JPH06259554A - Picture outline extracting device - Google Patents

Abstract

PURPOSE:To automatically initialize an outline model and to extract an outline by surely converging the initialized outline model. CONSTITUTION:The picture outline extracting device is constituted of a picture storing means 101 for storing a picture with a prescribed composition as an object from which an outline is exracted, a contact point extracting means 102 for finding out the coordinates of two points on which the outline extracting object of the picture comes into contact with a background by scanning the peripheral edge of the picture, an initial outline model acquiring means 103 for finding out an initial outline model with a shape along a line surrounding a prescribed range by using the coordinate positions of the two points found out by the means 102 as end points, an outline model string means 104 for storing the position of a sequence of dots expressing the outline model, and converging operation means 106 for finding out the moving position of the dot sequence in order to converge the obtained outline model to the outline of the outline extracting object and moving the dot sequence.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image contour extracting apparatus for extracting the contour of a target human image in the field of image processing.

[0002]

2. Description of the Related Art In the field of image processing, feature extraction and
The contour is extracted to cut out an image.

As one of the conventional contour extracting methods of this type, for example, M. Kass, A .; Witkin, D .;
"SNAKES: Activ" by Tezopoulos et al.
e Contour models, Proc. 1st
ICCV, pp259-268 ".

Active Con shown in this document
Tour models are a technique for capturing the contour of an image by minimizing an energy value defined as an amount corresponding to the state of a contour model composed of a point sequence. That is, this idea is as follows.

When it is desired to capture the contour EL of the image as shown in FIG. 9A, first, the contour EL of FIG.
As shown in (b), the points P1, P2, P3, ..., Pn are plotted, and the plotted points P1, P2, P3 ,.
A line image is obtained by connecting.

The obtained line image is a contour model EM, and the energy generated in the contour model EM consists of internal energy and external energy.

The internal energy is determined by the continuity of the point sequence and the degree of curve of the point sequence, and the external energy is determined by the positional relationship between the contour model and the contour of the image.

Therefore, M. Kass is a vector v with the length s of the arc as a parameter for the contour model as in the following equation.
It is represented by (s).

V (s) = (x (s), y (s)) Then, when the energy Esnake represented by the following equation has a minimum value, the contour model captures the contour of the image. .

[0010]

[Equation 1] Here, Eint in the above equation is Eimag, which is the internal energy related to the bending degree and discontinuity of the contour model.
e represents energy working with the image, and Econ represents external energy that is a constraint condition.

[0011] In addition, Amini describes the document "Proceed".
ings, Second International
Conference on Computer V
Ision, 1988, pp. 95-99 ",
An improved processing method is proposed by dynamic programming.

[0012] Further, Cohen is in the document "IMAGE".
UNDERSTANDING Vol. 53, No.
2, March, pp. 211-218, 1991 ″ proposes an improvement plan such as introducing energy that works so as to increase the area surrounded by the contour model.

By applying these ideas, the contour model of the image to be contour-extracted is given, and the contour calculation can be performed by converging the contour model to the contour of the image to perform the contour extraction of the image. Like

[0014]

Incidentally, the above-mentioned A
In the contour extraction method using the active contour models, the initial position of the contour model is usually manually input, and is generally determined by approximating with a polygon.

Therefore, if the number of images to be contour-extracted is small, this is not a serious problem, but the contour extraction processing must be performed on a large number of images such as a printer for issuing an ID card. In this case, the burden on the operator increases. Therefore, it is hoped that an image contour extraction device will be realized in which the initial setting of the contour model can be automated and the contour extraction can be performed automatically.

To automate the initial setting of the contour model,
Roughly set a relatively wide area that includes the image you want to extract the contour, plot the boundary of this set area, and use the point sequence obtained by the plot as the default contour model. Just do it.

However, with respect to the initial set contour model, in order to converge the point sequence forming the contour model, calculation is performed based on the above idea of energy, and the energy of each point of the point sequence is minimum as a sum. The position is moved so as to become, and the moving position is obtained by calculation, and one point is gradually converged.

Therefore, for example, when the point sequence of the initial contour model is applied to the image portion to be contour-extracted, and the image has a pattern confusing with the contour, etc. Does not expect good convergence of
It is unavoidable that it takes time to converge and that contour extraction fails. Therefore, even if the initial contour model is blindly set, many failures occur.
This is not applicable to automation.

Therefore, the object of the present invention is to
An object of the present invention is to provide an image contour extraction device capable of automating the initial setting of a contour model and surely converging the default contour model in a short time to perform contour extraction.

[0020]

In order to achieve the above object, the present invention is configured as follows. That is, an image storage means 101 for storing an image of a predetermined composition which is a target for contour extraction, and a contact point extraction means for scanning the peripheral edge of the image to obtain the coordinates of two points where the contour extraction target of the image and the background are in contact. 102, and an initial contour model acquisition means 103 for obtaining an initial contour model having a shape along a line surrounding a predetermined range with the coordinate positions of the two points obtained by the contact extraction means as endpoints.
And a contour model storage means 104 for storing the position of the point sequence representing the contour model, and a convergence calculation means for obtaining and moving the position of the point sequence so that the obtained contour model converges on the contour of the contour extraction target. It consists of 106 and.

[0021]

In such a structure, for example, the edge of the screen is scanned with respect to the contour extraction target image such as a person image photographed with a predetermined composition, a picture of a person, and an object image of a predetermined form. The boundary point between the contour extraction target image and the background is known from the pixel at the position, a predetermined range surrounding the image of the contour extraction target is set from that position, and a point sequence that traces the set range boundary is created. Generate a contour model. Then, for this initial contour model, the moving position of the point sequence of the above-mentioned contour model that converges to the contour of the contour extraction target image is obtained, a convergence calculation for moving is performed, and the contour model is converged to obtain the contour of the image. Then, the contour model that fits is obtained and used for the contour extraction of the image of the contour extraction target.

Therefore, according to the present invention, the initial setting of the contour model can be automated, and the image contour extracting apparatus can surely converge the contour model of the initial setting in a short time to perform the contour extraction. Can be provided.

[0023]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of this embodiment.

In FIG. 1, 101 is a frame memory for human images, 102 is a contact position extraction circuit, 103 is a contour model initial position setting device, 104 is a point sequence position memory, 10
5 is a contour model area calculation circuit, and 106 is a convergence calculation device.

Of these, the person image frame memory 101 is for storing the data of the person image to be the object of contour extraction, and the contact position extraction circuit 102.
Scans the left and right sides and the lower side at the outer frame (screen outer frame) position of the image in the frame memory 101, and uses the difference between the color of the person image and the background to contact the person region and the background region, that is, the image The points A and B at which the person area and the background area are in contact with each other along the screen periphery corresponding to the outer frame of the screen (two points on the left and right sides or the lower side)
This is a circuit for obtaining the coordinates of (point).

Further, the contour model initial position setting circuit 103
Is the above-mentioned 2 output obtained by the contact position extraction circuit 102.
This is to set the initial position of the contour model in which the coordinates of points (points A and B) are fixed at both ends. The initial position of the contour model calculated by the contour model initial position setting circuit 103 will be described later with reference to FIG. 4, but the coordinates of the above-mentioned two points (point A and point B) rise parallel to the Y-axis with both ends fixed. , X passing through a preset upper limit position y in the Y-axis direction
The boundary lines of the area surrounded by the line parallel to the axis are plotted at almost predetermined intervals.

The point sequence position memory 104 is a memory for storing the position of the point sequence of the contour model obtained by the contour model initial position setting circuit 103 as described above.
The contour model area calculation circuit 105 is a calculation device that calculates the area surrounded by the contour model, and the convergence calculation device 106.
Is an apparatus that performs a convergence process that is a process of moving the position of each point of the contour model so that the energy of the contour model is minimized.

In the present apparatus having such a configuration, the image of a person photographed by a predetermined framing (composition), for example, the image data of an upper half of a person photographed in a rectangular frame, is used as a frame for a person image. When the data is stored in the memory 101 and the operation is started, first, the contact position extraction circuit 10
2 scans along the edge of the frame of the person image data in the person image frame memory 101, and uses the difference between the person image and the background color to determine the position where the person area and the background area are in contact, that is, on the screen. A point A and a point B at which the person area and the background area are in contact with each other by tracing the edge 2
Obtain the coordinates of the points (two points on the left and right sides or the lower side).

When the coordinates of the points A and B are obtained, the contour model initial position setting circuit 103 then determines the A
Based on the coordinates of the point and the point B, an initial contour model having these coordinates as the ends is automatically created.

This contour model extends straight lines from the points A and B along the Y-axis, respectively, and
It has a rectangular shape along the boundary line of a region surrounded by a line extending straight along the X axis, passing through a preset Y coordinate position y. This contour model is created as a series of points in which the above-mentioned boundary lines are plotted at almost predetermined intervals. The position of the point sequence of the contour model thus obtained is stored in the point sequence position memory 1
It is stored in 04.

When the initial contour model is obtained, the area calculation of the contour model is started. This is performed by the contour model area calculation circuit 105. When the area enclosed by the contour model at the initial stage is obtained, the process of converging the contour model is started next.

This processing is performed by the convergence calculation device 106. The convergence calculation device 106 first receives the area data output from the contour model area calculation circuit 105 and other data, and obtains the energy value of the contour model. Then, the position where the contour model should move is calculated. This may be performed by a method similar to the conventional Active Contour models, focusing on energy generated by the positional relationship with the image, energy increasing according to the point interval, energy increasing as the radius of curvature decreases, The position of each point p of the contour model is moved so that the total energy value obtained by summing these energies becomes small.

By performing such processing, the initial contour model is converged, and the contour model converged on the contour of the person is obtained.

The processing in this apparatus as described above will be described more specifically.

FIG. 2 shows a method of determining the positions of both end points of the contour model. Here, initial data of the contour model is created by utilizing the characteristics of the person image prepared according to the purpose, such as a photograph of the upper half of the body used for an ID card or the like.

FIG. 2A shows a human image which is the object of contour extraction and is stored in the human image frame memory 101. The person image has a form in which the upper half image of the person is shown on a background of a single background color, for example, a single color such as blue, and the one-frame person image is composed of a rectangular area having a predetermined frame size. Therefore, it is assumed that a person image is captured with a size that fits within the frame size and the data of the person image is stored.

If the person image data to be contour-extracted is stored in the person image frame memory 101, the contour automatic extraction is started next. In the contour automatic extraction, first, the data of the person image is used. , Check the coordinate positions of the lower sleeves of the image of this data.

This is performed by the contact point position extraction circuit 102. That is, the contact point extraction circuit 102 first determines the specific point C1 of the background color area and the specific point C of the person image area of this image.
Sampling the two colors of 2. The specific point C1 of the background color area and the specific point C2 of the person image area are the upper-half portrait images of the person image used for identification, license, etc. However, it fits in the almost specified area. Therefore, the specific point C in the background color area
1 is set in the background portion and at a position where the human portion is considered not to be located even if the image is biased, and the specific point C2 in the human image area is where the human portion is located and the image is biased. Is set to a position where the background part is considered not to be located.

When the sampling of the color at each position of the specific points C1 and C2 is completed, the contact point position extraction circuit 102 then starts the operation of searching for the boundary position between the background and the human image. This is performed by scanning the pixels on the outermost periphery of the person image frame memory 101 downward from the intermediate points A0 and B0 at the left and right sides thereof (see FIG. 2C). Then, when the bottom point (the bottom edge of the screen) is reached, scanning is performed on the bottom side of the screen toward the inside.

The above-mentioned intermediate points A0 and B0 may belong to the background area of the image. Here, the midpoint of the left and right sides of the image is used. Then, while performing the above-described scanning, the color at each position (point) is read out, and the color and the specific point C are read.
Calculate the color distance from 1, C2. Then, the specific point C1
The scanning is terminated when the color distance from the specific point C2 is closer than the color distance from () (see FIG. 2D).

As a result, the points A and B at which the background area and the person area are in contact with each other can be extracted on the left and right sides or the lower side of the screen. A method such as edge detection using a differential operator may be used to extract the points A and B.

Further, as shown in FIG. 3, even when the person is displaced from the center of the image, the points where the person area and the background area are in contact are similarly extracted by scanning from the left and right sides. be able to.

In this way, when the points A and B where the background area and the person area contact each other are extracted, the contact point position extraction circuit 102
Outputs the coordinate information of the points A and B and supplies it to the contour model initial position setting circuit 103. Then, the contour model initial position setting circuit 103 starts the contour model setting process using the coordinate information of the points A and B output by the contact point position extraction circuit 102.

That is, the contour model initial position setting circuit 1
When 03 obtains the coordinate information of the points A and B output by the contact point extraction circuit 102, a straight line extending in the Y-axis direction (above the head of the image) is assumed with the coordinate position indicated by the coordinate information as a starting point.
In addition, as shown in FIG. 4A, a point p is set on the boundary of a rectangular area surrounded by a straight line parallel to the X axis passing through the coordinate y on the Y axis, and a frame of a point sequence is set. To do. This sequence of points becomes the initial position of the contour model.

The point interval d of this point sequence p and the upper limit position y
Is given in advance. The point intervals d do not have to be strictly equal intervals, so that the point intervals d may be adjusted appropriately at places where the point sequence is bent. Even if the person in the image is displaced from the center, as shown in FIG.
By setting the point sequence for the rectangular area surrounding the upper part of the head of the image with the coordinate position of the point B as the starting point, the initial position can be set as in the case of a normal image without bias.

The position of the top point of the person's head (point T in FIG. 5) in the image can be easily obtained by checking the color difference from the background color. The Y coordinate of the top point T may be used.

The information of the initial position of the point sequence set by the contour model initial position setting circuit 103 in this way is stored in the point sequence position memory 104 as the information of the initial contour model.

When the information of the initial contour model is stored in the point sequence position memory 104, the contour model area calculation circuit 105 calculates the area of the contour model using the information of the contour model.

Although various methods can be considered for this calculation,
An example will be described with reference to FIG. As shown in FIG. 8, the coordinates of the vertices of a polygon consisting of a sequence of n points are (X1, Y1), (X2, Y2), ..., (Xi,
Yi), (Xj, Yj), ..., (Xn, Yn), the area S of this polygon is calculated by the following equation.

[0051]

[Equation 2] However, if i ≠ n, j = i + 1, and if i = n, j = 1.

On the other hand, since the contour model is also a polygon consisting of the point sequence P as described above, the coordinates of each point constituting the contour model are read from the point sequence position memory 104 and are applied to the above formula to obtain the contour. The area S of the model can be obtained at any time.

When the area enclosed by the contour model at the initial stage is obtained, the process of converging the contour model is started next. This processing is performed by the convergence calculation device 106,
First, 6 receives the area data output from the contour model area calculation circuit 105 and other data, calculates the total energy value E _TOTAL of the contour model, and calculates the position to which the contour model should move. The aforementioned M. Kas
Similar to s, when the contour model is represented by a vector v (s) having the arc length s as a parameter, the total energy value E _TOTAL of the contour model is given by the following equation.

[0054]

[Equation 3] However, Earea = k · S (k is a constant, S is the area of the contour model) In the above equation, Eint is the internal energy related to the curvature of the contour model and discontinuity, and Eimage is the energy that acts between the image areas, Econ represents each external energy that is a constraint condition. Earea is energy corresponding to the area surrounded by the contour model.

The convergence calculation device 106 moves the position of each point P of the contour model so that E _TOTAL, which is the sum of these energies, becomes small.

As the energy generated in the positional relationship with the image, the amount of the edge component of each part of the image is generally used. In the present embodiment, a pixel having a larger color difference from the average color of the background area is By introducing the increasing energy, the background and the person can be separated.

A method for determining the destination position of each point p of the contour model will be described by taking the n-th point pn as an example.

FIG. 6 is a diagram showing a state near the nth point pn. In the figure, the point indicated by the white circle in the center is the n-th point pn. Further, the two line segments extending from the nth point pn connect the n-1th point pn-1 and the n + 1th point pn + 1 to the nth point pn, respectively.

First, the energy value is calculated with the n-th point pn placed at the current position indicated by the white circle (FIG. 6).
(See (a)). Next, the nth point pn is placed at eight positions indicated by black circles near the current position, and the energy value at each position is calculated (FIGS. 6B and 6B).
(See (c)). Then, in order to reduce the total energy of the contour model, a position pn ′ having the smallest energy value is obtained from the above 9 positions, and the n-th point pn is moved to that position.

When such a movement is repeated for each point p of the contour model shown in FIG.
After the state shown in FIG. 7A, the image converges on the contour of the person as shown in FIG. 7B. The convergence calculation device 106 performs such processing to converge the initial contour model, and obtains the contour model converged on the contour of the person.

Then, the contour in the image can be extracted from the coordinate information of each point of the converged contour model.

As described above, according to the apparatus of the present embodiment, from the contour extraction target image adjusted in advance to a predetermined composition, the boundary points of the background and the contour extraction target are determined from the contour extraction target image. The contour model is obtained by scanning the edge of the contour, and based on this boundary point, the area surrounding the contour extraction target is determined, and the initial contour model is obtained by plotting the boundary. it can.

Further, by performing the convergence processing in the direction of reducing the area enclosed by the contour model, the point sequence of the contour model located at a position distant from the human part can be easily converged to the human contour position. The appropriate range that can be set as the position becomes large. Therefore, it becomes easier to automatically set the initial position of the contour model.

Further, when considering a photograph of the upper body which is generally used for an ID card or the like, two points on the left and right sides or the lower side, which are the boundary points between the person and the background in the lower part of the photograph, are
It is the left and right limits of the range in which the person portion exists. Therefore,
By capturing the coordinates of these two points, the range suitable as the initial position of the contour model can be limited, so that the initial position of the contour model will not be too far from the actual contour, and the time required for convergence of the contour model can be shortened. can do. In other words, this device is a device that can automatically extract the contours of a human image without the need for human tracing work by incorporating the features of the image to be contour-extracted to the maximum and using it for initial setting of the contour model. Become.

The present invention is not limited to the above-mentioned embodiment, and can be carried out by appropriately modifying it within the scope of not changing the gist thereof. For example, in the explanation in FIG. The neighboring areas are not limited to the eight areas indicated by black circles, and may be wider or narrower.

Furthermore, the present invention can be used not only for the contour extraction of a person, but also for the contour extraction of an image in which a background region and an image region of an object are almost fixed. And
Of course, if the image is a target image with contours such as photographs and drawings,
The contour can be extracted. Further, although the shape of the contour model by the contour model initial position setting device is a rectangle in the embodiment, it may be set to an arch shape or a mountain shape.

[0067]

As described above in detail, according to the present invention,
It is possible to provide an image contour extraction device capable of automating the initial setting of a contour model and surely converging the default contour model in a short time to perform contour extraction.

[Brief description of drawings]

FIG. 1 is a block diagram showing the overall configuration of an embodiment of the present invention.

FIG. 2 is a diagram for explaining the operation of the present invention and is a diagram for explaining a method of determining the positions of both end points of the contour model.

FIG. 3 is a diagram for explaining the operation of the present invention and is a diagram for explaining a method of determining the positions of both end points of the contour model when the image is biased.

FIG. 4 is a diagram for explaining the operation of the present invention, showing an example of setting an initial contour model.

FIG. 5 is a diagram for explaining the operation of the present invention and is a diagram showing another setting example of the contour model at the initial stage.

FIG. 6 is a diagram for explaining the operation of the present invention and is a diagram for explaining a process of obtaining a moving position of a point sequence by convergence calculation.

FIG. 7 is a diagram for explaining the operation of the present invention, showing how the contour model is converged.

FIG. 8 is a diagram for explaining an example of an area calculation method of a contour model.

FIG. 9 is a diagram for explaining a state in which an object to be contour-extracted and a contour model at an initial stage are set by manual plotting.

[Explanation of symbols]

 101 ... Frame memory for person image 102 ... Contact position extraction circuit 103 ... Contour model initial position setting circuit 104 ... Point sequence position memory 105 ... Contour model area calculation circuit 106 ... Convergence calculation device

Claims (1)

[Claims] 1. An image storage means for storing an image of a predetermined composition which is a target of contour extraction, and a contact point for scanning the peripheral edge of the image to obtain the coordinates of two points where the contour extraction target of the image and the background are in contact with each other. Extraction means, initial contour model acquisition means for obtaining an initial contour model having a shape along a line surrounding a predetermined range with the coordinate positions of the two points obtained by the contact extraction means as endpoints, and a series of points representing the contour model. It is characterized by further comprising: a contour model storage means for storing a position; and a convergence calculation means for obtaining and moving the moving position of the point sequence so that the obtained contour model converges on the contour of the contour extraction target. Image contour extraction device.