way is to be recognized, and the feature amount is obtained for each connected component.

For example, as a system for detecting an intruder from a surveillance camera image, a difference image between a current image and an image immediately before the current image is obtained, and the difference image is binarized and labeled to obtain each connected component. It is known that the presence or absence of an intruder is determined from the area and the center of gravity of each connected component.

The following procedure is known as a conventional general labeling process. In this procedure, first, each image is examined by raster scanning from the upper left corner to the lower right corner of the image, and a temporary label is assigned to the pixel having the pixel value of interest.
After this scanning is completed, the raster label is scanned from the lower right corner to the upper left corner to adjust the temporary label. Then, the temporary label is readjusted by further performing raster scanning from the upper right corner to the lower left corner.

[0006] For example, in the case of judging the connection by the four connectivity, in the first raster scan of the above procedure, when a pixel having a target pixel value is detected, the pixel is set as a labeling object,
A temporary label is assigned to the same label as the adjacent pixel above and to the left of the pixel. When neither the upper nor the left adjacent pixel is labeled, a new label is assigned to the pixel to be labeled. In the next raster scan from the lower right corner to the upper left corner, when a pixel to which a temporary label is added is detected, the label of that pixel is matched with the labels of the lower and right adjacent pixels. If neither the lower nor the right adjacent pixel is labeled, the original temporary label is used as the final label. Then, in the last scan from the upper right corner to the lower left corner, when a pixel to which a temporary label is added is detected, the label of the pixel is adjusted to match the label of the upper and right adjacent pixels.

In the first scan, pixels belonging to the same connected component are generally given different temporary labels. However, if the connected component has a relatively simple shape by the middle scan and the last scan, the pixel belonging to the same connected component will be replaced by a different one. They all have the same label.

[0008]

However, in the above-described conventional processing, correct labeling may not be performed if the shape of the connected component becomes complicated. For example, as shown in FIG. 20, a connected component 10 having a complex shape including a large number of branches.
0 indicates that the same label cannot be assigned in the above-described conventional processing. If correct labeling is performed on such a complicated shape, it is necessary to take measures such as further increasing the scanning direction, and the processing time becomes longer.

In the conventional labeling process, the same label is finally assigned to pixels belonging to one connected component by sequentially adjusting the temporary labels. At the stage where the adjustment is completed, the label given to each connected component (for example, a cluster of white pixels) is likely to have discrete values. For this reason, if it is desired to know the total number of connected components, it is necessary to perform a process of changing the label to a serial number, which requires a processing time.

In the conventional labeling process, the final label of each pixel is not determined until all raster scans have been completed. For this reason, the feature amount such as the area and the center of gravity of the connected component performed after the labeling cannot be performed until the latter half of the scan is completed. Therefore, the above-described conventional labeling process is not suitable for application fields that require high-speed object recognition.

The present invention has been made to solve such a problem, and an object of the present invention is to provide a method capable of relatively quickly processing an image having a complicated shape. It is a further object of the present invention to provide a method capable of quickly obtaining the characteristic amount of a connected component in a labeling process.

[0012]

In an image processing method according to a first aspect of the present invention, an image is examined in a predetermined scanning order, and a pixel that satisfies a label condition that a label is unassigned and has a target pixel value is detected. Each connected component of a pixel having a pixel value of interest in the image by repeating a start point detecting step to determine a labeling start pixel and a labeling step of labeling a labeling start pixel and a pixel having connectivity thereto. Perform labeling. Here, in the labeling step, a step (a) in which the labeling start pixel is set as the line start point, and the line start point is set as the first target pixel, and the target pixel is sequentially moved one pixel at a time in one of right and left directions. And adding the label to the target pixel if the target pixel satisfies the label condition, and adding the adjacent pixel to the stack if adjacent pixels above and below the target pixel satisfy the label condition. (B)
In step (b), when the target pixel no longer satisfies the label condition or reaches the end of the image, the target pixel is sequentially shifted one pixel at a time from the labeling start pixel in the direction opposite to the step (b). While moving, if the target pixel satisfies the label condition, the label is assigned to the target pixel, and if adjacent pixels above and below the target pixel satisfy the label condition, the adjacent pixel is added to the stack. (C) and when the target pixel no longer satisfies the label condition or reaches the end of the image in this step (c), the top pixel of the stack is taken out and set as a new line start point in the step (c). Steps (d) for performing b) and (c), and steps (e) for repeating the above step (d) until there are no more pixels from the stack.

In this method, labeling is performed by sequentially examining the connectivity of pixels in the line direction of an image, and at this time, pixels of adjacent lines connected to the labeled pixels are registered in a stack, and labeling of connected components of the line is performed. Is completed, the pixel is then taken out of the stack and the same procedure is repeated starting from that pixel. As described above, the connectivity on the line is examined one pixel at a time, and pixels having connectivity on the upper and lower sides of the pixel (that is, adjacent lines) are completely examined and stacked as candidates for the labeling start point. Pixels belonging to the block (connected component) can be completely detected and labeled with the same label. Therefore, even a connected component having a complicated shape including many branches can be accurately labeled. In addition, according to this method, labeling is performed sequentially for each connected component, so that when the labeling of the entire image is completed, the labels of the connected components are always serial numbers.

[0014] In the second invention, the labeling step includes:
Step (a) of setting a labeling start pixel as a line start point
With the line start point as the first target pixel, while moving the target pixel one pixel at a time in one of the right and left directions until the target pixel no longer satisfies the label condition or reaches the end of the image. Each time, the target pixel and its upper and lower neighboring pixels are examined, and if the neighboring pixel satisfies the label condition, the neighboring pixel is added to the stack.If the target pixel satisfies the label condition, the label is added to the target pixel. When the target pixel does not satisfy the label condition or reaches the edge of the image in step (b) and step (b), the target pixel no longer satisfies the label condition or reaches the edge of the image. Until the target pixel is reached, the target pixel is sequentially incremented by 1 in the direction opposite to the step (b) from the labeling start pixel.
Each time the pixel is moved, the target pixel is examined above and below adjacent pixels.If the adjacent pixel satisfies the label condition, the adjacent pixel is added to the stack.If the target pixel satisfies the label condition, the target pixel is added. And (c) when the target pixel no longer satisfies the label condition or reaches the end of the image in step (c). (D) performing steps (b) and (c) as described above, and (e) repeating step (d) until there are no more pixels from the stack.
And repeating the start point detecting step and the label assigning step until scanning is completed for all the pixels of the image.

In this method, when the adjacent pixels of the scanning target pixel are stacked, whether or not the adjacent pixel satisfies the label condition is determined regardless of whether the target pixel is labeled. Judgment was made. Thereby, the stacking can be correctly performed even in the case of the diagonal connection in the 8-connection, so that the labeling processing of the 8-connected component can be realized. Also, this method can realize 8-neighbor labeling with almost the same calculation amount as the method according to the first invention.

In each of the preferred embodiments of the present invention, by counting the number and coordinates of the pixels labeled in parallel with the labeling, it is possible to calculate the characteristic amount such as the area and the center of gravity of the pixel connected component of the same label. , High-speed feature amount calculation becomes possible.

In a preferred aspect of the present invention, it is determined whether or not a pixel taken out of the stack is unlabeled (determination of the label condition), and the number of times when the label has been labeled is counted. The number of holes (regions of pixels other than the pixel value of interest) in the connected component of the pixel value of interest can be obtained.

Further, the method according to the present invention examines an image in a predetermined scanning order, detects a pixel which satisfies a label condition that a label is unassigned and has a pixel value of interest, and detects a starting point to be determined as a labeling starting pixel. And, a labeling starting pixel and a labeling step of giving a label to a pixel having connectivity thereto, including a labeling step, the labeling step is a starting step with a labeling starting pixel as a line starting point, and the line starting point The pixels on the included line are scanned one pixel at a time, and the label is added each time a pixel that has connectivity to the line start point and satisfies the label condition is added. A line scan for storing, in a predetermined storage means, a pixel satisfying the label condition with respect to a pixel of the scan line as a connection candidate pixel. And when the labeling process on the scanning line is completed in the line scanning step, a scanning repetition step of retrieving one connected candidate pixel from the storage means and repeating the line scanning step using this as a new line start point. And

In this method as well, pixels satisfying the label condition in a line adjacent to the line to be scanned are added to the stack, and after the line scanning is completed, the scanning start point of the next line is extracted from the stack and used for labeling. Perform line scanning. By such processing, even a complicated figure can be labeled at a high speed by one scan (that is, without repeating the scan for re-assigning the temporary label as in the related art).

[0020]

Embodiments of the present invention (hereinafter referred to as embodiments) will be described below with reference to the drawings.

<First Embodiment> [Apparatus Configuration] FIG. 1 is a functional block diagram showing a schematic configuration of an image processing apparatus according to the present invention. In the present embodiment, 2
Labeling processing is performed on the binary image stored in the value image memory 20 by the labeling processing unit 10, and the result is written in the labeled image memory 30. The labeling processing unit 10 uses the stack 12 as control information storage means for labeling. This stack 12
Is a last-in first-out (LIFO) data structure. In the present embodiment, the labeling is performed line by line in the binary image. In the stack 12, the coordinate information of the starting pixel of the labeling process of the next line obtained at the time of labeling the line is registered. This stack 1
The detailed usage of 2 will be described in detail later in the description of the processing procedure. In parallel with the labeling processing, the labeling processing unit 10 executes the area counter 14 and the X counter 1
6 and the Y counter 18 are used to calculate the area and centroid of each labeled connected component (i.e., a cluster of connected pixels having the pixel value of interest).

[Overall Processing Procedure] Next, the overall processing procedure of the image processing apparatus of the present embodiment will be described with reference to FIG. The following processing is performed by the labeling processing unit 10.

In the labeling process of this embodiment, first, the labeling processing unit 10 initializes the value n of the label to be added to 0 (step S10). In this embodiment,
The serial numbers are used as labels.

When the label value n is initialized, image scanning is performed to find a block of pixels of the target pixel value to be labeled. In the following, it is assumed that “white” is a pixel value of interest and a block of white pixels is labeled from among the black and white binary values forming the binary image.

In this embodiment, this scanning is performed in the raster scanning order from the upper left corner to the lower right corner of the image. For this scanning, first, the labeling processing unit 10 initializes the value of the vertical coordinate y to 0 (Step S12). The vertical coordinate y is a number sequentially assigned from the top to the image line to be scanned. In the present embodiment, as shown in FIG. 8, as a coordinate system of an image, a horizontal coordinate x is set rightward in the horizontal direction, a vertical coordinate y is set downward in the vertical direction, and the x coordinate of the leftmost column (column) of the image is set to 0, The y coordinate of the top line (line) is set to 0. When the vertical coordinate y is initialized, scanning in the vertical direction is controlled in the processing loop from step S14 to S26. In step S16, the horizontal coordinate x is initialized to 0 for horizontal scanning. Scanning in the horizontal direction is controlled in the processing loop from steps S18 to S24. According to such scanning, the 2D image data stored in the binary image memory 20 is stored.
Each pixel of the value image is examined in raster scanning order.

In step S20, it is determined whether or not the point (x, y) selected as a survey target in the scan is an unlabeled white pixel. In the figure, the affirmative determination is TRUE (true),
A negative determination is indicated as FALSE (false). If the determination result is affirmative (TRUE), the value of the label n is incremented by 1 (step S22), and this point (x, y)
Is performed (step S100) to assign a label of a value n to a cluster of white pixels (connected component) starting from. The processing in step S100 will be described later in detail with reference to FIG. Then, in step S24, the value of the horizontal coordinate x is incremented by one, thereby moving the investigation target one position to the right and returning to step S18. On the other hand, when the determination result of step S20 is negative (FALSE),
The process proceeds to step S24 without performing any operation, moves the target pixel by one to the right, and returns to step S18.

The process proceeds as described above, and step S1
If it is determined in step 8 that the pixel to be examined has exceeded the right end of the image, the vertical coordinate y is incremented by 1 in step S26, the process returns to step S14, and the processes in steps S16 to S24 are repeated for the next line. Then, when it is detected in step S14 that the line to be investigated has exceeded the lower end of the image, a series of labeling processes has been completed.

According to the above-described processing procedure, each pixel is examined in the raster scanning order from the upper left corner to the lower right corner of the image, and when a white pixel without a label is detected, the connected component of the pixel is determined as a starting point. Labeling process (step S10
0) will be performed. 1 in step S100
After labeling of one connected component is completed, a search for a white pixel serving as a starting point of the unlabeled connected component is performed in the raster scanning order.

As can be seen from the above procedure, step S
The unlabeled white pixel found at 20 is the leftmost pixel of the top line in the connected component (cluster) of one white pixel. With this pixel as a starting point, a subroutine for labeling one connected component, ie, step S10
0 is executed. Next, the processing procedure of step S100 will be described.

[Labeling of Connected Components] FIG. 3 shows a detailed processing procedure of step S100. In this procedure, as a preparation, first, the coordinates (xi, xi, i) of the starting point of the connected component detected in step S20 from the main routine shown in FIG.
yi) and the label value n obtained in step S22 are acquired (S102). For example, the label value n of the start point (xi, yi) found first becomes 1 according to step S22. In the following procedure, white pixels connected to the pixel at the start point are sequentially detected, and labels n are added.

For this purpose, first, the coordinates (xi, yi) of the starting point are added to the control stack 12 (see FIG. 1) (step S104). In the present embodiment, the area and the center of gravity of the connected component of the white pixel are calculated in parallel with the labeling process, and the area counter 14 (area S), the X counter 16 (Gx), and the Y counter 18 (Gy) used for the calculation are used. Initialization is performed in step S106. Thereafter, while the pixel coordinate data is being entered in the stack, step S1 is executed.
The processing loop from 08 to S114 is repeated.

In this processing loop, first, the top entry (that is, the last stacked pixel coordinate) of the stack 12 is fetched and set as the scanning start coordinate (x0, y0) (step S110). Then, in step S112, it is determined whether or not the point (x0, y0) is a pixel having the unlabeled target pixel value (in this case, "white"). In the drawings and the following description, for the sake of simplicity of expression, the condition determination as to whether or not the pixel at the point (x, y) is the pixel of the target pixel value with no label is denoted as K (x, y). I do. If the point (x, y) is the pixel of the unlabeled target pixel value, the result is T
RUE (true), otherwise FALSE (false).

If the processing result of step S112 is TRUE, the pixels are scanned in the line direction starting from the scanning start coordinates (x0, y0) to perform labeling (step S200). According to this step S200, the label n is assigned to each pixel having connectivity to the pixel of the image line to which the pixel at the scanning start coordinates (x0, y0) belongs, and the adjacent lines above and below the line are scanned. A candidate for the starting coordinates is obtained, and the stack 1
2 will be added. At this time, the counters 14, 16 and 1 for obtaining the area and the center of gravity of the cluster of white pixels are used.
A count-up of 8 is also performed. This step S
The detailed processing procedure of 200 will be described later in detail. Upon completion of the process in the step S200, the process returns to the step S108, the latest added entry is taken out from the stack 12, and the above process is repeated.

Now, if the processing result of step S112 is FA
The reason for the LSE is that the pixel at the coordinates stacked on the stack 12 is extracted from the stack 12 (step S10).
8) This is the case where the label has been attached before. As will be described later with reference to a specific example, in simple terms, this occurs when there is a hole (black pixel) in a connected mass of white pixels. In this case, since the labeling of the line to which the coordinates belong has been completed, the process returns to step S108 without doing anything.

When all entries of the stack 12 are exhausted as a result of repeating such processing, the starting point (xi, yi)
Means that label n has been assigned to all pixels belonging to the connected component of white pixels starting from.

The count value of the area counter 14 (S) at this time is the area S of the connected component, and the X counter 16
And the value obtained by dividing the count value of the Y counter 18 by the area S thereof becomes the x coordinate (Gx) and the y coordinate (Gy) of the center of gravity of the connected component. In the present embodiment, step S
At 114, the barycentric coordinates (Gx,
Gy) is calculated, and is calculated together with the value of the area S of the connected component.
It is stored in a predetermined storage means (step S116).

When the above processing is completed, step S10
After the processing of 0 is completed, the control is returned to the main routine of FIG. 2, and the starting coordinates of the next block of white pixels (connected component) are searched.

[Line Scanning] Next, a detailed processing procedure of the line scanning process in step S200 will be described. In this process, as shown in FIG.
Processing for scanning the line to the right from the pixel at the scanning start coordinates (x0, y0) extracted from Step 2 to label connected pixels (Step S210: rightward processing)
And a process of scanning the line from the pixel to the left and labeling the connected pixel (step S250: leftward processing). Hereinafter, the contents of each of these processes will be described.

FIG. 5 shows the detailed procedure of the rightward processing (step S210). In this process, first, two flags F1
And the value of F2 is initialized (reset) to FALSE (step S212). These flags are set in step S
This function is used in the process of registering the start point of the adjacent line of 300 in the stack 12, and its function will be described in detail in the description of the process of step S300.

Next, in step S214, the x coordinate value x0 of the scanning start coordinate is set to the horizontal coordinate x. Thereafter, the horizontal coordinate x is incremented by one in step S224 until it is detected in step S216 that x has become greater than or equal to the width of the binary image (that is, the right end of the image has been reached). Repeat the processing loop.

In this processing loop, it is determined whether or not the point (x, y0) is the pixel of the target pixel value to which no label is added (K
(X, y0)) (step S218), and the result is TR
In the case of the UE, a label with a value n is assigned to the pixel at the point (x, y0) (step S220).

Since x = x0 at the first time when the processing loop is entered, the condition judgment K (x, y) is applied to the pixel at the scanning start coordinates (x0, y0) extracted from the stack 12 in step S220. However, since this pixel is the same as the one that has been determined to be TRUE in step S112, the processing result here is also TRUE.
The label n is assigned to the pixel at the scanning start coordinate. Thereafter, in this processing loop, the coordinates of the pixel to be inspected are sequentially shifted to the right on the line one pixel at a time (step S224), and whether the pixel to be inspected is an unlabeled white pixel is inspected in step S218. Is done.
If the result of step S218 is FALSE, the process exits this processing loop. Conversely, while this processing loop is being repeated, the pixel to be inspected is compared with the pixel at the scanning start coordinate by the label n with the label n on the line. The connection is established only through the pixels. Therefore, step S21
If the result of step 8 is TRUE (that is, the pixel to be inspected is unlabeled and a white pixel), it is assured that the pixel to be inspected has connectivity to the pixel at the scanning start coordinate. The label n may be assigned to the inspection target pixel.

For the line to which the labeling start point (xi, yi) set in step S102 belongs,
It will be understood from the above description that a pixel having connectivity to the start point on the line is given the same label n as the start point. Stack 1 for other lines
The process is started from the pixel of the coordinates stacked on the stack 2, and the start pixel is a pixel to which the label n of the adjacent line has been added by the method of the process of adding coordinates to the stack 12 (step S <b> 300) described later. Is guaranteed to have connectivity. Accordingly, the start pixel and the pixel having the connectivity on the same line may also be given the label n.
After extracting the start pixel from 2, the execution of the processing loop of FIG. 5 (steps S216 to S224) results in the label n being assigned to each of those pixels.

Now, in step S220, the pixel (x, y0) to be inspected is labeled, and in parallel with this,
Area counter 14 for determining the area of the connected component of the white pixel
The count value of (S) is incremented by one, and the horizontal coordinate x and the vertical coordinate y0 of the pixel labeled for obtaining the center of gravity of the connected component are calculated by the X counter 16.
(Gx) and the count value of the Y counter 18 (Gy), respectively (step S222).

Thereafter, processing for registering the scanning start point of the adjacent line in the stack 12 is performed (step S).
300).

Referring to FIG. 6, the processing procedure of step S300 will be described. In this process, it is determined whether adjacent pixels above and below the inspection target pixel (x, y0) are suitable for the scanning start point of the adjacent line, and if so, the coordinates of the adjacent pixels are added to the stack 12. In this process, steps S302 to S310 are processes related to the upper adjacent pixel, and steps S312 to S320 are processes related to the lower adjacent pixel. Either of these may be performed first.

The operation will be described in the order shown in FIG.
In 02, the judgment K (x, y) is applied to the adjacent pixel (x, y0-1) above the inspection target pixel (x, y0),
It is determined whether or not the adjacent pixel is an unlabeled white pixel. If the result of this determination is TRUE, the adjacent pixel has connectivity to the inspection target pixel. Here, since the label to be inspected itself is given the label n, if the adjacent pixel has connectivity to the pixel to be inspected, it can be said that the adjacent pixel is also a pixel to be given the label n. . In the present embodiment, basically, such a label n
Is registered in the stack 12 as the scanning start point of the adjacent line.

However, if all such adjacent pixels are registered in the stack 12, the size of the stack becomes large, and the load of the process of taking out pixels from the stack 12 increases. 12, pixels to be registered are narrowed down. The aforementioned flags F1 and F2 are used for this narrowing down. F1
Is for an upper adjacent pixel, and F2 is for a lower adjacent pixel.

If it is determined in step S302 that the adjacent pixel above the pixel to be inspected is a white pixel with no label, it is determined whether the current value of the flag F1 is FALSE (step S304). Only when F1 is FALSE, the coordinates (x, y0-1) of the adjacent pixel are added to the stack 12 (step S306). Then, the value of F1 is set to TRUE (step S308).

At step S212, the flag F1 is set to F
Since it has been initialized to ALSE, F1 is always FALSE when first passing this step 304 in the line scanning process (step S200). Therefore, if the adjacent pixel above the scan start pixel (x0, y0) of the line is a white pixel without label, the adjacent pixel is always registered in the stack 12. Thus, once the upper adjacent pixel is registered in the stack 12, step S
At 308, the flag F1 is set to TRUE and F1 is set to T
While RUE, the upper neighboring pixel (x, y0-1) is
Even a white pixel with no label is not registered in the stack 12 by the determination in step S304. Therefore, for example, after first registering the adjacent pixel on the scanning start pixel (x0, y0) in the stack 12, even if white pixels continue rightward on the same line from the adjacent pixel on top, the White pixels are not registered in the stack 12. The reason that the flag F1 is reset from TRUE to FALSE is that the upper adjacent pixel (x, y0−
1) is a case where it is determined that the pixel is not an unlabeled white pixel (that is, a black pixel or a labeled pixel) (step S310). After F1 is reset to FALSE, when an adjacent pixel (x, y0-1), which is a white pixel without labeling, is found again (step S302).
The adjacent pixel is added to the stack 12 (step S
306).

The stacking process for the adjacent pixel above the inspection target pixel has been described above, but the stacking process for the lower adjacent pixel is performed in exactly the same manner. That is, first, in step S312, it is determined whether the lower adjacent pixel (x, y0 + 1) is an unlabeled white pixel. If the determination result is TRUE, it is further determined in step S314 whether the flag F2 is FALSE. Only if F2 is FALSE,
The pixel (x, y0 + 1) is registered in the stack 12 (Step S316). When registration in the stack 12 is performed, the flag F2 is set to TRU in step S318.
Set to E. The flag F2 once set to TRUE is reset to FALSE when scanning proceeds and the lower adjacent pixel (x, y0 + 1) is no longer a white pixel with no label in step S312 (step S32).
0).

The registration processing of the pixel coordinates in the stack 12 has been described with reference to FIG. Although not shown in FIG. 6, an error avoidance process for preventing a value outside the binary image from being erroneously read when referring to the pixel on the upper or lower side of the inspection target pixel in step S302 or S312. Needless to say that we are doing.

In this way, the inspection target pixel (x, y
When the stacking process for the upper and lower adjacent pixels in (0) is completed, the horizontal coordinate x of the inspection target pixel is incremented by 1 in step S224, and the inspection target pixel is shifted right by one, and the processing of steps S216 to S224 is repeated.
In this way, the inspection target pixel exceeds the left end of the binary image (step S216), or the inspection target pixel is no longer a pixel to which the label n should be added (step S21).
If the result of step 8 is FALSE, the rightward processing (step S210) is completed and the leftward processing (step S250)
Move on to

The leftward processing (step S250) is the same as that shown in FIG.
Is performed according to the procedure shown in FIG. This processing is exactly the same as the above-described rightward processing except that the processing is started from the pixel on the left of the scanning start coordinate (x0, y0) and the line is scanned leftward. I do.

In this process, first, in step S252,
The flag F1 is initialized to the determination result of K (x0, y0-1), and the flag F2 is initialized to the determination result of K (x0, y0 + 1). Here, the flag is initialized in accordance with the pixels on the upper and lower sides of the scanning start coordinates (x0, y0). Next, in step S254, the horizontal coordinate x is initialized to x0-1 so that the pixel on the left of the scan start coordinate is selected as the first inspection target pixel. Thereafter, until it is detected in step S256 that the horizontal coordinate x has become smaller than 0, the horizontal coordinate is sequentially decremented by 1 in step S264.
The labeling and stacking process (steps S218 to S218) is performed in the same manner as in the case of the rightward processing, while shifting to the left one by one.
300) is performed.

The leftward processing (step S2)
When step 50) is completed, the labeling process for one line (step S200) is completed, and the process returns to the flow of FIG. 3 to retrieve the next scan start coordinate from the stack 12 and repeat the same process.

As described above, the labeling processing unit 10 of the present embodiment
Has been described. In the above procedure, the point (x, y)
The condition determination K (x, y) that the upper pixel is unlabeled and the pixel value of interest was performed at various places. This processing is performed by performing the following initialization processing on the labeled image memory 30. It can be simplified by executing. In this initialization processing, the value of each pixel is read from the binary image memory 20 and
Depending on whether or not the pixel is a target pixel value, different temporary labels are given and written to the labeled image memory 30. As the temporary label used at this time, a value not used for final labeling (for example, 0 or -1 when labeling with a natural number serial number) is used. As a preferred example, a temporary label of “−1” is given to a pixel having a target pixel value, and a temporary label of “0” is given to a pixel having no target pixel value. By doing so, K in step S20 in FIG. 2, step S112 in FIG. 3, step S218 in FIG.
The condition determination of (x, y) is a determination as to whether or not the label value of that point (x, y) is “−1”. That is,
If the point (x, y) is “−1”, it is because the label has not been assigned and is the pixel value of interest (the label value is “0” if it is not the pixel value of interest, and the label has already been assigned). The label value is not "-1".

[Processing Specific Example] Next, referring to a specific example of a binary image, how the processing procedure is executed will be described.

As a specific example of the binary image, consider the image shown in FIG. In FIG. 8, a black square indicates a black pixel, a white square indicates a white pixel, and a white square including a numeral indicates a labeled white pixel. The number inside is the value of the label. The hatched rectangle indicates a white pixel whose coordinates are registered in the stack 12.

In the state shown in FIG. 8, labeling is completed for the connected component (lumps of white pixels) of label 1, and step S20 is performed.
Shows the state when the next labeling start point is detected and the start point is registered in the stack 12 in step S104. That is, coordinates (12, 2) are stacked on the stack, and thereafter, starting from the pixel at this coordinate, the labeling of the pixel connected to this pixel is performed.

In this case, first, the coordinates (1
2, 2) are taken out (step S110), and since these are unlabeled white pixels, a new label 2 is given (step S220). Of the upper and lower neighbors of this pixel, the latter is an unlabeled white pixel, so its coordinates (1
2, 3) are stacked on the stack 12 (step S31).
6). When the pixel on the left of the labeling start point (12, 2) is examined, since the pixel (13, 2) is a white pixel without labeling, the same label 2 as the labeling start point is added (step S220). The pixel (13,
3) is a white pixel with no label (step S3).
02), since the flag F2 is set to TRUE when the coordinates (12, 3) are stacked on the stack 12 earlier, they are not registered in the stack 12. FIG. 9 shows the situation at this point.

Further processing is performed, and when the scanning for the line y = 2 is completed, the state of labeling is as shown in FIG.
It becomes as shown in. Since the scanning of the line has been completed, the next entry from the stack 12, that is, the coordinate (1
2, 3) are taken out (step S110), and scanning is performed on the line of y = 3 using the coordinates as a starting point. Here, first, the label 2 is added to the start point (12, 3) of the line. The lower neighbors (12, 4) are unlabeled white pixels and are stacked on the stack 12. Line start point (12,
When the process proceeds to the right from 3), when the pixel to be inspected reaches (15, 3), the pixel immediately below the pixel (15, 3)
4) becomes a black pixel, and the flag F2 is reset to FALSE (S320). Thereafter, the pixel to be inspected is (1
When the pixel reaches (9, 3), the lower pixel (19, 4) is an unlabeled white pixel, and since the flag F2 is FALSE at this point, the coordinates (19, 4) are stacked on the stack 12. (Step S316). The labeling state at this time is shown in FIG.

Thereafter, the processing is advanced, and when the scanning of the line of y = 3 is completed, the labeling state becomes as shown in FIG. At this point, the coordinates (19, 4) and (12, 4) are stacked on the stack 12 in order from the top. This is because, below y = 4, the connected component of the white pixel is
9 and 4) and a portion connected to the pixel (12, 4).

Next, the coordinates (19, 4) stacked later are taken out of the stack 12, and scanning (labeling) is performed using the coordinates as the line start point. Thereafter, by repeating the above-described processing, the label 2 is assigned to the white pixel connected to the pixel (19, 4) among the two branched connected components. FIG. 13 shows a labeling state at the time when the processing is completed up to the line of y = 7.

Next, the coordinates (20, 8) are taken out of the stack 12, and labeling scanning of the line of y = 8 is performed with this point as a starting point. By this scanning, new coordinates (20, 9), (15, 7) and (9, 9) are stacked on the stack 12 in this order. FIG. 14 shows a labeling state at the time when scanning of the line of y = 8 is completed.

Next, coordinates (9, 9) are extracted from the stack 12, and labeling is performed downward from this point. When labeling is completed for all the white pixels of the connected component below (9, 9), the coordinates (15, 7) are extracted from the stack 12 and labeling is performed upward from this point. Go. FIG. 15 shows a labeling state at the time when the upward labeling scan is completed up to the line of y = 6.

FIG. 16 shows a state at the time when the labeling is completed up to one more line. At this point, the coordinates (13, 4) are stacked at the top of the stack 12.
Therefore, next, starting from this point (13, 4), y
Labeling is performed on the line of = 4. FIG.
Shows the result of this processing. By this processing, the label 2 is assigned to the pixel at the coordinates (12, 4) stacked at the bottom of the stack 12. In FIG. 17, this is indicated by a square of a white label. In this process, no new coordinates are stacked on the stack 12.

Next, the coordinates (20,
9) is taken out, and using this as a starting point, white pixels connected to this starting point are labeled downward. When labeling is completed for all the connected components in the downward direction from the start point, the state shown in FIG. 18 is obtained. In this state, only the point at the coordinates (12, 4) remains in the stack 12. Finally, this point is taken out of the stack 12 (step S110). Since this point has already been labeled, the determination result of step S112 is FALSE, and the line scanning processing step S200 is not performed.
Therefore, no entry is added to the stack 12, and as a result, the stack 12 becomes empty as a result. Thus, the end condition of the loop A is satisfied in step S118, and the labeling of the label 2 starting from the labeling start point (12, 2) is completed.

As can be clearly understood from the above specific example, when a set of white pixels connected to each other on each line is considered as a node, a cluster of white pixels having the same label is formed by connecting the node to the adjacent lines. Can be thought of as a tree structure connected by. It can be seen that the labeling method of the present embodiment is equivalent to a process of searching the multitree according to the depth-first search algorithm. In the method of the present embodiment, since the starting points (scanning start coordinates of lines) of all branches of the multi-tree structure are reliably stacked on the stack 12, it is possible to obtain and label four connected components without fail.

In this embodiment, only one pixel (scanning start coordinate) is selected and registered in the stack 12 for a set of white pixels connected to each other on each line. Therefore, when examining a pixel taken out of the stack 12 and examining it, if it is already labeled, the pixel is stacked from a certain direction (for example, from the top) as a candidate of a starting point of the branch of the white pixel block. Is
It indicates that the label has already been labeled by the approach from the opposite direction, that is, there is one hole (black pixel area) in the mass of white pixels. Therefore, by counting the number of times that the pixels taken out of the stack 12 have been labeled, the result can be used to determine the number of holes in the labeled mass of white pixels. Topological information such as the number of holes is useful information for object recognition.

[Effects] The labeling processing according to the present embodiment has been described above. In the present embodiment, the connectivity is examined one pixel at a time along the line, and pixels having connectivity above and below the pixel are stacked as labeling start point candidates (scanning start coordinates). Even if the connected component is complicatedly branched upward or downward from the line, any pixel of each branch part is always stacked on the stack. Therefore, if the process is repeated until there are no more entries in the stack, the same label is always assigned to all branch portions, and the same label can be reliably assigned to each pixel even if the connected component has a complicated shape.

Further, in the prior art, it was necessary to perform scanning in various directions in order to cope with a connected component having a complicated shape. Therefore, a long time was required for processing. Can also perform the labeling process at high speed.

For example, the inventor has set a 20 × 25 shown in FIG.
In an experiment performed using a binary image of pixels, the processing time for labeling white pixel connected components between a program based on the conventional method and a program based on the method according to the present embodiment is about 4: 1 (that is, the processing time according to the present embodiment). Is about four times the speed of the prior art). This experiment was performed using a general personal computer.

In the prior art, the calculation of the feature amount had to be waited until the labeling process for the entire image was completed. On the other hand, in the present embodiment, the labeling of one white pixel block (connected component) was performed. Since the feature amount such as the area and the center of gravity of the block can be obtained at the stage when the calculation is completed, it can be said that the effect of speeding up is even higher if the calculation up to the feature amount is considered comprehensively. In particular, in applications where it is only necessary to find one that satisfies a specific condition from an image, such as intruder detection, the effect of speeding up is even more remarkable.

In the above example, the target pixel value to be labeled is white, but the present invention is naturally applicable to a case where the target pixel value is black.

[Intruder Detecting Apparatus] Finally, an application example of the image processing of the present embodiment to an intruder detecting apparatus will be described with reference to FIG. 19, the same components as those in FIG. 1 are denoted by the same reference numerals.

In this intruder detection device, a monitoring CCD
The NTSC composite signal from the camera 100 is received by the image input unit 40, and the image obtained by the image input unit 40 is transmitted to the first original image memory 42 and the second original image memory 44.
Writing is performed alternately for each frame. Difference calculation unit 46
Then, the first and second original image memories 4 are stored for each frame.
The difference for each pixel of 2 and 44 is calculated, and the result is written to the difference image memory 48. The difference image between a certain time and the time one frame before the time formed in the difference image memory 48 is binarized by a binarization processing unit 50 based on a predetermined threshold, and the result is binarized. The data is written to the value image memory 20. The labeling processing unit 10 stores the binary image memory 2
Labeling and feature value calculation processing are performed on the binary image within 0 according to the above-described processing procedure, and the result is written to the labeled image memory 30. In this figure, illustration of a control stack and various counters used for the labeling process is omitted. The intruder determination unit 52 receives the labeling result and the calculation result of the feature amount, and determines the presence or absence of an intruder from these data according to a predetermined criterion. This determination result is transmitted to the control unit 54 that controls the entire intruder determination device. For example, when it is determined that there is an intruder, the control unit 54 activates a predetermined alarm device.

<Embodiment 2> [Overview] Next, another embodiment of the image processing according to the present invention will be described. In contrast to the first embodiment in which four connected components are labeled, this embodiment will describe a configuration for eight connected components.

The configuration of the apparatus for realizing this embodiment may be the same as that of the first embodiment shown in FIG. 1, and the basic processing procedure is almost the same as that of the first embodiment. However, in the present embodiment, in order to enable labeling of eight connected components, the method of stacking the starting points of adjacent lines is slightly changed from the method of the first embodiment.

In the first embodiment, when the next pixel is examined and labeled one by one from the start point of the line, after the labeling is performed on the pixel to be examined, the stack processing is performed on the adjacent pixels above and below the pixel. (For example, FIG. 5
S218 to S300). In this method, unless the target pixel is labeled, adjacent pixels above and below the target pixel are not stacked on the stack 12 (FIG. 1) as candidates for a labeling start point. In this method, since the pixels stacked on the stack as the scanning start point of the adjacent line are limited to the upper and lower adjacent pixels of the labeled pixel, the connectivity with the four adjacent pixels of the target pixel is determined. It conforms to the method of connected components. For example, image 2 as shown in FIG.
In the case of 00, according to the method of the first embodiment, the cluster of white pixels 210
And the chunk 220 are determined to be unconnected (in terms of 4-connectivity), and another label is given. For example, when scanning the pixels from left to right and from top to bottom to check the connectivity, when scanning in the right direction from (7, 6) on the sixth line, (9, Since the pixel 6) is a black pixel, the label condition is not satisfied, and thus no labeling is performed. Therefore, although (9, 7) below it is a white pixel, it is not added to the stack 12. Therefore, the labeling scan cannot continue to line 7 and the block 220
Will be given a different label than the chunk 210.

On the other hand, in the second embodiment in which the eight connected components are considered, in order to give the same label to the blocks 210 and 220 in the case shown in FIG. The algorithm is changed so that (9, 7) is stacked on the stack 12. That is, in the present embodiment, unlike the first embodiment, regardless of whether or not a label is given to the target pixel of the labeling scan, if the adjacent pixels above and below the target pixel satisfy the label condition, they are added to the stack 12. To do it. For example, in the example of FIG. 22, in the scanning of the sixth line, the target pixel is (9, 6).
Is satisfied, the lower adjacent pixel (9, 7) satisfies the label condition (unlabeled and white pixel), so that pixel is stacked on the stack 12 (whether the target pixel satisfies the label condition or not). Is not considered). In the line scanning, connectivity is guaranteed up to the pixel (86) immediately before the target pixel (9, 6), and this pixel (8, 6) and the pixel (9, 7) stacked on the stack have eight connectivity. There is. Therefore, if the labeling scan of the adjacent line is restarted from the pixels (9, 7) stacked on the stack 12, labeling satisfying the 8 connectivity can be performed.

[Processing Procedure] Next, a labeling procedure of the present embodiment will be described with reference to a flowchart. In this procedure, besides changing the stack processing as described above, the label condition is easily determined by performing an initialization process on the labeled image memory 30.

FIG. 23 is a flowchart of the overall processing procedure. In this procedure, first, a binary image to be labeled is prepared in the binary image memory 20 (S50).
The labeled image memory 30 is initialized according to the binary image (S500). FIG. 24 shows the details of this initialization processing.

In the initialization processing of the labeled image memory 30, first, a binary image is prepared (S502), and the coordinates x and y of the coordinates indicating the scanning position are initialized to 0 (S504). Thereafter, in the processing loop from step S506 to S518, scanning in the vertical direction is controlled,
In the processing loop from step S508 to S516, scanning in the horizontal direction is controlled. According to this raster scanning, the value B (x, y) of each pixel (x, y) of the binary image
Are checked (S510).

If B (x, y) = 0 (that is, black) in S510, the value L (x, y) of the pixel (x, y) in the labeled image memory 30 is set to “0” ( S5
12), if B (x, y) = 0 (that is, a black pixel), the label value L (x, y) of the pixel is set to “−”.
It is set to 1 "(S514).

That is, in this example, a label of a natural number (1, 2, 3,...) Is given to a cluster of white pixels. And a temporary label value “−1” is given to each white pixel. This temporary label value "-1" is later replaced with a formal label value. Conversely, in the subsequent labeling process, if a pixel with a label value of “−1” is found, it is immediately known that the pixel satisfies the label condition (white pixel and no label attached). It is easy to determine whether a pixel satisfies the label condition.

In this way, the labeled image memory 30
Is completed, the label value n is initialized to 0 (S
52), the coordinate index x and y of the scanning point are each set to 0
(S54), and the labeled image memory 3 is initialized.
0 is raster-scanned (S58 to S64 in the x direction, S56 to S66 in the y direction). This scan is to find the first point of a new unlabeled cluster of white pixels,
In S60, the value L of the pixel to be investigated (x, y) in the memory 30
(X, y) is checked, and if the value is -1, it means that an unlabeled white pixel has been found, and the label value n
Is incremented (S62), and the connected component is labeled with the pixel as a start point (S600). Each time the labeling of one block (S600) is completed, S56 to S
The processing loop of 66 is repeated, and the process ends when scanning is completed up to the last pixel of the image.

The detailed procedure of the labeling process (S600) is almost the same as the procedure of the first embodiment shown in FIG.
However, in the first embodiment, the condition determination K (x0, y0) is performed to determine whether or not the pixel (x0, y0) extracted from the stack 12 in S112 satisfies the label condition. , The label value L (x
(0, y0) is "-1". If L = -1, it means that the label condition is satisfied, so the process proceeds to the line scanning process (S200).
The next pixel is taken out from 2 and the process is repeated.

Now, in the line scanning process of S200, the rightward process (S210) and the leftward process (S250) are sequentially performed as in the first embodiment (see FIG. 4). Detailed procedures of rightward processing and leftward processing in the present embodiment are shown in FIGS. 25 and 26, respectively. In these figures, FIG. 5, FIG.
Steps for performing the same processing as in the above are denoted by the same reference numerals, and description thereof is omitted.

Rightward processing in this embodiment (see FIG. 25)
Before judging whether or not the target pixel (x, y0) satisfies the line condition (S232), the adjacent line start point stacking process (S3) is performed on the adjacent pixels above and below the target pixel.
00) is different from the first embodiment (first embodiment).
Then, only when the target pixel satisfies the line condition, the stack processing (S300) is performed on the upper and lower adjacent pixels). If it is determined in S232 that the target pixel satisfies the line condition, the target pixel (x,
The label value L (x, y0) of (y0) is set to n (S23).
4) Perform processing for calculating the area and the center of gravity (S22)
4).

The stack processing of this embodiment (S30
The processing content of (0) is to determine whether the upper and lower adjacent pixels satisfy the line condition (S302, S312), and to determine whether the pixel values L (x, y0-1) and L (x, y0 + 1) of those pixels are " -1 "
The process may be the same as the process of the first embodiment shown in FIG.

In the leftward process (see FIG. 26), first, a flag adjustment process (S270) is performed. Details of this flag adjustment processing are shown in FIG. That is, in the flag adjustment processing, first, it is determined whether or not y0 (the number of the line being scanned) is greater than 0 (that is, whether or not there is an adjacent line above) (S700). The flag F1 for determining the start point pixel of the upper adjacent line is set to L (x
0, y0-1) is set. This value L (x0, y0-1) is
This is the label value of the pixel immediately above the start pixel (x0, y0) of the line currently being scanned. However, the flag F1 is 2 of 0 and 1
Since it is a value, F1 = 1 in cases other than L = 0. By this S702, F1 = 0 (FALSE, that is, reset state) only when the pixel immediately above the line start pixel is L = 0, and in other cases, F1 = 1 (TRU
E).

When L (x0, y0-1) = 0, it means that the pixel (x0, y0-1) above the line start point does not satisfy the label condition. A pixel that satisfies the label condition of the line adjacent to the line above is disconnected at least at (x0, y0-1). Therefore, by resetting F1 to FALSE, it is possible to stack the label condition satisfying pixels of the upper adjacent line, which are found first in the subsequent leftward line scanning, on the stack.

On the other hand, when L (x0, y0-1) is not 0, the following two cases can be considered. The first case is L
(X0, y0-1) =-1. In this case, since the pixel (x0, y0-1) must already be in the stack at the time of the rightward processing (FIG. 25), F1 is TRUE. There is no problem.

The second case is a case where L (x0, y0-1) is n (natural number), that is, a pixel (x0, y0-1) has already been labeled, and this value n is Has the same value as the label of the starting pixel (x0, y0). in this case,
Since the connection of the unlabeled pixel is disconnected in the upper adjacent line, the flag F1 should be reset to FALSE normally, but is set to TRUE in S702. But this is no problem. Because the label of the pixel (x0, y0-1) is n (natural number), the label L (x0-1,
This is because y0-1) should be set to n for a white pixel and to 0 for a black pixel. Therefore, even if F1 is set to TRUE in S702 in the second case, F1 is reset to FALSE when the next pixel is processed and the stack processing (see FIG. 6) is performed. Absent.

Next, the flag is adjusted similarly for the flag F2. That is, first, it is confirmed that the line being scanned is not the last line of the image (S704). If it is confirmed that the line is not the last line, the flag F2 is set to F2 = L (x0, y0) in the same manner as the aforementioned F1. +1). S70
In the determination of 4, if it is determined that the line being scanned is the last line, the flag F2 is not required, and no adjustment is performed.

As described above, in the present embodiment, the flag adjustment at the time of shifting from rightward processing to leftward processing is realized by simple processing.

When the flag adjustment is completed in this manner, labeling processing is performed leftward from the start point in the same manner as in the above-described rightward processing (S254 to S264). This leftward processing is also similar to the rightward processing in that the target pixel (x, y0)
It is determined whether or not satisfies the line condition (S232).
First, the adjacent line start point stacking process (S300) is performed on the adjacent pixels above and below the target pixel. Subsequent processing may be basically the same as the rightward processing.

Thus, the leftward processing (step S2)
When step 50) is completed, the labeling process for one line (step S600) is completed, and the process returns to the flow of FIG. 23, the next scanning start coordinate is extracted from the stack 12, and the same process is repeated. Thus, the first embodiment
In the same way as in the case of, the labeling process in consideration of the 8 connectivity can be performed. Also in the present embodiment, similarly to the first embodiment, the area S and the center of gravity of the cluster of white pixels, the number of holes in the cluster of white pixels, and the like can be obtained in parallel with the labeling process. If the maximum and minimum values of the x and y coordinates of the scanned pixel are stored during the scanning for labeling, the width and height of the white pixel block are obtained in parallel with the labeling process. It is also possible.

As described above, in this embodiment, when stacking the scanning start points of the adjacent lines, regardless of whether or not the target pixels in the line scanning are labeled, the adjacent pixels above and below the target line are The determination as to whether or not stacking is made only based on whether or not the label condition is satisfied is made. Thereby, even in the case of the diagonal connection in the eight connection (such as the relationship between (8, 6) and (9, 7) in FIG. 22), stacking can be performed correctly, so that the labeling processing of the eight connected components can be realized. Can be.

Further, since the processing procedure of the present embodiment is almost the same as the processing procedure of the first embodiment, the amount of calculation is almost the same as that of the first embodiment. In this embodiment,
Compared with the case of the four-neighbor labeling of the first embodiment, for each labeling of one line, only four pixels diagonally adjacent to both ends of the line are additionally inspected. Almost negligible.

That is, in the 4-neighbor labeling of the first embodiment, when labeling one line of length L (pixel), the number of pixels for which the label condition is determined is 3L + 2 (the target line and the lines immediately above and below the target line). L × 3, plus two pixels on the left and right sides of the target line). On the other hand, in the eight-neighbor labeling according to the present embodiment, the number of pixels for which the label condition is determined for one same line is 3L + 6. The worst case is when black and white are in a checkered pattern one pixel at a time, and in this case L = 1, so the number of inspection pixels per line for 8-neighbor labeling in the present embodiment is 1.8 times of However, in practical applications (eg, intruder detection, etc.), the image does not become such an extreme and generally has a sufficient line length, so that the number of pixels is increased by four pixels as compared with four neighbor labeling. Even so, it is negligible (eg, only 2.6% increase for L = 50). Therefore, the method of the present embodiment has an advantage that 8-neighbor labeling can be realized with substantially the same calculation amount as that of the first embodiment in which 4-neighbor labeling is performed.

The method of the present embodiment described above can be applied to various application fields such as an intruder detection device as in the case of the first embodiment.

[Effect Verification] The effects of the present embodiment were verified using actual examples. The images used are shown in FIGS.
(C) three images, and (a)
1) Four-neighbor labeling according to the conventional method (described in the prior art), (a2) eight-neighbor labeling according to the conventional method (the method shown in the prior art is expanded to take into account eight neighbors), (a3) Embodiment 1 4-neighbor labeling by
(A4) Labeling was performed by four methods of eight neighborhood labeling according to the second embodiment, and the processing time was measured. The processing is performed on the Pentium 2 (registered trademark) 2 as a CPU.
The measurement was performed on a personal computer equipped with 66 MHz. The processing time at that time is shown in FIG.

Since the image (a) is a simple image, the final label is determined by updating the temporary label once or twice in the conventional method. Nevertheless, it can be seen that the method (a3) shown in the first embodiment is four times or more faster than the conventional method (a1). In addition, it can be seen that the method (a4) of the second embodiment is at least four times faster than the conventional method (a2), and is almost the same speed as the first embodiment (a3).

Image (b) is a rather complicated image.
In applications such as intruder surveillance, labeling images of this complexity is not uncommon, and images of this complexity are a category of practical images. At such a level of complexity, the conventional methods (a1) and (a2) require re-assignment of temporary labels many times to determine the final label, so that the processing time becomes extremely long. On the other hand, the methods (a3) and (a4) of the first and second embodiments
Shows that the processing is completed in the same time as the case of the simple image (a), which clearly shows the high speed of the method according to the present invention. Also in this example, the processing time according to the second embodiment is comparable to that of the first embodiment.

The image (c) is a very complicated image, and it is unlikely to handle such a complicated image in an actual application field. However, in order to examine the effectiveness of the method according to the present invention, such a complicated image was also verified. In the conventional methods (a1) and (a2), it is necessary to repeat the re-attachment of the temporary label over and over again, so that a very long processing time is required. Moreover,
In the conventional 4-neighbor processing, since the number of pixels for which connection is determined is small, the speed at which a valid label propagates through the block by re-labeling becomes slow. Is getting faster. However,
In any case, the methods (a3) and (a4) according to the present invention
In each case, the processing is completed in the same processing time as a simple image. Also in this example, it is understood that the neighborhood of eight can be labeled in the second embodiment in the same processing time as the first embodiment.

As described above, according to the present embodiment, 8-neighbor labeling can be realized at a processing speed much higher than that of the prior art and at a speed almost equal to that of the first embodiment in which four-neighbor labeling is performed. Can be. In addition, the time required for the processing is not so affected by the complexity of the image, and even a somewhat complicated image can be processed at a high speed. Thereby, the real-time property of the image processing can be further enhanced. It is considered that such a margin of time caused by the speeding up of the labeling makes it possible to devote time to other processing which has been postponed from the viewpoint of the processing time.

[Brief description of the drawings]

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

FIG. 2 is a flowchart illustrating an overall processing procedure of a labeling processing unit according to the first embodiment.

FIG. 3 is a flowchart illustrating a procedure of a process of labeling one lump (connected component).

FIG. 4 is a flowchart showing the procedure of one-line scanning in the procedure of FIG. 3;

FIG. 5 is a flowchart showing a procedure of a scanning process in a rightward direction from scanning start coordinates in one-line scanning.

FIG. 6 is a flowchart illustrating a procedure of a process of registering a scan start coordinate of an adjacent line in a stack.

FIG. 7 is a flowchart illustrating a procedure of a scanning process in a leftward direction from scanning start coordinates in one-line scanning.

FIG. 8 is a diagram illustrating a processing state at a certain point in time when the labeling processing of the embodiment is applied to a specific binary image.

FIG. 9 is a diagram illustrating a processing state at a certain point in time when the labeling processing of the embodiment is applied to a specific binary image.

FIG. 10 is a diagram showing a processing state at a certain point in time when the labeling processing of the embodiment is applied to a specific binary image.

FIG. 11 is a diagram showing a processing state at a certain point in time when the labeling processing of the embodiment is applied to a specific binary image.

FIG. 12 is a diagram illustrating a processing state at a certain point in time when the labeling processing of the embodiment is applied to a specific binary image.

FIG. 13 is a diagram illustrating a processing state at a certain point in time when the labeling processing of the embodiment is applied to a specific binary image.

FIG. 14 is a diagram illustrating a processing state at a certain point in time when the labeling processing of the embodiment is applied to a specific binary image.

FIG. 15 is a diagram showing a processing state at a certain point in time when the labeling processing of the embodiment is applied to a specific binary image.

FIG. 16 is a diagram illustrating a processing state at a certain point in time when the labeling processing of the embodiment is applied to a specific binary image.

FIG. 17 is a diagram illustrating a processing state at a certain point in time when the labeling processing of the embodiment is applied to a specific binary image.

FIG. 18 is a diagram illustrating a processing state at a certain point in time when the labeling processing of the embodiment is applied to a specific binary image.

FIG. 19 is a diagram showing a configuration of an intruder detection device to which the present invention is applied.

FIG. 20 is a diagram illustrating an example of an image including a connected component that is complicatedly branched.

FIG. 21 is a diagram showing images used in a comparison experiment of processing time between the conventional method and the method of the embodiment.

FIG. 22 is a diagram illustrating an example of an image having 8-neighbor connectivity.

FIG. 23 is a flowchart illustrating an overall processing procedure of labeling according to the second embodiment.

FIG. 24 is a flowchart illustrating a procedure of an initialization process of a memory for a labeled image.

FIG. 25 is a flowchart illustrating a procedure of a rightward scanning process according to the second embodiment.

FIG. 26 is a flowchart illustrating a procedure of a leftward scanning process according to the second embodiment.

FIG. 27 is a flowchart illustrating a procedure of a flag adjustment process.

FIG. 28 is a diagram illustrating an example of an image used for verifying the effect of the method of each embodiment.

29 is a diagram illustrating measurement results of labeling processing times for the image of FIG. 28 according to the conventional method and the method of each embodiment.

[Explanation of symbols]

10 labeling processing unit, 12 stacks, 14 area counter, 16 X counter, 18 Y counter, 20
Binary image memory, 30 Labeled image memory.

Claims (7)

[Claims] 1. A starting point detecting step of examining an image in a predetermined scanning order, detecting a pixel that satisfies a label condition that a label is unassigned and has a pixel value of interest, and that determines a labeling start pixel; And a label assigning step of assigning a label to a pixel having connectivity thereto, wherein the label assigning step comprises: (a) a step of setting a labeling start pixel as a line start point; When the target pixel satisfies the label condition while sequentially moving the target pixel one pixel at a time in a predetermined one of right and left directions, the label is assigned to the target pixel, and the target pixel is Adding adjacent pixels to the stack when adjacent pixels above and below satisfy the label condition; and (c) in step (b), when the target pixel is When the label condition is not satisfied or the end of the image is reached, the step (b) is performed from the labeling start pixel.
When the target pixel satisfies the label condition while sequentially moving the target pixel one pixel at a time in the reverse direction, the label is given to the target pixel, and adjacent pixels above and below the target pixel are set to the label condition. If the target pixel no longer satisfies the label condition or reaches the end of the image in step (c), the pixel at the top of the stack is added. And performing steps (b) and (c) as a new line start point, and (e) repeating step (d) until there are no more pixels from the stack. An image processing method in which the start point detecting step and the label assigning step are repeated until scanning is completed. 2. A starting point detecting step for examining an image in a predetermined scanning order, detecting a pixel which satisfies a label condition that a label is unassigned and has a target pixel value, and determines a starting point pixel for labeling; And a label assigning step of assigning a label to a pixel having connectivity thereto, wherein the label assigning step comprises: (a) a step of setting a labeling start pixel as a line start point; Until the target pixel no longer satisfies the label condition or reaches the edge of the image, the target pixel is sequentially moved one pixel at a time in one of right and left directions, and the target pixel and its upper and lower portions Is checked, and if the adjacent pixel satisfies the label condition, the adjacent pixel is added to the stack, and the target pixel satisfies the label condition. Assigning the label to the target pixel, and (c) when the target pixel no longer satisfies the label condition or reaches the end of the image in step (b), the target pixel satisfies the label condition. Until the pixel disappears or reaches the end of the image, the target pixel is sequentially moved one pixel at a time from the labeling start pixel in the direction opposite to the step (b), and the adjacent pixel above and below the target pixel is checked each time. Adding a label to the stack if the label condition is satisfied, and adding the label to the target pixel if the target pixel satisfies the label condition; Satisfies the label condition or reaches the end of the image, fetches the top pixel of the stack and sets it as a new line start point in steps (b) and (c) And (e) repeating the above step (d) until there are no more pixels in the stack, wherein the starting point detecting step and the labeling step are repeated until scanning is completed for all the pixels of the image. Method. 3. The steps (b) and (c) are set when an adjacent pixel above the target pixel is added to the stack, and reset when the adjacent pixel above the target pixel no longer satisfies the label condition. And a flag that is set when an adjacent pixel below the target pixel is added to the stack, and is reset when the adjacent pixel below the target pixel satisfies the label condition. 3. The image processing method according to claim 1, wherein the addition of the pixel to the stack is performed only when the corresponding flag is reset. 4. An area counter is counted up when a label is given to a target pixel in the steps (b) and (c), and the area counter at the point when no pixels are left from the stack in the step (e). 4. The image processing method according to claim 3, wherein the count value is output as an area of a pixel group having connectivity to the labeling start pixel. 5. When assigning a label to a target pixel in the steps (b) and (c), the x and y coordinates of the target pixel are added to an x counter and a y counter, respectively. By dividing the count value of the x counter and the count value of the y counter at the time when there are no more pixels from the count value of the area counter, the center of gravity of a group of pixels having connectivity to the labeling start pixel is calculated. The image processing method according to claim 4, wherein outputting is performed. 6. A counter for counting the number of times that the pixel extracted from the stack in the step (d) is determined not to satisfy the label condition in the step (b), and the counter is provided in the step (e). The count value of this counter at the time when there are no more pixels from the stack is output as the number of holes included in a pixel area having connectivity to the labeling start pixel. The image processing method according to any one of the above. 7. A starting point detecting step of examining an image in a predetermined scanning order, detecting a pixel that satisfies a label condition that a label is unassigned and has a pixel value of interest, and that determines a labeling start pixel. And a label applying step of applying a label to a pixel having connectivity thereto, and wherein the label applying step includes: a starting step in which a labeling starting pixel is a line starting point; and a pixel on a line including the line starting point. Scanning one pixel at a time, applying the label each time a pixel that has connectivity to the line start point and satisfies the label condition is added, and in adjacent lines above and below the scan line,
A line scanning step of storing a pixel satisfying the label condition for a pixel of the scan line as a connection candidate pixel in a predetermined storage unit, and when the labeling process on the scan line is completed in the line scanning step, A scanning repetition step of retrieving one connection candidate pixel from the storage unit and repeating the line scanning step using this as a new line start point.