JPH0130180B2 - - Google Patents

Description

[Detailed description of the invention] Technical field to which the invention pertains As a method for separating and identifying the shape of a pattern from an image captured by an industrial television camera or the like, there is a method of tracing the outline of a pattern.
The present invention relates to a method for performing this contour tracing at high speed.

Conventional technology and its problems Conventional pattern shape recognition uses the so-called Run
There are area-based tracking methods such as the Length Code method that check the overlapping parts of pattern scanning lines. In order to find other curved parts, etc.) and the perimeter, it is effective to directly trace the outline of the pattern. Here, a basic method for tracing the outline of this type of pattern will be described.

FIG. 1 shows a binary image F including a pattern S captured by an imaging device (not shown). This image F is created by converting an image signal obtained by so-called raster scanning in which scanning in the X direction is repeated in the Y direction as shown by the arrow AR to each pixel through a binarization means (not shown).
Consists of binarized pixel data for each PE, and the pixel data is "1" in the area of pattern S,
In the background, the coordinates (x,
y). Hereinafter, this pixel data will be expressed as F(x,y).

Next, contour tracking is started based on the image memory. First, scanning is started from the pixel PO at coordinates (0, 0) as in the raster scanning described above, and the tip surface element BO in the area of the pattern S is found. The coordinates of this pixel BO are stored as the contour pixel (first) as the starting point for contour tracing of the area of the pattern S, and as the end point. The tracking direction is assumed to be around the pattern S in a counterclockwise (clockwise) direction.

Pixel BO to find the direction to track next
is replaced with pixel P in a general sense, and considering a local region of 3 x 3 pixels centered on pixel P as shown in Fig. ) surrounding adjacent pixels A1 to A
8 pixel data is read from the image memory using the following coordinates (addresses). That is, if the coordinates of pixel P are (x, y), the coordinates of adjacent pixels A1 to A8 are A1; (x+1, y), A5; (x-1, y), A2; (x+1, y-1), A6; (x-1, y+
1), A3; (x, y-1), A7; (x, y+1), A4; (x-1, y-1), A8; (x+1, y+1).

Further, the definition of a direction code (hereinafter simply referred to as direction) for expressing the direction of tracking is shown in FIG. That is, it is expressed as a number from 1 to 8 at intervals of 45 degrees.

Now, to describe the method of finding the tracking direction from the starting point (pixel BO) (referred to as the initial tracking method), there is no pattern in the direction 5 where raster scanning has been performed, so the pattern is If it rotates in a counterclockwise (clockwise) direction as shown in FIG. Adjacent pixel A1 in the order of 1, 1 →...
- Start examining the pixel data of A8 and proceed in the direction in which "1" is found (next direction code (second)). At this time, the coordinates of the contour pixel (second) adjacent to the tracking direction are the fourth
As shown in the figure, the coordinates (x, y) of the pixel BO as the pixel of interest are determined by adding the operation amount determined by the next direction code.

Note that in determining the next direction code (second), one step is wasted due to the method of determination, but direction 5 (-
The pixel data may be examined starting from the X direction).

Next, a third contour pixel tracing method will be described. The second contour pixel is set as the pixel of interest, and the pixel data of the adjacent pixels A1 to A8 are read in the same manner as described above. Principles for determining the subsequent tracking direction (referred to as normal tracking method)
is a direction perpendicular to the direction of the previous direction code (in this case, the above-mentioned next direction code (second)) that has arrived at the pixel of interest (this is called the scan start direction code). The pixel data of adjacent pixels A1 to A8 are scanned in the same way as above in the counterclockwise (clockwise) direction, and the direction in which "1" is hit for the first time (the next direction code (third)) Proceed to. Further, the adjacent third contour pixel at this time is also obtained in the same manner as described above. Note that FIG. 5 shows the relationship between the forward direction code and the scanning start direction code. Thereafter, the contour is traced in the same manner as the third contour pixel and the next direction code (third) are obtained from the second contour pixel and the next direction code (second), respectively.

The contour tracing described above ends when the coordinates of the contour pixels, which are sequentially obtained while going around the pattern, coincide with the starting point.

From the coordinates of each contour pixel determined in this way and their direction codes, the detailed features and perimeter of the contour are determined as described above.

The disadvantage of the above method is that the coordinates and pixel data of the pixel of interest and eight pixels adjacent thereto must be read out each time one contour pixel to be tracked is determined, which takes processing time. Furthermore, even if the software for performing the above processing is attempted to be constructed simply by hardware, the construction will be complicated and the number of accesses to the image memory will be large, making it difficult to shorten the processing time.

Objects and Points of the Present Invention The present invention aims to eliminate the drawbacks of the conventional pattern contour tracing methods described above, and to provide a pattern contour tracing method that is easy to implement in hardware and capable of high-speed processing.

The key point of the present invention is that, although the basic principle of pattern contour tracking is the same as the conventional one, after converting the image signal obtained by raster scanning into binarized serial pixel data, this serial pixel data is further processed. Multi-value pixel data (multi-value code) that corresponds to the pixel of interest in a local area of 3×3 pixels centered on the pixel and is determined by the pixel data of its adjacent pixels.
is stored in a multi-value image memory in the form of a multi-value image through a multi-value code conversion means that converts the multi-value code into a multi-value image, and the multi-value code of the contour pixel and the forward code that traced the contour pixel are used as addresses and then The next direction code is immediately searched by the next direction search means written in advance so that the next direction code to the adjacent contour pixel to be tracked can be searched.

Embodiment of the present invention FIG. 6 is a block diagram showing the configuration of the main part of the present invention. That is, 601 is a multi-value code encoder, 602 is a multi-value image memory, 603 is a memory scanning section, 604 is an end monitoring section, 605 is a next direction search means, 605a is a next direction table, and 605b is a forward direction code latch. Next, the operation of FIG. 6 will be described with reference to FIGS. 7 to 9. The multilevel code encoder 601 converts the binary image F shown in FIG. 1 into the binary image F shown in FIG. The pixel F is converted into a multivalued image FM consisting of multivalued codes corresponding to each coordinate point.

Here, the internal configuration of the multi-level code encoder 601 is as shown in FIG. 8 in Figure 8
01 is a 3-bit shift register as a local register, and 802 is a multi-bit shift register as a remaining register, which has a bit length smaller by 3 bits of the local register 801 than the number of scanning pixels in the X and Y directions. In the local register 801, the binary images F connected in series by raster scanning are shown by the arrow AR.
Since it is input in the order of 8 and shifted, these 3
The nine pixel data in the two local registers are inverted vertically and horizontally for convenience of illustration, but they correspond to the 3×
This is pixel data for a local area of 3 pixels, and since the pixel P of interest sequentially moves through each coordinate of the binary image F by raster scanning, it is as if each coordinate point of the binary image F is a window of 3 x 3 pixels. It will be shaped like raster scanning. However, if the coordinates of the pixel P of interest fall within two rows or two columns of the edge of the screen of the binary image F, the pixel data for the 3 x 3 pixels will therefore contain an incorrect value in this multi-value code (described later). Therefore, ignore it and write “OOH” in this case.

Next, to describe how to convert the pixel data of the local area obtained in this way into the multi-level code, the adjacent pixels to the pixel of interest P are A1 to
It is 8 of A8. When the pixel data in the current pixel P is “O” (background), in FIG. 8, the pixel data and the adjacent pixel A1 are
Since an AND condition is applied to each pixel data in -A8, the values b0 to b7 in the multi-value code latch 804 are output as "OOH" regardless of the value of each pixel data. When the pixel data in the pixel of interest P is "1" (pattern), each pixel data in the adjacent pixels A1 to A8 corresponds to the values b0 to b7 in the multi-value code latch 804 and is output as 8-bit information. be done. This is the multi-value code mentioned above, and becomes the output of the multi-value code encoder 601 in FIG.
It is written into the multi-value pixel memory 602 in synchronization with the input of 1 and stored in the form of the multi-value image FM shown in FIG. now,
The multi-value code of the pixel of interest (referred to as the former) at the coordinates (x, y) on the multi-value image FM is expressed as FM(x, y)
The pixel of interest corresponds to the pixel of interest (referred to as the latter) at the coordinates (x, y) on the binary image F in FIG. y).

In other words, the multi-valued image FM has the same size as the binary image F, and each coordinate point corresponds to the other, and instead of the 1-bit pixel data F(x,y) in the latter, 8-bit FM is used.
It has (x, y). Furthermore, as is clear from the above description, the shape of the multivalued pattern SM in the multivalued image FM is exactly the same as the pattern S in the binary image F. In addition, in Fig. 7, the multivalued code FM (x, y) of each coordinate point is
Shown in hexadecimal representation.

In this case, accessing one pixel of the multilevel image memory 602 corresponds to accessing nine pixels of the conventional binary image memory. In other words, adjacent pixels A1~
These are 8 pixels (8 bits) of A8 and the pixel P of interest itself (1 bit of "0" or "1").

By the way, as explained in the prior art, the tracking direction (next direction code) from a contour pixel that is a pixel of interest to the next adjacent contour pixel is not determined only by the pixel of interest and its adjacent pixels, but is determined by the forward direction code that reaches the pixel of interest. It can only be determined by knowing the That is, in FIG. 6, the next direction table 605a is the multilevel image memory 6.
A search table made of ROM or RAM that is accessed by a total of 11 bits of input, including 8 bits of adjacent pixel information obtained from 02 and the forward code (8 directions, 3 bit information) output from the forward code latch 605b. The next direction code for the next tracking direction for each of the 11-bit values is written in advance, and the next direction code is immediately output as 3 bits. However, this output is input to a memory scanning unit 603 that generates an address for accessing the multilevel image memory 602, and a forward code latch 605b.
It is designed to be used both when writing to 02 and during tracking processing to trace the outline of a pattern. Therefore, the next direction table 605a stores the FM (x, y) when writing the multi-value code FM (x, y) to the multi-value image memory 602 and during the tracking process.
When the value of is "OOH" (background part), a direction code "1" is output which increases the scanning by +1 in the X and Y directions.

FIG. 9 shows part of the contents of the next direction table 605a. This is based on the normal tracking method described in the prior art section. That is, a local area LA having five types of memory contents, a multilevel code TA1 corresponding to the local area LA, three types of forward codes (1, 5, 6) TA2, and the multilevel code TA1 and the corresponding multilevel code TA1. Table address TA as part of the combination of forward code TA2 causes next direction table 605
The correspondence of eight examples of next direction codes TO as table outputs retrieved from a is shown. However, this figure is an example in which the pattern outline is tracked counterclockwise. An application example of FIG. 9 will be described later.

Again in FIG. 6, the memory scanning unit 603 generates a table for obtaining the operation amount as shown in FIG. 4 in response to the output of the next direction code of the next direction table 605a, and the x and y coordinates after the operation. It consists of an accumulator (not shown) that holds the accumulated value of additions and subtractions. Therefore, the memory scanning section 60
3 is the next direction table 605 when storing the multi-value image FM in the multi-value image memory 602 as described above.
The output of a advances the x coordinate by +1, and when the total number of pixels in the . In this way, 2 on each coordinate point (x, y) in the binary image F
The value pixel data F (x, y) is added to the multi-value code FM (x, y) on each coordinate point (x, y) in the multi-value image FM.
y).

Next, the operation of the memory scanning section 603 when performing pattern tracking processing based on the multivalued image FM in the multivalued image memory 602 will be described with reference to FIG. In this case, memory scanning to the multivalued image memory 602 is a random access, unlike when storing data. That is, at the start of the tracking process, the memory scanning unit 60
The value of the accumulator corresponding to access coordinates (x,y) in 3 is set to (0,0). Next, the address in the multivalued image memory 602, in other words, the address in FIG.
The memory (multi-value code) of PO point coordinates (0, 0) is read out. The multi-value code is the next direction table 6
05a, and the next direction code is output from the next direction table 605a. As mentioned above, when the multi-value code FM (x, y) of the coordinate (x, y) point is "OOH", the output value becomes a direction code that increases the X direction scan by +1, and also for the Y direction scan. The memory scanning unit 603 is a multilevel image memory 60
It operates in the same way as writing to 2. Therefore, scanning and tracing is performed by raster scanning up to the coordinate point BO in FIG. 7 where the multivalued code FM (x, y) is not "OOH". The coordinate point BO is the starting point for tracing the pattern outline described above, and thereafter random access is performed along the outline. Termination monitoring unit 604
is the output of the multi-value image memory 602, that is, the multi-value code
FM (x, y) is monitored, the coordinates of the starting point BO where the multi-valued code first becomes a value other than "OOH" are stored, and the output coordinates of the memory scanning unit 603 are thereafter stored while the contour tracking process is being executed. If the output coordinates match the coordinates of the starting point BO, it is determined that one round of contour tracing has been completed and the tracing process is terminated. Also note BO
Then, the end monitoring unit 604 sends a starting point signal to the forward code latch 605b to inform it that it is the starting point, and when the pattern contour tracing direction is counterclockwise (clockwise), a virtual forward code is set as "8," for convenience.
2” is output. This is the next direction table 605.
This is because the memory contents of a are designed to search for the next direction code based on the normal tracking method described in the prior art section. 2'' address causes the next direction table 605a to operate in such a manner that the scanning start direction code described in the prior art section is equal to ``6,4'', in other words, the tracking direction is changed counterclockwise from the direction of ``6,4'' ( Outputs a new direction code that is found by scanning in the clockwise direction. Therefore, this method is equivalent to the method of tracing the second contour pixel from the starting point (first contour pixel) described in the prior art section. In this case as well, as mentioned in the prior art section, the scanning start direction code is made to correspond to "5 (-X direction)", that is, the forward direction code is "7,3" instead of "8,2". You can also use it as

In addition, completely apart from the above method, the next direction table 6
05a, the next direction code (second) from the starting point is based on the initial tracking method described in the prior art section, while being aware that the memory amount of 05a is increased and the circuit configuration is slightly complicated.
A dedicated memory area for finding the code is provided in the next direction table 605a separately from the memory area for the normal tracking method, and only the next direction code (second) is based on the starting point signal from the end monitoring unit 604 and the multi-valued starting point BO. code and the next direction table 605a as an address.
A method may be adopted in which the information is found from the dedicated memory area, and this idea is also included in the present invention.

After the next direction code (second) is determined, the forward direction code latch 605b is set to the next direction table 605a.
The output of is latched and output as a forward code for direction tracking of the next contour pixel.

Figure 7 shows an example of contour tracking performed in this way.
This is shown in FIGS. 9 and 1. That is, among the multi-value codes in the contour pixels on the multi-value pattern SM in FIG. 7, TA11 to TA14 are the multi-value codes in FIG.
Corresponding to the same code in TA1, when going around the outline of the pattern in a counterclockwise direction, the above multivalued code TA11~
The direction of contour tracing of the portion of each example contour pixel corresponding to TA14 corresponds to the direction of the next direction code TO corresponding to the respective forward direction code TA2 in FIG. In addition, in FIG. 1, the local areas LA1 to LA4 in FIG. 9 centered on the above-mentioned exemplary contour pixels are indicated by the same reference numerals, and at the same time, the direction of contour tracking centered on each of the above-mentioned exemplary contour pixels is indicated by arrows. There is. With the above configuration and functions, if the outputs of the memory scanning unit 603 and the next direction table 605a are stored in another memory (not shown) from the start point to the end point of contour tracking, the position of each contour point (pixel) ( List information of coordinates) and directions is obtained, and the number of pixels of the contour part (boundary point) is obtained by counting the number of tracking processes from the starting point to the ending point. Note that these processes can be performed particularly at high speed.

In this example, the case of 8 adjacent pixels was explained, but
The same thing is possible in the case of four adjacent pixels and is included in the present invention. In the case of four adjacent pixels, multi-level coding is performed using adjacent pixels A1, A3, A5, and A7 in FIG. 2.

In the embodiments described above, the pattern is completed by going around the outline of the pattern once, but for example, in the case of "one-step writing" or character recognition, tracking may be temporarily interrupted and proceeded. In other words, when separating the drawing strokes for a pattern after thinning the input image, the starting point in the above embodiment is taken as the endpoint (the multi-value code becomes a special value), the thin line is traced, and a new endpoint is set as the end point. Alternatively, stroke separation can be efficiently performed by taking an intersection.

In the above description, words and numerical values that change depending on whether the pattern is traced counterclockwise or clockwise are shown outside the parentheses when the pattern is traced counterclockwise, and those inside the parentheses when the pattern is traced clockwise. Furthermore, it is obvious that even if the horizontal, vertical, or X-direction and Y-direction relationships are interchanged, they are irrelevant to the idea of the present invention.

Effects of the Invention and Field of Application According to the present invention, as a preprocessing for tracing the outline of a pattern, the pixel data of each pixel in a binary image is converted into a multi-value code representing the pixel data of a local area centered on the pixel. By applying the multi-value code conversion means, for example, in the case of eight adjacent pixels, one access to the image memory in contour tracking processing is equivalent to nine accesses in the past, and at the same time, the multi-value code for the contour pixel of interest and its By using the next direction search means that searches for the next direction code from the previous direction code for the pixel of interest, the contour tracking process that should be performed for each image can be performed at extremely high speed. It becomes possible to eliminate problems.

The present invention can be widely applied as a basic method to various fields such as measurement using images, inspection, positioning control, and character recognition.

[Brief explanation of drawings]

Fig. 1 is a diagram showing an example of a conventional binary image, Fig. 2 is an explanatory diagram of a local area, Fig. 3 is an explanatory diagram of a direction code, and Fig. 4 is an illustration of the next direction code and the coordinate value of the pixel of interest. FIG. 5 is a diagram showing the relationship between the operation amount and the forward direction code, and FIG. The figure is a block diagram showing the configuration of the main part of the present invention, FIG. 7 is a diagram showing a multivalued image corresponding to FIG. 1 according to the present invention, and FIG.
9 is a diagram showing the contents of the multi-value code conversion means in the present invention, and FIG. 9 is a diagram showing part of the contents of the next direction table in the present invention. Code explanation, 601...Multi-value code encoder,
602...Multi-valued image memory, 603...Memory scanning section, 604...End monitoring section, 605...Next direction search means, 605a...Next direction table, 605
b...Forward code latch, F...Binary image, S
...Pattern, BO...Starting point, P...Pixel of interest,
A1 to A8...adjacent pixels, FM...multivalued image,
SM...Multi-value pattern.

Claims (1)

[Claims] 1. Converting an image signal obtained by repeating scanning in the horizontal (vertical) direction (referred to as the X direction) in the vertical (horizontal) direction (referred to as the Y direction) to binary serial pixel data. a binarization means for converting the pixel data of the pixel of interest and the pixel data of adjacent pixels surrounding the pixel of interest into multi-value pixel data (referred to as a multi-value code) for the pixel of interest; A multi-value image storage means for storing a multi-value image made up of the multi-value code, and a next direction search for outputting a next direction code indicating the tracking direction from an arbitrary contour pixel (referred to as A) of the pattern to an adjacent contour pixel. have the means,
Inputting the multivalued code of the contour pixel (A) as an address to the next direction search means, and identifying that the contour pixel (A) is the starting point when it is the starting point of pattern contour tracing. A pattern characterized in that a starting point signal is inputted, and when the contour pixel (A) is not the starting point, a forward direction code indicating a tracking direction from an adjacent contour pixel that traced the contour pixel of the instep is inputted. Contour tracking method. 2. In the pattern contour tracing method according to claim 1, the starting point signal is -X
1. A method for tracing contours of a pattern, comprising inputting a virtual forward code in a tracking direction in which a direction code in the (-Y) direction or an adjacent direction is in a relationship with a scan start direction code.