JPS62251879A - Image contour extracting system - Google Patents

Abstract

PURPOSE:To simply extract the contour of an image with the small volume of calculation by executing a projection calculating process, a binarization calculating process and a continuity inspection to extract the contour. CONSTITUTION:An image inputted from a camera 1 is stored in a current image frame memory 3 through an A/D converter 2 and its changing area and changing picture element are detected by a changing area detecting part 4 and a changing picture element detecting part 7 respectively and the detected results are applied to a contour extracting part 8. The extracting part 8 consists of an X-direction slit projection calculating part 9, a Y-direction slit projection calculating part 10, a unit pattern table forming part 11, an contour pattern table forming part 12, and a link pattern table forming part 13. The extracted contour is applied to a collating part 17 through a shape drawing/extracting part 14 and an object model forming part 15 and collated with a model obtained from a model registering part 16.

Description

[Detailed description of the invention] [Technical field] The present invention relates to an image contour extraction method.

[Background technology] As a conventional contour extraction method, input image binarization Wi
A method has been proposed in which an image is created and tracked using, for example, a 3×3 pixel mask (spatial filter). This binarization process generally requires input j! ! There are two methods: one is to binarize an intermediate threshold value of the luminance difference of the i-image, and the other is to apply a differential operator to the input image and then binarize it to obtain a line image. This method of differentiation is used in research on artificial intelligence called Computeravikran, and is useful for creating computer models of visual field, but it requires a huge amount of time and is fast and expensive. It requires a computer device, so it can be said that it is hardly ever put into practical use. On the other hand, binarization processing based on the luminance difference of input images is widely applied to 7-actor automation, and the amount of calculation is relatively small. However, an image that is simply binarized based on the brightness difference with respect to a threshold value (!!I) has a lot of noise such as convex points and isolated points, and this binarized image is subjected to propagation, degeneration processing, etc. to remove isolated points, and edge detection is performed. spatial filter (e.g. 3x3 pixel Robi
There is a method of determining the contour using the nson operator), but in this case, the amount of calculation is enormous because all the spatial filters of the edge components are applied and the one with the highest degree of matching is extracted.

[Object of the Invention] The present invention was made in view of the above-mentioned problems, and its purpose is to provide an image outline extraction method that can easily and reliably extract the outline of an image with a small total gt. is to provide.

[Disclosure of the Invention] The present invention provides an extraction method that includes a plurality of pixels on the short side of an extracted image such as a binarized image based on a luminance difference of an input image or a changed pixel image detected by comparing an input image and a reference image. Set a rectangular slit, find the projection of the short side of this rectangular slit in the axial direction, binarize this projection based on the setting value corresponding to the size, and find the end points of the binarized line figure inside the rectangular slit. In addition, the rectangular slit is moved in the short side direction to extract the contour of the image to be extracted based on the connectivity of the edges of the line figures within the rectangular slit.

The present invention will be explained below with reference to Examples.

Figure 1 shows the circuit configuration of an embodiment of the present invention, and 1'
The V camera 1 is, for example, for displaying a surveillance image. Δ/D converter 21゛゛■ A/D converts the image input from the camera 1 into, for example, 256×25 G pixels, gradation 8-bit digital value, and imports it into the memory 3 as the current image 7. The change area detection unit 4 compares the previous image stored in the previous image 7 frame memory 5 or the reference (background) image stored in the reference image frame memory 61 in advance to examine the magnitude of change for each pixel. The interval between the pixels to be tested is 4P! ! It is designed to be taken roughly, such as by each element. In other words, the magnitude of change for each pixel is greater than the set value R1.
As shown in Figure 13, the projection, which is the accumulated value of pixels with large changes on the X-axis, is obtained, and the projection is similarly obtained on the Y-axis, and this projection ( A rectangular area whose size (projection) is larger than the set value R2 is extracted as a change ata, and furthermore, a marnon set around this detection area is taken and data of the change area is output. FIG. 2 shows an example of an input image, and shows that the coordinates (xa, ya) and (xb, yb) of a changing region are extracted by projection of a coarse Iil element.

The changed pixel detection unit 7 is for creating a changed pixel image in detail (for example, every III!+i pixels) for the pixels in the extracted changed area, and compares the current image with the reference image, Also, pixels with a large difference are binarized as 1, and the others are binarized as O. If isolated points are removed through degeneration and propagation during this changed pixel detection, more effective changed pixel detection can be achieved. Note that the set value R1' is set smaller than the previous set value R1.

The image consisting of the changed pixels detected by the changed pixel detection section 7, that is, the data of the image to be extracted, is taken into the contour extraction section 8 which employs the method of the present invention. The contour extraction unit 8 includes an X-direction slit projection calculation unit 9, a Y-direction slit projection calculation unit 10,
Uniy) Pattern table creation gIS11, contour pattern table creation section 12/link table creation section 13
The X-direction slit projection calculation section 9 calculates
As shown in Figure 3, from the changed pixels in the changed area, the projection in the short side direction of a rectangular slit elongated in the
It takes a value that indicates that there are excessive pixels depending on the size of the target object, such as pixels. In other words, the projections on the X axis of the changed pixels cut out as shown in Fig. 4 (a) and (C) by this rectangular slit in the X direction are shown in Fig. 4 (b) and (d).
The line figure inside the rectangular slit is obtained by binarizing with the set value R1O. This #i closed figure end point can be expressed by coordinates of the form Xff1lX3□, for example, as shown in FIG. 4(b).

Of course, it is easier to obtain the block number of a square lattice block (hereinafter referred to as a slit lattice) divided by the length of the short side of a rectangular slit as shown in FIG.

Now, if we collect the line figure created by this X-direction rectangular slit by moving the rectangular slit in the axis direction of the short side, then fjrJ
A binary image as shown in Figure 6 is obtained. Here, as shown by the broken line in Fig. 1, it is possible to obtain the contour by the method described below from the mapping situation of the slit grid only by calculating the projection of the slit in the X direction, but in this example, the size of the target object is sufficiently large compared to the slit grid, and in order to be strong against noise and prevent expansion of the contour, a rectangular slit elongated in the Y direction is formed using a Y direction slit projector 81510 as shown in FIG. Projection of short side direction (
The projection is also the same), that is, the X
The projection on the axis is determined as shown in FIG. 8(b), and it is binarized using the set value RIO, and the line figure inside the rectangular slit is determined to obtain a binary image as shown in FIG. 9. Furthermore, Fn+ in Figure 8
, ynz... indicate the end points of the line figure.

Now, as shown in FIG. 10, a binary image is obtained by the horizontal projection calculation using the Y-direction slit, and a common area is obtained by overlapping with the binary image obtained by the above-mentioned X-direction slit.
Obtain a binary image using a slit in the Y direction.

The binary image of the slit grid shown in Fig. 10 is
The unit patterns are expressed using the unit patterns shown in Figures (,) to (11) and created in the unit pattern table creation 4111. The same holds true when calculating the projections in each of the X and Y directions for the square slits of the slit lattice. Here, the unit pattern of fj411 figure (a) is above.

The 8 directions (down, left, right, left, right° upward diagonal, left, right downward diagonal) are all connected, and the others are (b) (cHd)...7 directions, 6 directions, 5 directions... ...and the number of connections is decreasing. From the slit binary image in FIG. 10, the 1st it!
If the 8-direction connection pattern I(a) is removed, a binary image of the contour as shown in FIG. 5 is obtained, but in this embodiment, the unit pattern of FIG. Create a contour pattern table that represents a contour figure like .

The link pattern table creation unit 13 regards the patterns within the slit grid as points (hereinafter referred to as slit points), and
From the contour pattern table corresponding to the contour figure as shown in FIG. 5, a link pattern table shown by connecting lines between slit points as shown in FIG. 12 is created. In FIG. 12, the mouth indicates the pixel of interest, and the connection relationship is indicated by the mouth.

In this link pattern table, FIGS. 12(a) to (C)
) The features of the outer periphery of the contour are expressed by expressing links (connections) in four or more directions using only the link patterns in the 1st to 3rd directions shown in (). The point where the outer peripheries intersect is expressed as an overlap of a plurality of link patterns as shown in FIG. 4(d).

Note that since the same unit pattern and link pattern are used for rotation, the direction of each pattern is also shown in the contour pattern table and link pattern table. Searching for links becomes easier by determining them from unit patterns in order to create a link pattern table.

Unit patterns and link patterns are written in the same format in the table, so even if they are redundant, the outline and outer shape of an object can be accurately and easily extracted and described, which is useful for object matching, recognition, etc. It is valid. Since the points and unit patterns are determined by projection calculation without using a spatial filter, the total number of fl-t can be reduced. The unit pattern is X.

It is determined by the shape of the projection in both Y directions.

In addition, for example, as shown in Fig. 14, the unit pattern of each slit lattice may be obtained by calculating the projection in both the X and Y directions using square slits that are larger than the spacing a of the slit lattice. In this case, each side of the square corresponds to the short side of the rectangular slit, and the unit pattern shown in FIG. 11(b) is obtained.

Of course, even when rectangular slits are used, an accurate link pattern cannot be obtained only from the projection information within the slit lattice, so the link pattern is extracted by also referring to the surrounding slit lattices from the result of the unit pattern. In addition, dividing one square block image from the beginning and calculating the average value within the block of the binary image, binarizing it in block units and finding it from the connection relationship of the surrounding blocks is rougher than projection, but simpler. This is a great method.

Therefore, in this embodiment, the process from creating a unit pattern to creating a link pattern can be performed on the same table, and the unit pattern in FIG. 11 and the link pattern in FIG. 12 can be expressed in the same way on the table. Although they can be written as and equated, they are separated to make the process clear.

When binarizing the slit lattice block by block, the unit pattern may be used as a link pattern, and the contour pattern table and link pattern table may be treated as one. However, it is better to preserve the shape information by calculating the projection in the X, Y directions and ＼ for each slit than by simply adding blocks, and global features can also be extracted by increasing the size of the slit. It can be said that there are advantages.

As a result of the above, it is clear that the amount of calculation is smaller than edge/contour detection using a spatial filter or the like, and macro features such as contours can be efficiently extracted even from noise such as isolation and α. Therefore, efficiency is improved by expressing binary images as a list in a table.

Now, in the embodiment circuit, shape factors (vt characteristic points) such as bending points and protrusions are extracted by the shape factor extraction section 14 based on the contour information expressed by the link pattern extracted by the contour extraction section 8 as described above. An object model is created in the object model creation section 15 based on the mutual relationship of the shape factors, and is compared with the object model registered in advance in the model registration section 16 in the matching section 1.
7, the detected object is determined by comparison, the movement of the detected object is detected by the tracking unit 18, and the shape of the object that has been compared is determined according to the pre-registered monitoring conditions of the condition setting unit 19. The abnormality determination unit 20 compares the motion with the shape and motion, and if the shape and motion match the above conditions, an alarm is issued.

[Effect of the invention 1] The present invention is capable of adding a plurality of pixels to the short side of an image to be extracted, such as a binarized image based on the luminance difference of the input image, or an image with changed pixels detected by comparing the input image and a reference image. Set a rectangular slit containing the rectangular slit, find the projection of the short side of this rectangular slit in the axial direction, binarize this projection based on the setting value corresponding to the size, and calculate the end point of the binarized line figure inside the rectangular slit. At the same time, this rectangular slit is moved in the short side direction and the outline of the image to be extracted is extracted from the connectivity of the end points of the line figure within the rectangular slit, so the projection calculation process and the binarization calculation process are Since contours can be extracted only by testing connectivity, image contours can be easily extracted with less calculation than conventional contour detection using spatial filters, etc., making object identification and recognition more efficient. This has the effect of making it possible to perform well.

[Brief explanation of drawings]

FIG. 1 is a circuit configuration diagram of an image contour extraction circuit according to the present invention, FIGS. 2 to 10 are operation explanatory diagrams of an embodiment of the present invention, and FIG. 11 is an example of a unit pattern used in the above, tjS1
FIG. 2 is an example of a link pattern used in the above, FIG. 13 is an explanatory diagram of the operation of detecting a changed area in the same as the above, FIG. Extraction part, 9
10 is an X-direction slit projection calculation unit; 11 is a unit pattern table creation unit; 1
2 is a contour pattern table creator r&8B, and 13 is a link pattern table creator. Agent Patent Attorney Stone 1) Ai 7 Figure 2 Figure 3 Xat Xat X++thXl+4 Figure 5 Figure 8 (0) (b) Figure 11 Figure 13 Sot x Axis Figure 12 Procedural Amendment-8: (Voluntary) 1. Display of the case 1985 Patent 1FFPt495213 2. Name of the invention Image outline extraction method 3. Person making the amendment Relationship with the case Patent applicant address 1048 Kadoma, Kadoma City, Osaka Prefecture Name (58
3) Matsushita Electric Works Co., Ltd. Representative Fujii 4, Agent postal code 530 5, Date of amendment order 6. Number of inventions increased due to 11111 No. 7, Subject of amendment Description

Claims (5)

[Claims] (1) A rectangular slit containing multiple pixels on the short side is set for the extracted image, such as a binarized image based on the brightness difference of the input image, or an image with changed pixels detected by comparing the input image and a reference image. , find the projection of the short side of this rectangular slit in the axial direction, binarize this projection based on the setting value corresponding to the size, find the end points of the binarized line figure inside the rectangular slit, and An image contour extraction method characterized in that the contour of the image to be extracted is extracted from the connectivity of the end points of the line figure within the rectangular slit by moving the line shape in the short side direction. (2) The rectangular slit is set in both the X and Y directions, and the line figure obtained by projecting the rectangular slit in the X direction onto the X axis and the line figure obtained by projecting the rectangular slit in the Y direction are obtained. 2. The image contour extraction method according to claim 1, wherein a unit pattern of overlapping line figures in both directions is obtained and the contour of the image to be extracted is extracted based on the connectivity of this unit pattern. (3) While moving the rectangular slit in the axial direction of the short side at a pitch narrower than the width of the short side, find the projection of the midpoint of the short side of the rectangular slit onto the long side axis, and find the end point of the line figure inside the rectangular slit. 2. The image contour extraction method according to claim 1, wherein the contour of the slit grid interval shorter than the length of the short side of the rectangular slit is determined by determining the length of the short side of the rectangular slit. (4) The axial contour of the image according to claim 1, characterized in that the projection in the X direction and the Y direction of the square slit having the length of the short piece of the rectangular slit is determined to determine the unit pattern of the lattice slit. Extraction method. (5) After binarizing each block of the slit lattice depending on whether the average value within the square slit is higher or lower than a predetermined set value, a unit pattern or link pattern is determined from the connection relationship with surrounding blocks. An image contour extraction method as claimed in claim 4.