JPH10136258A - Outline extraction device/method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To uniformly detect the intensity of an edge without depending on the side of a color difference by normalizing the intensity of the edge by means of the section area of the edge. SOLUTION: A gradient vector calculation part 7 receives outline candidate area information S3 in an edge intensity calculation circuit 21 and receives an edge position (e) and an edge direction (d) in an integrating route calculation circuit 22. The edge intensity calculation circuit 21 generates an edge intensity picture Ii based on outline candidate area information S3 as an edge intensity picture generation means. The integrating route calculation circuit 22 calculates an integrating route P based on the edge intensity picture Ii, the edge position (e) and the edge direction (d) as an integrating route calculation means. A normalized coefficient calculation circuit 23 calculates a normalized coefficient S based on the edge intensity picture Ii and the integrating route P. A normalization circuit 24 normalizes the edge intensity picture Ii by dividing the edge intensity picture Ii by the normalized coefficient S. Thus, the edge intensity picture Ii is normalized and it is transmitted to an outline route probe part as an edge intensity picture Io.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[Table of Contents] The present invention will be described in the following order. BACKGROUND OF THE INVENTION Problems to be Solved by the Invention (FIG. 18) Means for Solving the Problems Embodiments of the Invention (FIGS. 1 to 17) (1) Overall Configuration (FIG. 1) (2) Basic principle of the present invention (FIGS. 2 and 3) (3) First embodiment (FIGS. 4 to 9) (3-1) Basic configuration of gradient vector calculating section (FIG. 4) (3-2) Integration path Configuration of calculation circuit (FIGS. 5 to 7) (3-3) Configuration of normalization coefficient calculation circuit (FIG. 8) (3-4) Operation and effect of first embodiment (FIG. 9) (4) Second implementation Example (FIGS. 10 and 11) (5) Third Embodiment (FIGS. 12 to 14) (6) Fourth Embodiment (FIGS. 15 and 16) (7) Other Embodiments (FIG. 17) Effects of the Invention

[0002]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a contour extracting apparatus and a contour extracting method, and more particularly to a contour extracting apparatus and a contour extracting apparatus for extracting a contour of an object from an image in a special effect processing in producing a video such as a television or a movie. It is suitable for application to a method.

[0003]

2. Description of the Related Art Conventionally, in this kind of special effect processing, accuracy in extracting an arbitrary partial image from an image is important. In this extraction work, processing such as edge detection, area detection, and association with a known object is necessary. Particularly, when the background in the image or the target object is complicated, the edge forming the outline of the target object is required. Need to be detected and tracked with high accuracy.

Here, the edge detection is a process of detecting a portion where a pixel value changes rapidly in a grayscale image (a process of detecting a color change). Usually, such a sharp change in the pixel value occurs at the contour of the object, so that the contour of the object can be extracted from the image based on the result of edge detection. This contour extraction is a basic and important technique used in object recognition, object tracking, image synthesis, and the like. Therefore, it can be said that edge detection is the most basic technique for obtaining information on an object from an image.

For this edge detection, a method using a first derivative of an image, a method using a second derivative of an image, and various other methods have been proposed (“Image Analysis Handbook” (Mikio Takagi, Hiroo Shimoda, Tokyo, Japan)). University Press, 1991).

An edge detection method using the first derivative of an image calculates a magnitude of a color change (a degree of a change in a pixel value, hereinafter, referred to as an edge intensity) by a differential operation, and calculates a magnitude of the differential value. This is a method of making edge strength, and is often used as the most basic method. Where the first derivative of the image I is

(Equation 1) Can be calculated by calculating the gradient G defined by Here, u and v are unit vectors in x and y directions, respectively, dI (x, y) / dx, d
I (x, y) / dy represents partial differentiation of the image I in the x and y directions, respectively. The gradient is defined as a target pixel (x,
represents the gradient of the image in y).

As shown by the equation (1), the gradient G is a vector. The edge intensity g uses the magnitude of the gradient vector, and an image representing the edge intensity for each pixel is referred to as an edge intensity image g (x, y). This edge intensity image g (x, y) is expressed by the following equation.

(Equation 2) Can be represented by Usually, since an image is given as an array of discrete data, a method of approximately obtaining a gradient using a local difference operator or the like is used for actual gradient calculation.

After calculating the edge intensity image g, a pixel corresponding to the contour position is detected, and a process of generating a binarized image from the edge intensity image g (hereinafter referred to as edge point detection process). May be performed. Normally, binarization by threshold processing is used for edge point detection processing. That is, since the edge point image E is clearly judged to be an edge point, the following equation is obtained by a fixed threshold value T in which a sufficient edge intensity is set.

(Equation 3) Is represented as

By the way, the threshold value for edge point detection needs to satisfy two conditions that the detected edge point has no break as a contour and does not include a noise point that is not a contour. However, these two conditions are inconsistent with respect to the magnitude of the threshold value,
The fixed threshold value as described above cannot cope with the problem, and there is a problem that the edge intensity becomes unstable or causes noise.

As one method for solving such a problem, a method has been proposed in which a threshold value is variable and an appropriate value is obtained for each area. In the edge point detection using the variable threshold value, the threshold value is set based on the statistical properties of surrounding values.

For example, the document "Foreground / background Se
gmentation of Color Images by Integration of Multi
ple Cues "(Huang et al., Proceedings of ICIP 199
5, Vol.I, pp.246-249, 1995), the threshold setting method calculates the average value and standard deviation of edge strength and sets the upper limit of the noise level to three times the standard deviation. Things. According to this method, the edge point is detected when the edge intensity is detected to be larger than the average by three times the standard deviation. Since what can be predicted from the surrounding values is the magnitude of the random noise level, the fluctuation of the base level, and the like, this method is an effective method for removing noise.

[0012]

However, the conventional edge detection using the first derivative has the following problems. That is, the conventional edge detection method utilizes the property that the color change is large at the contour position. However, the magnitude of the color change in the contour depends on the color difference between the colors of the objects on both sides of the contour.

Here, the effect of the change in the color difference on the edge intensity will be described with reference to FIG. FIG. 18A shows a feature of an example of an image in which an edge exists. FIG.
(B) shows a cross section of an image at two line segments l0 and l1. FIG. 18C shows a cross section of the gradient along two line segments l0 and l1. FIG. 18D is the same as FIG.
FIG. 4 shows characteristics of the gradient intensity image obtained from FIG.

An edge is present at the center of the image shown in FIG. 18A, and a left area of the edge has a color F.
It is assumed that the area on the right side of the edge has two areas having colors B0 and B1 respectively above and below. In FIG. 18B, the change from the color F area to the color B0 area is represented by a cross-section along the line l0, and the color F area is changed to the color B1 area in order to easily explain the color change of the image. To the line segment l
Represented by 1.

Here, the colors in the two line segments are the same as the color F in the left region of the edge, but have different colors B0 and B1 in the right region of the edge, so as shown in FIG. , The slopes at the edge positions are not equal. Further, since the gradient intensity indicates the magnitude of the gradient of the color change, there is a difference in the magnitude of the gradient intensity at the edge position, as shown in FIG. Accordingly, as shown in FIG. 18 (D), the obtained gradient intensity image is one in which the intensity value changes in the middle of the image, although it is a continuous edge.

As described above, in the conventional edge detection, although the edge exists continuously along a certain object, the edge intensity varies depending on the magnitude of the color difference, so that the edge intensity is detected uniformly. And the contour of the object cannot be stably extracted from the image.

In the edge point detection processing using the variable threshold value, the method focuses on random noise and fluctuations in the base level. Therefore, the fluctuations in the edge intensity itself due to the above-described color difference cause the edge intensity to be detected to be the noise level. When the camera approaches, the edge point cannot be detected, and the edge point may be interrupted. In particular, when extracting the shape of the contour, the edge points must not be interrupted, so it is necessary to perform edge point connection processing after edge point detection processing, and as a result, the edge detection processing becomes complicated. Was.

The present invention has been made in view of the above points, and has as its object to propose a contour extraction device and a contour extraction method capable of stably extracting a contour of an object from an image.

[0019]

According to the present invention, there is provided a contour candidate for an outline of an object determined based on the position and direction of an edge inputted by a predetermined input means. An edge intensity image is generated by calculating a gradient vector in the region, an integration path crossing the edge is calculated based on the edge intensity image, the edge position and the edge direction, and the edge intensity image is integrated on the integration path. The value is output as a normalization coefficient, and the edge intensity image is normalized based on the edge intensity image and the normalization coefficient. According to the present invention, since the edge intensity is standardized by the cross-sectional area of the edge, it is possible to uniformly detect the edge intensity without depending on the magnitude of the color difference in the input image.

In the present invention, a contour candidate area for a contour of an object is determined based on a plurality of line segments along the edge inputted as information on the position of the edge and the direction of the edge. Generates a mask image specifying a neighboring area, outputs each mask image and each line segment as an edge position and an edge direction, calculates a gradient vector in a contour candidate area, generates an edge intensity image, and generates an edge intensity. Based on the image, the corresponding mask image and the line segment, for each of the corresponding mask image and line segment, an integration path crossing each line segment is calculated, and the value obtained by integrating each mask image on each corresponding integration path is defined as a standard. Are output as standardized coefficients, each mask image is standardized based on the corresponding mask image and each standardized coefficient, and the corresponding standardized mask image is output. And using each mask image to generate a edge strength image is normalized. Further, according to the present invention, the contour of the object is divided by each segment of the polygonal line, and the edge strength is standardized by the cross-sectional area of the edge for each region, so that the shape of the contour of the object is complicated. Even when the direction of the contour changes significantly, the edge intensity can be detected uniformly without depending on the magnitude of the color difference in the input image.

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below in detail with reference to the drawings.

(1) Overall Configuration In FIG. 1, reference numeral 1 denotes a key signal generating apparatus to which the present invention is applied as a whole, and generates soft keys of objects existing in a plurality of continuous images. The key signal generation device 1
It is composed of a contour extraction unit 2 and a key calculation unit 3,
Based on the contour curve of the object calculated by the contour extraction unit 2 and the gradient vector of the outline of the object, the key calculation unit 3 generates soft keys of the object.

The key signal generating apparatus 1 is a line signal, polygonal line information, curve information, or a thick stroke drawn by an operator using a mouse 4 as an input means. Mask image information S1
Is received by the estimated contour calculation unit 5 of the contour extraction unit 2. The estimated contour calculation unit 5 serves as an outline candidate area determination unit in accordance with an estimated shape of the outline of the target object (hereinafter, referred to as an estimated outline) based on line segment information, broken line information, curve information, or mask image information S1. The estimated contour information S2 is obtained, and the estimated contour information S2 is sent to the contour candidate area determination unit 6.

The contour candidate area determination section 6 calculates a region where a contour is considered to exist (hereinafter, referred to as a contour candidate area) based on the estimated contour information S2, as a contour candidate area determining means, and also executes edge detection. The position and the direction of the edge are calculated, and the contour candidate area information S3 corresponding to the contour candidate area, the edge position e as the edge position information, and the edge direction d as the edge direction information are sent to the gradient vector calculation unit 7. .

The gradient vector calculator 7 calculates a gradient intensity value in the contour candidate area based on the contour candidate area information S3, thereby forming a gradient vector of the contour (hereinafter, this is referred to as an edge intensity image).
Generate Ii. The gradient vector calculation unit 7
By standardizing the edge intensity image Ii based on the edge intensity image Ii, the edge position e, and the edge direction d, the standardized edge intensity image Io is transmitted to the contour path searching unit 8 and the key calculating unit 3. Alternatively, the edge point binary image Ie obtained by subjecting the standardized edge intensity image Ii to threshold processing using a predetermined threshold value is sent to the contour path searching unit 8 and the key calculating unit 3. .

The contour path searching section 8 serves as a contour path searching means, based on the edge intensity image Io or the edge point binary image Ie, a list of pixels passing through the center of the outline (hereinafter, referred to as a list).
This is used to generate a contour coordinate list, and contour coordinate list information S4 corresponding to the contour coordinate list is sent to the curve generating unit 9. The curve generation unit 9 generates a contour curve by performing curve approximation on a contour obtained based on the contour coordinate list information S4 as a curve generation unit, and outputs the contour curve information S5 corresponding to the contour curve to the key calculation unit 3. Send out.

Here, in the key signal generation device 1, apart from the method of obtaining the estimated contour information S2 from the line segment information, broken line information, curve information or mask image information S1, the image i- which is the immediately preceding frame is obtained. The estimated contour information S2 in the current image i can be obtained by estimating where the contour curve of the image i-1 has moved in the current image i by the motion vector estimation processing using the one contour curve information S5. It has been made possible.

That is, the motion vector estimating section 10 serves as a contour estimating means, for the previous image i-1, a contour candidate area determining section 6, a gradient vector calculating section 7, a contour path searching section 8, and a curve generating section 9. Is used to estimate where the contour curve corresponding to the contour curve information S5 has moved in the current image i.
The result is sent to the contour candidate area determination unit 6 as estimated contour information S2. In this case, the processes from the contour candidate area determination unit 6 to the curve generation unit 9 are performed in the same manner as described above. Accordingly, when the motion vector estimating unit 10 is selected as a method for obtaining the estimated contour information S2, the curve generating unit 9 feeds back the contour curve information S5 to the motion vector estimating unit 10 via the delay circuit 11. I have.

The key calculating section 3 receives the standardized edge intensity image Io and the contour curve information S5 at the both end curve calculating section 12. The both-ends curve calculation unit 12 calculates the width between two curves representing both ends of the outline (hereinafter, referred to as both-ends curves) based on the edge intensity image Io and the outline curve information S5, and calculates the width between the both-ends curves. Is sent to the soft key generation unit 13 in accordance with the width of both ends. Soft key generator 13
Generates a curved surface whose height smoothly changes from "0" to "1" between two end curves based on both end curve information S6, and uses the height of this curved surface as a soft key value in an image. The writing process is performed to calculate and output the soft keys.

Thus, the key signal generating apparatus 1 can generate a soft key of an object existing in an image.

(2) Basic Principle of the Present Invention The present invention calculates the magnitude of the color difference and sets the calculated magnitude of the color difference to "1" in order to eliminate the influence of the variation of the gradient intensity due to the color difference. Perform normalization processing. The theoretical background of this normalization processing will be described below.

First, the gradient G * not affected by the color difference is a gradient obtained from an image in which the color difference is ideally constant everywhere in the outline, that is, the combination of colors in the regions on both sides of the outline is determined by the outline of the outline. It is assumed that the gradient is obtained from the image I * which does not change even at the place. Here, the magnitude of the color difference is “1”.
Generality is not lost.

The relationship between the image I * and the image I having a color difference is given by the following equation.

(Equation 4) Can be represented by Here, C represents the color difference of the outline of the image I. Since the color difference C is a color difference between the regions on both sides of the contour, the color I (p (a)) of a certain pixel p (a) in one region of the image I and a certain color in the other region are bordered by the contour. The color difference C can be represented by the color I (p (b)) of the pixel p (b). That is, the color difference C
Is

(Equation 5) Can be represented by

Here, p (a) and p (b) must not be pixels spanning the contour. Therefore,

(Equation 6) The following conditional expression can be derived. The color difference C in the equation (5) can be calculated by the following equation by using definite integration.

(Equation 7) Can be expressed as

Here, ∫ (P) represents the integral in the path P on the image, and P: p (t) is the pixel p (a) to the pixel p (a).
This is an appropriate route to (b). Therefore, when the equations (4) and (7) are converted to the equation for the gradient G * to be obtained, the following equation is obtained.

(Equation 8) Can be expressed as Accordingly, by calculating the gradient G of the image I and determining the path P and calculating the equation (8), it is possible to obtain the gradient G * which is not affected by the color difference.

Further, utilizing the fact that the direction of the gradient vector in the contour is perpendicular to the edge forming direction, and setting a path P so as to vertically cross the edge, the following equation (8) is obtained.

(Equation 9) Can be simplified as follows. That is, this (9)
The formula is the gradient intensity g as seen across the edge.
Is standardized so that the cross-sectional area of “1” is “1”.

Here, an outline of the calculation method of the normalization processing is shown in FIG.
This will be described with reference to FIG. FIG. 2A schematically illustrates an image having an edge at the center, and a broken line in the figure indicates the edge. Further, it is assumed that a left area has a color F and a right area has a color B with an edge as a boundary. FIG. 2B shows a color change when a cross section is taken along a line segment P of this image.

First, as a first operation, a path P that vertically crosses an edge is set. Next, the gradient intensity on the path P is determined. Fig. 2 shows the obtained gradient intensity.
It is shown in (C). The gradient intensity has a value at the edge, and has no value except at the edge. Therefore, points where the gradient intensity is "0" are detected on both sides of the base of the edge so as to satisfy the condition of the expression (6), and these points are defined as p (a) and p (b), respectively. Next, an area S (corresponding to a standardization coefficient described later) formed by the gradient intensity surrounded by p (a) and p (b) is obtained. This area S is a calculation result of the denominator on the right side of the equation (9). Finally, the gradient intensities of all pixels obtained separately are calculated using (9)
By dividing by the area S according to the formula, a standardized gradient intensity g * can be obtained.

FIG. 2D schematically shows an image obtained as a result. Actually, an image whose value changes smoothly, but a region other than a region (area S) where the gradient intensity is formed is shown by a blank. In order to show the difference between the gradient g and the gradient g *, the gradient g * on the route P is shown in FIG. The gradient intensity changes so that the area across the edge becomes “1”.

Further, the present invention uses a standardized gradient to realize edge point detection without edge discontinuity, so that the peak value of the gradient intensity at the edge position always exceeds the threshold value. Set the value. This will be described with reference to FIG.

FIG. 3A schematically shows an image having an edge at the center. In the figure, a broken line indicates an edge.
It is assumed that a left area has a color F and a right area has a color B with an edge as a boundary. FIG. 3B shows the normalized gradient intensity g * on the path P obtained by the above-described gradient normalization calculation. The cross-sectional area S of the gradient strength g * in the direction crossing the edge is “1”. According to the threshold value setting method of the present invention, the area is S,
A rectangle whose two vertices on the bottom surface are p (a) and p (b) is set, and the height T of the rectangle is set as a threshold (FIG. 3).
(C)).

In general, the cross section of the gradient intensity in the edge shows a mountain-like shape in which the shoulder is gradually lowered due to the influence of a filter or blurring of an object at the time of imaging. Therefore, by setting the threshold value according to the present invention, the peak value of the gradient intensity in the edge is always larger than the threshold value T, so that the edge point can be detected.
As a result, interruption of the edge point can be avoided.
The edge point binary image obtained in this way is shown in FIG.
Is shown schematically in FIG.

(3) First Embodiment (3-1) Basic Configuration of Gradient Vector Calculator FIG. 4 shows the basic configuration of the gradient vector calculator 6 according to the first embodiment of the present invention. The gradient vector calculation unit 6 outputs a normalized edge intensity image Io by performing a normalization process for setting the magnitude of the color difference to “1”.

The gradient vector calculation section 6 receives the contour candidate area information S 3 by the edge strength calculation circuit 21, and receives the edge position e and the edge direction d by the integration path calculation circuit 22. The edge intensity calculation circuit 21 generates an edge intensity image Ii based on the contour candidate area information S3 as an edge intensity image generating means, and outputs the edge intensity image Ii.
i is sent to the integration path calculation circuit 22, the normalization coefficient calculation circuit 23, and the normalization circuit 24.

The integration path calculation circuit 22 serves as integration path calculation means based on the edge intensity image Ii, the edge position e, and the edge direction d, as will be described later.
Is calculated, and the integration path P is sent to the normalization coefficient calculation circuit 23. Where edge position e
And the edge direction d are given as a vector in the image space (x, y).

As described above, this integration path P can theoretically use a curve segment having an arbitrary shape that crosses the edge. In the subsequent integration value calculation processing, a vector indicating the direction of the integration path and an integration path are used. Dot product with the gradient vector on the path (corresponding to the denominator on the right side of equation (9))
, The output of the integration path calculation circuit 22 can be made a curve. However, as described above, the calculation can be simplified by using a path that vertically crosses the edge. Therefore, a case in which an integration path using a line segment is used will be described here.

The normalization coefficient calculating circuit 23 calculates a normalization coefficient S by a normalization coefficient calculation process described later based on the edge intensity image Ii and the integration path P, as a normalization coefficient calculation means. S is sent to the normalization circuit 24. The normalizing circuit 24 normalizes the edge intensity image Ii by dividing the edge intensity image Ii by the standardization coefficient S (corresponding to the equation (9)) as a standardizing means, and as a standardized edge intensity image Io. It is sent to the contour route search unit 8.

(3-2) Configuration of Integration Path Calculation Circuit The configuration of the integration path calculation circuit 22 is shown in FIG. 5, and the procedure of the integration path calculation processing is shown in FIG. The integration path calculation circuit 22 starts the integration path calculation process from step SP1, and in step SP2, calculates the edge intensity image Ii as a traversing line calculation circuit 22A as a traversing line calculation means and an integration path end point calculation as an integration path end point calculation means. Circuit 22
B, and in step SP3 and step SP4, the edge position e and the edge direction d are received by the transverse straight line calculation circuit 22A.

Next, in step SP5, the integration path calculation circuit 22 calculates a cross line c perpendicular to the edge based on the edge position e and the edge direction d in the cross line calculation circuit 22A. Here, dn represents a vector perpendicular to the edge direction d, and k represents a scalar.

Subsequently, in step SP6, the integration path calculation circuit 22 determines the edge strength maximum point pmax at which the edge strength is the maximum on the traversing line c, by using the integration path end point calculation circuit 22.
After the calculation at B, in step SP7, the edge intensity is smaller than the edge intensity at the edge intensity maximum point pmax on one side of the half-line starting from the edge intensity maximum point pmax on the transverse line c, and The point p0 at which the value is equal to or smaller than the threshold value is detected by the integration path end point calculation circuit 22B. Here, in step SP7, "WH
"ILE A {B}" means "process B while condition A is satisfied.
H "is a positive scalar smaller than" 1 ".

Next, at step SP8, the integration path calculation circuit 22 determines the maximum edge intensity point pma on the transverse line c.
A point p1 at which the edge intensity is smaller than the edge intensity at the edge intensity maximum point pmax and equal to or less than a predetermined threshold value on the half line on the opposite side of the half line at step SP7 starting from x is calculated as an integral path end point. Circuit 22B
Is detected at After that, the integration path calculation circuit 22
At step SP9, the points p0 and p1 are sent out from the integration path end point calculation circuit 22B to the normalization coefficient calculation circuit 23 as the integration path P, and the integration path calculation processing ends at step SP10. Accordingly, the integration path P is calculated as a line segment on the crossing line c divided by the integration section of the points p0 and p1.

Here, in the above-mentioned steps SP7 and SP8, when the integration path end point calculating circuit 22B calculates the edge strength of a point on the transverse line c, the position of a point on the transverse line c is not necessarily the pixel grid of the image. In the present invention, the integration path end point calculation circuit 22B receives the current image and the real number coordinates in order to obtain the edge intensity at a position not on the pixel grid. Image interpolation means for outputting image values (shown)
have.

The image interpolation processing in the image interpolation means will be described with reference to a flowchart shown in FIG. The image interpolation means starts the image interpolation processing from step SP1, and receives the edge intensity image Ii and the real number coordinates (xf, yf) at step SP2 and step SP3, respectively.

Subsequently, in step SP4, the image interpolation means defines a weight function of the interpolation filter. Here, as the weighting function of the interpolation filter, "Digital image processing for image understanding (I)" (Junichiro Toriwaki, Shokodo) and "Image analysis handbook" (Mikio Takagi, Hiroo Shimoda, University of Tokyo Press, 1991) ) Can be used. For example,

(Equation 10) There is a function as shown in Here, x ** y represents x to the power of y.

Next, the image interpolation means generates a function Ic (x, y) by calculating convolution of the edge intensity image Ii and the weight function in step SP5, and then generates the function Ic (xf, yf) in step SP6. The pixel value is output as a pixel value on the real number coordinates, and the image interpolation processing ends in step SP7.

(3-3) Configuration of Normalization Coefficient Calculation Circuit The normalization coefficient calculation processing in the normalization coefficient calculation circuit 23 will be described with reference to a flowchart shown in FIG. The normalization coefficient calculation circuit 23 starts the normalization coefficient calculation processing from step SP1, and receives the edge intensity image Ii and the integration path P at step SP2 and step SP3, respectively. Here, as described above, the integration path P is a line segment that can be represented as the end points p0 and p1.

Subsequently, in step SP4, the normalization coefficient calculation circuit 23 changes the end point p0 to the end point p1 of the integration path P.
The edge intensity values at arbitrary plural points up to are calculated, and the sum S of the edge intensity values at each of these points is calculated. Here, the accuracy of the sum S of the edge intensity values can be improved by reducing the interval s between the edge intensity value calculation positions (that is, the positions of the points from the end point p0 to the end point p1). In this case, if the edge intensity value at each point is calculated using the above-described image interpolation means, the end point p0 to the end point p1
Of these points, the edge intensity value at a point not on the pixel grid can be calculated with high accuracy. Next, in step SP5, the normalization coefficient calculation circuit 23 sends the sum S of the edge intensity values to the normalization circuit 24 as the normalization coefficient S, and ends the normalization coefficient calculation processing in step SP6.

(3-4) Operation and Effect of First Embodiment In the above configuration, the gradient vector calculation unit 7
Starts the normalization processing of the edge intensity image Ii shown in FIG. 9 from step SP1, and proceeds to step SP2 and step SP2.
After obtaining the edge intensity image Ii, the edge position e, and the edge direction d in Step 3 and Step SP4, respectively, the integration path P is obtained based on the edge intensity image Ii, the edge position e, and the edge direction d in Step SP5.

Subsequently, the gradient vector calculating section 7
Calculates the normalization coefficient S based on the edge intensity image Ii and the integration path information P in step SP6.
In step SP7, the standardized edge intensity image Io is calculated by dividing the edge intensity image Ii by the normalization coefficient S (calculating the equation (9)), and the process proceeds to step SP8.
In step SP9, the edge intensity image Io is sent to the contour path searching unit 8, and in step SP9, the edge intensity image Ii is normalized.

Accordingly, the gradient vector calculation unit 7 normalizes the edge strength by the cross-sectional area of the edge.
The edge intensity can be detected uniformly without depending on the magnitude of the color difference in the input image. Further, since the gradient vector calculation unit 7 calculates the cross-sectional area of the edge so as to cross the edge vertically, the edge intensity image Ii can be standardized by a simple method.

Further, the gradient vector calculator 7 calculates edge points p0 and p1 forming the integration path P, and calculates edge intensity values at a plurality of arbitrary points from the end point p0 to the end point p1 of the integration path P. In the calculation, the image value on the real number coordinates is calculated, and the edge intensity value at each point is calculated based on the pixel value on the real number coordinates. Therefore, the normalization coefficient S can be detected with high accuracy. Thus, the edge intensity image Ii can be accurately standardized.

According to the above configuration, the integration path P crossing the edge in the vertical direction is calculated, and the sum S of the edge intensity values at a plurality of arbitrary points on the integration path P is determined by the standardization coefficient S
By dividing the edge intensity image Ii by the normalization coefficient S to normalize the edge intensity image Ii, the edge intensity can be normalized by the cross-sectional area of the edge. Edge intensity can be detected uniformly without depending on the edge intensity. Thus, a contour extracting apparatus and a contour extracting method capable of stably extracting the contour of an object from an image can be realized.

(4) Second Embodiment FIG. 10 shows the configuration of the gradient vector calculation unit 30 according to a second embodiment of the present invention, in which parts corresponding to those in FIG. The threshold value calculation circuit 31 receives the normalization coefficient S from the normalization coefficient calculation circuit 22 as the threshold value calculation means, and transmits the normalization coefficient S to the length of the integration path P (integration section (points p0 and p1 The value obtained by the division in the period is sent to the edge point detection circuit 32 as the threshold value T.

The edge point detecting circuit 32 receives the standardized edge intensity image Io from the normalizing circuit 24 as edge point detecting means, and uses the threshold value T at each position in the edge intensity image Io. The edge point bit map is calculated by performing threshold processing on the intensity, and the edge point bit map is sent to the contour path searching unit 8 as the edge point binary image Ie.

In the above configuration, the gradient vector calculation unit 30 starts the edge point binary image calculation processing shown in FIG. 11 from step SP1, and proceeds to step SP2.
After receiving the integration path P and the normalization coefficient S in step SP3, the threshold value T is calculated by dividing the normalization coefficient S by the length (integration section) of the integration path P in step SP4.

Next, the gradient vector calculation unit 30
Receives the standardized edge intensity image Io at step SP5, and receives the threshold value T at step SP6.
Is used to calculate the edge point binary image Ie by performing threshold processing on the edge intensity at each position in the standardized edge intensity image Io, and at step SP7, the edge point binary image Ie Is output, and the edge point binary image calculation processing is terminated in step SP8. Here, in step SP6, "For each A {B}"
This means "execute process B for each of all A".

Here, the gradient vector calculating section 30 sets a rectangular shape whose area is S (normalization coefficient) and whose bottom two vertices are p0 and p1, and determines the height of this rectangular shape. Since the threshold value T is used, the peak value of the gradient intensity at the edge can always be made larger than the threshold value T. Therefore, the edge point can be always detected, so that interruption of the edge point can be reliably avoided.

According to the above configuration, each position in the standardized edge intensity image Io is determined using the value obtained by dividing the standardization coefficient S by the length of the integration path P as the threshold value T. Since the threshold value processing is performed on the edge intensity in (1), the peak value of the gradient intensity in the edge can always be made larger than the threshold value T, so that the edge point can always be detected. As a result, an edge point binary image Ie without edge points can be obtained, and thus a contour extraction device and a contour extraction method capable of stably extracting a contour of an object from an image can be realized. .

(5) Third Embodiment FIG. 12 shows the configuration of the gradient vector calculating section 40 according to a third embodiment of the present invention, in which the same reference numerals are given to the portions corresponding to FIGS. The gradient vector calculation unit 40 inputs a continuous polygonal line along the edge from the input means 4 to provide information on the edge position e and the edge direction d, thereby extracting the contour of an object having a complicated shape. It is made to be able to correspond to.

The contour candidate area determination unit 6 defines each line segment of the polygonal line L as a line segment l _j based on the estimated outline information S2 composed of a polygonal line supplied from the estimated outline calculation unit 5, and
The mask image Im _j that specifies the neighboring pixels of the line segment l _j as the region near the line segments generated for each segment l _j, as a counter and a line segment l _j and the mask image Im _j respectively corresponding, in the pair The pairs of the line segment l _j and the mask image Im _j are respectively sent to the gradient vector calculation units 7 provided by the number (n), and each mask image Im _j is transmitted to the image synthesis unit 41.
To send to.

Accordingly, in the case of the third embodiment, the mask image Im _j is sent to the gradient vector calculation section 7 as the edge position e, and the line segment l is set as the edge direction d.
_j is sent to the gradient vector calculator 7. The contour candidate area determination unit 6 calculates a contour candidate area based on the estimated contour information S2, and sends contour candidate area information S3 corresponding to the contour candidate area to each gradient vector calculation unit 7.

In the case of the third embodiment, each gradient vector calculation section 7 is provided with an edge strength calculation circuit 21.
In, based on the line segment l _j and the mask image Im _j ,
After generating the edge strength image Ii _j corresponding to the number of line segments l _j, by performing the normalization processing described above with respect to neighboring pixels of the line segment l _j in the respective edge intensity image Ii _j, normalized The edge intensity image Is _j is generated, and the image edge intensity image Is _j is transmitted to the image combining unit 41.

[0073] That is, in the graph Day entry vector calculation unit 7, the integral path calculating circuit 22, edge intensity image Ii _j, based on the corresponding mask image Im _j and the line segment l _j, integral path crosses the line segment l _j P Is calculated. Normalization coefficient calculation circuit 23 calculates the integrated value in the integration path P corresponding mask image Im _j as normalization factor. Normalization circuit 24 normalizes the edge strength image Ii _j based on the corresponding mask image Im _j and the normalized coefficient,
Edge strength image Is _j image synthesizing unit 4 which is the standardized
Send to 1.

The image synthesizing section 41 includes an edge intensity image Is _j supplied from each gradient vector calculating section 7 and
With each pair of the beauty mask image Im _j corresponding supplied from the outline candidate area decision unit 6, and generates an edge intensity normalized edge intensity image Io. Here, the image synthesizing unit 4
1 generates a standardized edge intensity image Io by synthesizing a pair of an image Is _j and a mask image Im _j using, for example, an image synthesis method generally called Alpha synthesis. The alpha synthesis and edge intensity image Is _j which is standardized, it shall be made by the product sum calculation of the mask image Im _j indicating the weight.

Here, the area dividing process in the contour candidate area determining section 6 will be described with reference to a flowchart shown in FIG. The outline direction area determination unit 6 starts the area division processing from step SP1, and receives the estimated outline information S2 consisting of continuous broken lines L in step SP2. Here, as the polygonal line L, what is created so as to approximate the shape of the outline and to place the vertex at a position where the color difference in the outline changes is input. Each vertex of the polygonal line L has coordinates p0, .., pn-1.

[0076] The contour candidate area determination section 5 Subsequently, in step SP3, after defining a segment l _j each line segment of polyline L, at step SP4, each line segment l _j, all the pixels to "0" to produce a clear and mask image Im _j.
Next, in step SP5, the contour candidate area determination unit 6 calculates the closest line segment lk for all the pixels (x, y) in the original image, and calculates the distance d (lk, x,
Only when y) is smaller than a preset threshold value h, the value of the pixel Im _k (x, y) of the image Im _k is set to “1”. As a result, it is possible to obtain a mask image Im _{k in} which the value of the neighboring pixel of the line segment lk is represented by “1”, and obtain n mask images Im _j corresponding to the line segment l _j .

[0077] The distance d (l _j from where the line segment l _j, x,
y) can be easily calculated because both end points of the line segment are known. Then outline candidate area decision unit 6, and sends at step SP6, the corresponding mask image Im pair of _j and the line segment l _j (n-number) respectively Gras Day entry vector calculator 7 as edge position d and edge direction e a, In step SP7, the area division processing ends.

In the above configuration, the gradient vector calculating section 40 starts the normalizing process shown in FIG. 14 from step SP1, and in step SP2, converts the estimated contour information S2 consisting of the polygonal line L into the contour candidate area determining section 6. At step SP3, the corresponding line segment l _j
And calculating a pair of mask image Im _j.

Subsequently, the gradient vector calculating section 40
, In step SP4, after generating an edge intensity image Ii _j each Gras Day entry vector calculation unit 7, in step SP5, edge intensity image Ii _j, based on the corresponding line segment l _j and the mask image Im _j, standard calculating a pair of the reduction has been edge intensity image is _j and the mask image Im _j.

Next, the gradient vector calculation section 40
Is the standardized image Is _j in step SP6
And using each pair of the mask image Im _j, after generating the normalized edge intensity image Io, in step SP7, and outputs the normalized edge intensity image Io, the normalization process in step SP8 finish.

Accordingly, the gradient vector calculating section 40 divides the contour candidate region of the object by each line segment l _j of the polygonal line L to be inputted, and calculates the neighboring pixels of each line segment l _j in each of the divided regions. Even if the shape of the contour of the object is complicated and the direction of the contour changes greatly, the edge intensity is detected uniformly without depending on the magnitude of the color difference in the input image. Can be.

[0082] According to the above configuration, the edge as the information of the edge orientation 'd' with use of each line segment l _j of沿Tsuta polygonal line L, the mask image to specify the neighboring pixels of each line segment l _j as information edge position e Im generate _j and the corresponding line segment l
generates a normalized edge intensity images Is _j by performing the normalization processing for neighboring pixels of the line segment l _j based on each pair of _j and the mask image Im _j, the corresponding edge intensity images Is _j with each pair of the mask image Im _j, by generating a normalized edge strength image Io, complicated the shape of the contour of the object even when the direction of the contour is greatly changed, the color difference of the input image Edge intensity can be detected uniformly regardless of the magnitude. Thus, a contour extracting apparatus and a contour extracting method capable of stably extracting the contour of an object from an image can be realized.

(6) Fourth Embodiment FIG. 15 shows the configuration of a gradient vector calculation unit 50 according to a fourth embodiment of the present invention, where parts corresponding to those in FIG. 1 and FIG. The gradient vector calculation unit 50 inputs the continuous curve along the edge from the input means 4 to provide information on the edge position e and the edge direction d, thereby obtaining the contour of the object having a more complicated shape. It is designed to support extraction.

The contour candidate area determination section 51 receives the estimated contour information S2, which is a curve supplied from the estimated contour calculation section 5, at the feature point extraction circuit 51A and the curve division circuit 51B. The feature point extraction circuit 51A includes a curve C
, Pn-1 are extracted, and feature point information CP corresponding to the feature points p0, p1,..., Pn-1 is extracted.
To the curve dividing circuit 51B.

Here, the characteristic point is a point showing a unique characteristic on the curve, and is, for example, a bending point at which the shape of the curve is bent or a color change point at which the color of the image changes on the locus of the curve.
The curve dividing circuit 51B serves as a curve converting means,
Is approximated to a continuous broken line L along an edge having each feature point p0, p1,..., Pn-1 as a vertex, and the broken line L is sent to the contour candidate area determination unit 6.

In the above configuration, the gradient vector calculation section 50 starts the normalization processing shown in FIG. 16 from step SP1, and in step SP2, converts the estimated contour information S2 consisting of the curve C into the contour candidate area determination section 51. At the characteristic point extraction circuit 51A and the curve division circuit 51B. Subsequently, in step SP3, the gradient vector calculation unit 50 determines the characteristic points p0, p2,.
, Pn-1 are detected by the feature point extraction circuit 51A.

Next, the gradient vector calculation unit 50
.., Pn−1 in the curve dividing circuit 51B in each of the characteristic points p0, p1,.
Is approximated to a polygonal line L having a vertex as a vertex, and the polygonal line L is sent to the contour candidate area determination unit 6 in step SP5.
Subsequently Gras Day entry vector calculating unit 50, calculates at step SP6, and each line segment l _j fold line L, and each pair of the mask image Im _j that specifies the neighboring pixels of the line segment l _j contour candidate area decision unit 6 I do.

Subsequently, the gradient vector calculating section 50
, In step SP7, after generating each graphene Day entry vector calculating unit 7 the edge strength image Ii _j, at step SP8, edge intensity image Ii _j,
Based on the line segment l _j and the mask image Im _j , a pair of the normalized edge intensity image Is _j and the mask image Im _j is calculated. Next, the gradient vector calculation unit 50
In step SP9, using the pair of the normalized edge intensity image Is _j and the mask image Im _j, after generating the normalized edge intensity image Io, step SP9
In step S10, the standardized edge intensity image Io is output, and the normalization process ends in step SP10.

Accordingly, the gradient vector calculating section 50 converts the input curve C into a polygonal line L having vertices at characteristic points p0, p1,...
, The outline of the object is divided by each line segment l _j of the polygonal line L, and the normalization process is performed on the pixels near each line segment l _j , so that the shape of the outline of the object is Even in the case where the direction of the outline changes greatly due to the complexity, the edge intensity can be detected uniformly without depending on the magnitude of the color difference in the input image.

According to the above configuration, as information in the edge direction d, a polygonal line having the vertices of the characteristic points p0, p1,..., Pn-1 which are the inflection points and / or color change points of the curve C along the edge. Using each line segment l _{j of} L, the edge position e
And the information generating mask image Im _j that specifies the neighboring pixels each line segment l _j fold line L, with respect to neighboring pixels of the corresponding line segment l _j and the line segment l _j based on each pair of the mask image Im _j generates edge intensity image is _j which is standardized for each pair by performing normalization process Te, with each pair of the corresponding edge intensity image is _j and the mask image Im _j, the normalized edge By generating the intensity image Io, even when the shape of the outline of the object is more complicated and the direction of the outline changes greatly, the edge intensity can be detected uniformly without depending on the magnitude of the color difference in the input image. . Thus, a contour extracting apparatus and a contour extracting method capable of stably extracting the contour of an object from an image can be realized.

Further, according to the above configuration, since the inflection point and the color change point are extracted as the characteristic points of the curve C, the area can be divided for each combination of colors constituting the contour. As a result, the normalization processing of each edge intensity image Ij can be performed with higher accuracy, and the edge intensity can be detected more uniformly without depending on the magnitude of the color difference in the input image.

(7) Other Embodiments In the first, second, third and fourth embodiments, the case where the operator inputs a line segment, a polygonal line, or a curve using the input means 4 has been described. However, the present invention is not limited to this. As described above, the current image i, the image i-1 one frame before the contour curve has already been obtained as the contour curve information S5, and the contour of the image i-1 The estimated contour information S2 of the current image i may be obtained based on the curve information S5 (FIG. 1).

In this case, it is not necessary to input the edge position e and the edge direction d for each image, so that the outline of the object can be extracted from the image by a simple method. In particular, when contours are extracted from a moving image, the work for inputting the edge position e and the edge direction d for each frame can be omitted, so that the work efficiency can be significantly improved.

Here, as the motion vector detection, for example, “SNAKES: Active Contour models” (Kass M., Wit
ikin A., Terzopoulos D., Proc. 1st ICCV, pp259-268
1987), an active contour model called "SNAKES", and an image processing system (Japanese Patent Laid-Open No. 4-117079).
And motion vector detection by matching, which can be used.

In the first, second, third and fourth embodiments described above, the case where a mouse is used as an input means for inputting a line segment, a polygonal line, or a curve for specifying the edge position e and the edge direction d is described. As described above, the present invention is not limited to this, and a line segment, a polygonal line, or a curve may be input using a tablet or the like.

Further, in the third and fourth embodiments described above, a case where a plurality of gradient vector calculation units 7 are used has been described. However, the present invention is not limited to this, and a configuration in which one gradient vector calculation unit 7 is used You may make it. In this case, the configuration of the gradient vector calculation units 40 and 50 can be simplified.

In the third and fourth embodiments, the case where the gradient vector calculation section 7 is used has been described. However, the present invention is not limited to this, and the gradient vector calculation section 7 described in the second embodiment may be used. The unit 30 may be used. In this case, edge point detection circuit 32 detects the edge points in the vicinity of pixels of each line segment l _j, it generates a edge point binary image Ie in accordance with the detection result.

Further, in the above third and fourth embodiments, a case has been described in which a polygonal line and a curve are input from the input means 4 as information on the edge position e and the edge direction d, respectively. However, the present invention is not limited to this. A plurality of line segments along the edge may be input from the input means 4 as information on the edge position e and the edge direction d. in this case,
Since information on both the edge position e and the edge direction d can be given by the line segment, the work efficiency can be improved.

[0099] In yet a third embodiment described above, the line segment l Distance from _{k d (lk, x, y} ) has dealt with the case of calculating in each case, the present invention is not limited to this, FIG. 17
The distance between the pixel through which the line segment passes and the coordinates (x, y) may be obtained using a square distance table as shown in FIG. Thereby, the calculation of the distance can be simplified. Here the distance table, by matching the coordinates (x, y) at the upper left position, which is created as the value of the table corresponding to the position where the line segment l _j passes becomes the square distance. This distance table is created when obtaining the distance in the lower right direction. Therefore, if three new tables are created by rotating this distance table by 90 °, distances in all directions can be easily obtained.

Further, in the above-described embodiment, the case where the present invention is applied to the key signal generating device 1 having the estimated contour calculating circuit 5 and the motion vector estimating circuit 10 has been described. However, the present invention is not limited to this. The present invention can be applied to a key signal generation device having the contour calculation circuit 5 or the motion vector estimation circuit 10. In the key signal generation device 1,
A switch for switching between the estimated contour calculating circuit 5 and the motion vector estimating circuit 10 may be provided so that the estimated contour calculating circuit 5 and the motion vector estimating circuit 10 can be switched.

[0101]

As described above, according to the present invention, the gradient in the contour candidate area for the contour of the object determined based on the position of the edge and the direction of the edge inputted by the predetermined input means. A vector is calculated to generate an edge intensity image, an integration path crossing the edge is calculated based on the edge intensity image, the edge position, and the edge direction, and a value obtained by integrating the edge intensity image on the integration path is standardized. The edge intensity can be normalized by the cross-sectional area of the edge by calculating the edge intensity image and normalizing the edge intensity image based on the edge intensity image and the normalization factor. The edge intensity can be detected uniformly without the need. Thus, a contour extracting apparatus and a contour extracting method capable of stably extracting the contour of an object from an image can be realized.

According to the present invention, a contour candidate area for a contour of an object is determined based on a plurality of line segments along the edge, which are input as information on the position of the edge and the direction of the edge. , A mask image designating a neighboring area of the image is generated, each mask image and each line segment are output as an edge position and an edge direction, and a gradient vector is calculated in a contour candidate area to generate an edge intensity image. On the basis of the intensity image, the corresponding mask image and the line segment, for each of the corresponding mask image and line segment, an integration path crossing each line segment is calculated, and the value obtained by integrating each mask image on each corresponding integration path is Output as a normalization coefficient, standardize each mask image based on each corresponding mask image and each standardization coefficient, and output the corresponding standardized mask image and By generating a standardized edge intensity image using each mask image, the contour of the object is divided by each line segment, and the edge intensity is normalized by the cross-sectional area of the edge for each region. Therefore, even when the shape of the contour of the target object is complicated and the direction of the contour changes greatly, the edge intensity can be detected uniformly without depending on the magnitude of the color difference in the input image. Thus, a contour extracting apparatus and a contour extracting method capable of stably extracting the contour of an object from an image can be realized.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an overall configuration of a key signal generation device to which the present invention is applied.

FIG. 2 is a schematic diagram for explaining a basic principle of the present invention.

FIG. 3 is a schematic diagram for explaining a basic principle of the present invention.

FIG. 4 is a block diagram showing a configuration of a gradient vector calculator according to the first embodiment of the present invention.

FIG. 5 is a block diagram showing a configuration of an integration path calculation circuit.

FIG. 6 is a flowchart for explaining a processing procedure of an integration path calculating process;

FIG. 7 is a flowchart for explaining a processing procedure of an image interpolation process;

FIG. 8 is a flowchart for explaining a processing procedure of a normalization coefficient calculation process.

FIG. 9 is a flowchart for explaining a processing procedure of a normalization process in the first embodiment.

FIG. 10 is a block diagram showing a configuration of a gradient vector calculation unit according to a second embodiment of the present invention.

FIG. 11 is a flowchart for explaining a processing procedure of a binarized image calculation process in the second embodiment.

FIG. 12 is a block diagram showing a configuration of a gradient vector calculator according to a third embodiment of the present invention.

FIG. 13 is a flowchart for explaining a processing procedure of a region dividing process.

FIG. 14 is a flowchart for explaining a processing procedure of a normalization process in the third embodiment.

FIG. 15 is a block diagram showing a configuration of a gradient vector calculation unit according to a fourth embodiment of the present invention.

FIG. 16 is a flowchart for explaining a procedure of a normalization process in a fourth embodiment.

FIG. 17 is a chart showing an example of a distance table.

FIG. 18 is a schematic diagram used to explain the effect on edge strength due to a change in color difference.

[Explanation of symbols]

1 ... key signal generation device, 2 ... contour extraction unit, 3 ... key calculation unit, 4 ... input means, 5 ... estimated contour calculation unit,
6, 51... Contour candidate area determination section, 7, 30, 40, 5
0 ... Gradient vector calculation unit, 8 ... Contour path search unit, 9 ... Curve generation unit, 10 ... Motion vector estimation unit, 11 ... Delay circuit, 12 ... End-end curve calculation unit, 13
... Soft key generator, 21... Edge strength calculation circuit,
22 integration path calculation circuit, 22A crossing straight line calculation circuit, 22B integration path end point calculation circuit, 23 standardization coefficient calculation circuit, 24 standardization circuit, 31 threshold calculation circuit 32: an edge point detecting circuit; 41: an image synthesizing unit.

Claims (30)

[Claims] 1. A contour extracting apparatus for extracting a contour of a target object from an input image as an edge, wherein the edge is detected based on a position of the edge and a direction of the edge according to a trajectory input by a predetermined input means. Contour candidate area determining means for determining and outputting a contour candidate area for the contour of the object; calculating a gradient vector in the contour candidate area; generating and outputting an edge intensity image based on the calculation result Edge intensity image generation means; integration path calculation means for calculating and outputting an integration path crossing the edge based on the edge intensity image, the position of the edge, and the direction of the edge; and integrating the edge intensity image with the integration A normalization coefficient calculating means for outputting a value integrated on the path as a normalization coefficient; and There the edge intensity image and normalized, contour extraction apparatus characterized by comprising a normalization means for outputting an edge strength image is the standardized. 2. A threshold calculating means for outputting, as a threshold, a value obtained by dividing a value obtained by integrating the edge intensity image on the integration path by an integration section of the integration path; Edge point detection means for detecting an edge point by subjecting the converted edge intensity image to a threshold value using the threshold value, and outputting an edge point binary image according to the detection result. The contour extraction device according to claim 1, wherein: 3. A crossing straight line calculating means for calculating and outputting a crossing straight line perpendicular to the edge based on the edge intensity image, the position of the edge and the direction of the edge. Based on the edge intensity image and the traversing line, two points on the traversing line at which the edge intensity value is initially equal to or less than a predetermined threshold are defined as end points with the edge as the boundary, and the respective end points 2. The contour extraction device according to claim 1, further comprising: integration path end point calculation means for outputting a line segment or the two end points on the traversing straight line separated by the following as the integration path. 4. An image interpolating means for calculating and outputting a value of the input image at arbitrary real number coordinates, wherein the integration path end point calculating means calculates the value of the input image on the crossing straight line based on an output of the image interpolating means. The contour extraction device according to claim 3, wherein two end points are calculated. 5. An image interpolation means for calculating and outputting a value of the input image at arbitrary real number coordinates, wherein the normalization coefficient calculating means converts the edge intensity image based on an output of the image interpolation means. 2. The method according to claim 1, wherein a value integrated on the integration path is output as a normalization coefficient.
4. The contour extraction device according to claim 1. 6. A contour path searching means for calculating a contour path passing through the center of the contour of the object based on the standardized edge intensity image, and outputting the contour path as contour path information; Contour curve generating means for generating a contour curve by approximating the contour path based on the contour path information; and contour estimation for detecting an estimated contour of the contour of the object in the current image based on the contour curve. Means for determining a contour candidate area for the contour of the object in the current image based on the estimated contour detected by the contour estimating means. 2. The contour extraction device according to claim 1, wherein: 7. A contour extraction device for extracting a contour of a target object from an input image as an edge, wherein the information on the position of the edge and the direction of the edge is
Input means for inputting a trajectory corresponding to a plurality of line segments along the edge, and generating a mask image specifying a neighboring area of each line segment based on each line segment, and generating each mask image and each line segment A contour candidate area determining means for outputting the position of the edge and the direction of the edge, respectively; an edge intensity image for calculating a gradient vector in the contour candidate area, generating an edge intensity image based on the calculation result, and outputting the edge intensity image Generating means for calculating, based on the edge intensity image, the corresponding mask image and the line segment, an integrated path crossing each line segment for each of the corresponding mask image and line segment, and outputting the calculated integrated path; And a normalization coefficient calculating means for outputting a value obtained by integrating each of the mask images on the corresponding integration path as a normalization coefficient. Standardizing means for normalizing the respective mask images based on the respective mask images and the respective standardization coefficients and outputting the standardized mask images; and a corresponding standardized mask image and the respective standardized mask images. A contour extracting apparatus, comprising: image synthesizing means for generating and outputting a standardized edge intensity image using a mask image. 8. A threshold value calculating means for outputting, as a threshold value, a value obtained by dividing a value obtained by integrating each of the mask images on each of the corresponding integration paths by an integration section of each of the integration paths. And performing threshold processing on the standardized edge intensity image using the threshold value, thereby detecting an edge point in a region near each of the line segments, and detecting an edge point binary image corresponding to the detection result. The edge extraction device according to claim 7, further comprising edge point detection means for outputting the edge point. 9. A crossing straight line calculating means for calculating and outputting a crossing straight line perpendicular to the edge based on the edge intensity image, the mask images and the line segments, respectively. Based on the edge intensity image and the respective traversing lines, two points on the traversing line at which the edge intensity value is initially equal to or less than a predetermined threshold are defined as end points, with the edge as a boundary. 8. The contour extraction device according to claim 7, further comprising an integration path end point calculating means for outputting each line segment or each of the two end points on each of the transverse straight lines separated by each end point as each of the integration paths. 10. An image interpolating means for calculating and outputting a value of the input image at an arbitrary real number coordinate, wherein said integration path end point calculating means calculates the value of said input image based on an output of said image interpolating means. 10. The contour extraction device according to claim 9, wherein the two end points are calculated. 11. An image interpolation means for calculating and outputting a value of the input image at an arbitrary real number coordinate, wherein the normalization coefficient calculating means calculates a value of the input image based on an output of the image interpolation means.
8. The contour extracting device according to claim 7, wherein a value obtained by integrating each of the mask images on the corresponding integration path is output as a normalization coefficient. 12. The input means includes information on the position of the edge and the direction of the edge.
A trajectory corresponding to a continuous polygonal line along the edge is input, and the contour candidate area determination means generates a mask image specifying a neighborhood area of each line of the polygonal line based on the polygonal line, and generates the mask image 8. The contour extracting apparatus according to claim 7, wherein each of the line segments is output as a position of the edge and a direction of the edge. 13. The information processing apparatus according to claim 11, wherein the input means includes information on a position of the edge and a direction of the edge.
A trajectory corresponding to a continuous curve along the edge is input, and the contour candidate area determining means approximates the curve to a polygonal line, and specifies a neighborhood area of each line of the polygonal line based on the polygonal line. 8. The contour extraction device according to claim 7, wherein the contour extraction device outputs the mask image and the line segments as the position of the edge and the direction of the edge, respectively. 14. A contour point candidate area determining means for extracting a bending point and / or a color change point on the curve inputted by the input means as a feature point; 14. The contour extraction device according to claim 13, further comprising a curve conversion unit that converts the inflection point and / or the color change point extracted by the feature point extraction unit into the polygonal line having the vertex. 15. A contour path searching means for calculating a contour path passing through the center of the contour of the object based on the standardized edge intensity image, and outputting the contour path as contour path information; Contour curve generating means for generating a contour curve by approximating the contour path based on the path information; and contour estimating means for detecting an estimated contour of the contour of the object in the current image based on the contour curve. Wherein the contour candidate area determining means determines a contour candidate area for the contour of the object in the current image based on the estimated contour detected by the contour estimating means. The contour extraction device according to claim 7, wherein: 16. A contour extraction method for extracting a contour of a target object from an input image as an edge, wherein the edge is detected based on a position of the edge and a direction of the edge according to a trajectory inputted by a predetermined input means. A contour candidate area determining step of determining and outputting a contour candidate area for the contour of the object; calculating a gradient vector in the contour candidate area; generating and outputting an edge intensity image based on the calculation result; An edge intensity image generation step, an integration path calculation step for calculating and outputting an integration path crossing the edge based on the edge intensity image, the position of the edge, and the direction of the edge, and the integration of the edge intensity image. A normalization coefficient calculation step of outputting a value integrated on the path as a normalization coefficient; A contour extraction method comprising: normalizing the edge intensity image based on the standardization coefficient; and a normalization step for outputting the standardized edge intensity image. 17. A threshold value calculating step of outputting a value obtained by dividing a value obtained by integrating the edge intensity image on the integration path by an integration section of the integration path as a threshold value, Edge point detection step for detecting edge points by subjecting the converted edge intensity image to a threshold value using the threshold value, and outputting an edge point binary image corresponding to the detection result. 17. The contour extraction method according to claim 16, wherein: 18. The traversing line calculation step for calculating and outputting a traversing line perpendicular to the edge based on the edge intensity image, the position of the edge, and the direction of the edge. Based on the edge intensity image and the traversing line, two points on the traversing line at which the edge intensity value is initially equal to or less than a predetermined threshold are defined as end points with the edge as the boundary, and the respective end points 2. An integration path end point calculation step for outputting a line segment on the traversing straight line or the two end points delimited by the following as the integration path.
7. The contour extraction method according to 6. 19. The integration path end point calculation step calculates the value of the input image at arbitrary real number coordinates, and calculates the two end points on the transverse line based on the calculation result. The contour extraction method according to claim 18. 20. The normalization coefficient calculating step calculates a value of the input image at arbitrary real coordinates, and normalizes a value obtained by integrating the edge intensity image on the integration path based on the calculation result. 17. The contour extraction method according to claim 16, wherein the method is output as a coefficient. 21. A contour path search step for calculating a contour path passing through the center of the contour of the object based on the standardized edge intensity image, and outputting the contour path as contour path information; A contour curve generation step for generating a contour curve by approximating the contour path based on the contour path information; and a contour estimation for detecting an estimated contour of an object in the current image based on the contour curve. The outline candidate area determination step is performed based on the estimated outline detected in the outline estimation step.
17. The contour extraction method according to claim 16, wherein a contour candidate area for a contour of the object in the current image is determined. 22. A contour extracting method for extracting a contour of a target object from an input image as an edge, wherein information on the position of the edge and the direction of the edge is provided as:
An input step of inputting a trajectory corresponding to a plurality of line segments along the edge; and generating a mask image designating a neighborhood area of each line segment based on each line segment, and generating each mask image and each line segment. A contour candidate area determination step to be output as the edge position and the edge direction, respectively; an edge intensity image for calculating a gradient vector in the contour candidate area and generating and outputting an edge intensity image based on the calculation result A generation step and, based on the edge intensity image, the corresponding mask image and the line segment, an integration path calculation step of calculating and outputting an integration path crossing each line segment for each of the corresponding mask image and line segment. And a normalization coefficient calculation for outputting a value obtained by integrating the respective mask images on the corresponding integration paths as a normalization coefficient. A step of normalizing each of the mask images based on the corresponding mask images and the standardization coefficients, and outputting the standardized mask images; and a corresponding one of the standardized mask images. An edge synthesizing method, comprising: a mask image and an image synthesizing step of generating and outputting a standardized edge intensity image using the mask images. 23. A threshold value calculating step of outputting a value obtained by dividing a value obtained by integrating each of the mask images on each of the corresponding integration paths by an integration section of each of the integration paths as a threshold value. And performing threshold processing on the standardized edge intensity image using the threshold value, thereby detecting an edge point in a region near each of the line segments, and detecting an edge point binary image corresponding to the detection result. 23. The edge extraction method according to claim 22, further comprising an edge point detection step for outputting the edge point. 24. A crossing straight line calculating step for calculating and outputting a crossing straight line which crosses the edge vertically based on the edge intensity image, the mask images and the line segments, respectively. Based on the edge intensity image and the respective traversing lines, two points on the traversing line at which the edge intensity value is initially equal to or less than a predetermined threshold are defined as end points, with the edge as a boundary. 23. The contour extraction method according to claim 22, further comprising: an integration path end point calculation step of outputting each line segment or each of the two end points on each of the transverse straight lines separated by each end point as each of the integration paths. 25. The integration path end point calculation step calculates the value of the input image at arbitrary real number coordinates, and calculates the two end points on each of the transverse straight lines based on the calculation result. 25. The contour extraction method according to claim 24, wherein: 26. The normalization coefficient calculation step calculates a value of the input image at an arbitrary real number coordinate, and integrates each mask image on the corresponding integration path based on the calculation result. 23. The contour extraction method according to claim 22, wherein is output as a normalization coefficient. 27. The input step includes information on the position of the edge and the direction of the edge.
A trajectory corresponding to a continuous polygonal line along the edge is input, and the contour candidate area determination means generates a mask image specifying a neighborhood area of each line of the polygonal line based on the polygonal line, and generates the mask image 23. The contour extracting method according to claim 22, wherein each of the line segments is output as a position of the edge and a direction of the edge. 28. The input step includes: information on the position of the edge and the direction of the edge;
A trajectory corresponding to a curve that is continuous along the edge is input, and the contour candidate area determination step approximates the curve to a polygonal line, and specifies a neighboring area of each of the polygonal lines based on the polygonal line. 23. The contour extraction method according to claim 22, further comprising: generating a mask image and outputting each of the mask images and the line segments as a position of the edge and a direction of the edge. 29. The contour candidate area determining step includes: a feature point extracting step of extracting a bending point and / or a color change point on the curve input in the input step as a feature point; 29. The contour extraction method according to claim 28, further comprising a curve conversion step for converting the inflection point and / or the color change point extracted in the extraction step into the polygonal line having the vertex. 30. A contour path searching step for calculating a contour path passing through the center of the contour of the object based on the standardized edge intensity image, and outputting the contour path as contour path information; A contour curve generating step of generating a contour curve by approximating the contour path based on the path information; and a contour estimating step of detecting an estimated contour of the object in the current image based on the contour curve. Wherein the contour candidate area determining step determines a contour candidate area for the contour of the object in the current image based on the estimated contour detected in the contour estimating step. 23. The contour extraction method according to 22.