JPH07271989A - Boundary detecting method of background and object area, contour extracting method of object and object contour extracting device - Google Patents

Abstract

PURPOSE:To drastically shorten the time till when a contour is converged by providing a background/object area decision process successively performing a decision on a background or an object area for the plural sections of the strip area generated in a division process. CONSTITUTION:The analog image signal photographed by a CCD camera 21 is converted into a digital signal in an A/D conversion part 22 and the signal is stored as an original image in an image storage part 23. The inside of the strip area set at the both sides of the input image stored in the image storage part 23 is investigated, a decision on a background or a person area is performed and the boundary is detected by a background/person boundary location detection part 25. By using the detected background, the boundary location of a person and the Y coordinate of the top part detected in a top part location detection part 26, the contour is set by an initial contour setting part 27 to be used in a person contour search part 28. The person contour extracted in the person contour search part 28 is displayed by overlapping it with the original image stored in the image storage part 23. Therefore, the contour of the person can be automatically extraced.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a background / object region boundary detection method, an object contour extraction method, and an object contour extraction apparatus, which are capable of automatically obtaining the contour of a person against a background.

As a method for extracting the contour line of an object, which is the boundary between the object region and the background region, from an image showing a certain object, for example, there is one shown in FIG.
In this method, an operator manually specifies a rough outline called a rough outline before extracting the outline. This rough contour line is specified as a thick strip-shaped area along the contour that includes the true contour line, and the coarse contour line is thinned to preserve the connectivity of the pixels and the connected contour lines are extracted. .

In the method called elastic contour model, an initial closed contour a is set in the vicinity of the contour line of the object to be extracted as shown in FIG. 16, and it is determined by the shape of the contour line, the position in the image and the like. There is a method in which the contour is moved so as to optimize the energy function, and the contour b of the object is finally obtained.

[0004]

However, in the case of the method of thinning the rough contour line, it is troublesome because the operator needs to specify the rough contour line, and a large amount of different data must be processed. If not,
There is a problem that the burden is heavy.

In the case of a closed contour energy optimization method using an elastic contour model, the initial contour is set and
There is a problem that the amount of calculation required for energy optimization becomes large, and it takes time to converge the contour of the final object from the initial contour.

The present invention has been made in view of the above points, and an object thereof is to eliminate the need for operation by an operator and to significantly reduce the time until the contour is converged. An object is to provide a boundary detection method, an object contour extraction method, and an object contour extraction device.

[0007]

A method for detecting a boundary between a background and an object area according to a first aspect of the present invention is a projection waveform of an edge value for a pixel in a band-shaped area set so as to cross a boundary line between the background and the object area. An edge value projection waveform calculating step of calculating a projection waveform of the edge value calculated in the edge value projection waveform calculating step, and a dividing step of dividing the strip-shaped region into a plurality of sections with a first reference value as a boundary; The present invention is characterized by further comprising a background / object region determination process of sequentially determining whether a background or an object region is generated for a plurality of sections generated in this division process.

In the background / object region boundary detecting method according to claim 2, in the background / object region determining step according to claim 1, the average brightness in each section is within the allowable range of the reference brightness and within the section. When the average value of the edge values is smaller than the second reference value, it is determined to be the background, and in other cases, it is determined to be the object region.

In the background / object region boundary detection method according to the third aspect, when the background / object region determination step of the second aspect determines the background, the average brightness of the determined section is changed to the next section. It is characterized in that it is updated as the reference luminance of.

According to a fourth aspect of the present invention, there is provided a contour extracting method, wherein an evaluation value calculating step of calculating an evaluation value from an edge strength on an initial contour preset outside the object area, and a direction of narrowing the initial contour toward the object area. Contour movement process to move to
An evaluation value calculation step of calculating the evaluation value of the contour from each edge strength of the control points arranged at substantially equal intervals on the contour obtained in this contour movement step, the previous evaluation value and the present evaluation value And the convergence determination step for determining whether the contour is converged, and if the convergence determination step determines that the contour is not converged, move the contour further in the direction of narrowing the object area. It is characterized in that the evaluation value of the contour is calculated again by the evaluation value calculation step, and is repeated until the contour converges by the convergence determination step.

In the contour extraction method according to the fifth aspect, it is considered that the contour moving step of the fourth aspect receives the tension from the first vector toward the maximum edge strength position near the control point on the contour and the control point. In this case, the destination of the control point is determined by the weighted sum of the second vector that moves toward the equilibrium position of the control point and the third vector that forcibly moves the control point toward the person.

According to a sixth aspect of the present invention, there is provided an object contour extracting device, an image pickup means for taking an image of an object with a background in the background, a storage means for storing image data of the object taken by the image pickup means, and the storage means. A background object boundary detection means for calculating the projection waveform of the edge value for the pixels in the strip-shaped area set so as to cross the boundary line between the background and the object area stored in, and detecting the boundary between the background and the object. , An initial contour setting means for setting an initial contour of the object based on the boundary detected by the background object boundary detection section, and an initial contour of the object set by the initial contour setting means for controlling the vicinity of the control point on the control point. A first vector heading to the maximum edge strength position, a second vector heading to a balance position of the control points when it is considered that tension is applied from the control point, and a third vector forcibly moving the control point toward the person. Vector addition And the object contour searching means for determining the destination of the control points by the sum, constituted by a determining convergence determining means for converging the evaluation value of the contour of the control points of the contour of searching by the object outline search means.

[0013]

According to the present invention, the projection waveform of the edge value is obtained for the pixels in the area set so as to cross the boundary line between the background and the object area, and each section divided by the position of the peak value of the projection waveform is determined. , The pixel value of the original image and the feature of the edge value of the original image are sequentially evaluated to identify whether it is the background or the object region, and detect the boundary between the background and the object region.

According to a fourth aspect of the present invention, the outline of the person is extracted by updating the outline while repeatedly searching for the outline of the person in accordance with the movement rule from the initial outline determined outside the area of the person. .

According to a sixth aspect of the present invention, the boundary between the background and the object area is automatically detected from the image of the object with the background imaged by the imaging means, the initial contour is set based on the boundary, and the initial contour is controlled. The first vector heading to the maximum edge strength position in the vicinity of the control point on the point, the second vector heading to the balanced position of the control point and the control point are forced when the tension from the control point is considered. Further, the moving destination of the control point is determined by the weighted sum of the third vector moved toward the person, and the convergence of the contour is determined from the evaluation value of the control point of the contour.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An object contour extracting apparatus according to an embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of an apparatus for photographing an upper half of a person under a substantially uniform background using a CCD camera and automatically extracting a contour line of a person region in the photographed image.

In FIG. 1, reference numeral 20 denotes a control unit which controls the apparatus as a whole, and is composed of, for example, a microcomputer and its peripheral circuits, and outputs a control signal to each processing unit described later.

Reference numeral 21 is a CCD camera for image input. The analog image signal photographed by the CCD camera 21 is converted by the A / D converter 22 into a digital signal having, for example, 256 gradation levels according to the brightness of the subject, and then the original image is stored in the image storage 23. Is stored as

The edge image creating section 24 is an image storage section 23.
An edge image is created from the original image stored in the image storage unit 23 and stored in the image storage unit 23. Although various types of edge images can be used in the method described in the present embodiment, in the present embodiment, the Sobel filter, which is one of the edge extraction filters, is used as the input image stored in the image storage unit 23. What was made to act was used.

The background / person boundary position detecting section 25 detects the boundary between the background and the person area by examining the strip-shaped areas set on both sides of the input image stored in the image storage section 23. Detailed functions of the background / person boundary position detection unit 25 will be described later with reference to FIG.

The crown position detecting unit 26 detects the Y coordinate of the crown in the input image stored in the image storage unit 23. The initial contour setting unit 27 uses the boundary position between the background and the person detected by the background / person boundary position detection unit 25, and the Y coordinate of the crown detected by the crown position detection unit 26 to determine the outside of the person region. Then, the initial contour used in the next person contour search unit 28 is set.

The person contour search unit 28 is provided with an initial contour setting unit 2
Based on the initial contour set in step 7, edge search and contour movement are repeated, and the contour line of the person is finally extracted.

The person contour output unit 29 is a person contour search unit 28.
The person outline extracted in step 3 is superimposed on the original image stored in the image storage unit 23 and displayed on a display (not shown).

Next, the operation of the embodiment of the present invention constructed as above will be described. The image of the upper half of the body of the person captured at a certain time after being photographed by the CCD camera 21 is a digital image, and each pixel has 256 gradation values according to the brightness (luminance) of the subject. This original image is stored in the image storage unit 23. Then, the edge image creation unit 24 creates an edge image by applying a Sobel filter for edge extraction from the original image stored in the image storage unit 23, and stores it in the image storage unit 23.

The process of detecting the boundary between the background and the person performed by the background / person boundary detecting section 25 will be described below with reference to the flowchart of FIG. First, as shown in FIG. 3, the strip areas used in the projection calculation in the strip area of FIG. 2 (step 30) are set in advance on both sides of the image. The strip area on the right side is the strip area R42 and the strip area on the left side is the strip area R42. L43. The orientations of the band-shaped regions 42 and 43 are set so as to cross the boundary for which the position is to be detected, but in the case of the present embodiment, since the boundary of the shoulder position extending in the X direction 40 is to be detected as the main boundary, Band regions R42, L4
The direction of 3 was set in the Y direction 41.

Then, these strip regions R42, L43
For each pixel, the sum of pixel values with the same Y coordinate is calculated for each Y
The projected waveform is obtained by obtaining the coordinates. Here, the projected waveform is such that the strip regions R42 and L43 are projected in the Y direction 41.
In order to obtain a boundary candidate for dividing into a plurality of sections, or to derive a feature for identifying whether each divided section is a background or a person.

In the present embodiment, the former Y direction component image having high sensitivity to an edge having a large change in pixel value in the Y direction 41 (created by the edge image creating unit 26 and stored in the image storage unit 23). The original image was used for the latter.

Then, the projected waveform of the edge image is smoothed by an appropriate scale (step 31). As a result, the projected waveform of the original image, that is, the luminance image is 44, and the projected waveform of the Y-direction component image of the edge after smoothing (referred to as the projected waveform of the edge image) is 45.

Next, processing for selecting a boundary candidate point (step 3)
In 2), local maxima whose values of the projected waveform 45 of the edge image are equal to or larger than the first threshold TH1 determined in advance are extracted, and the positions of these maxima are designated as the boundary candidate positions 46 of the background and the human area. To be done.

Here, each section divided at the boundary candidate position 46 is referred to as a level 1 section 48. Next, each section divided at the boundary candidate position 46 will be described with reference to the flowchart of FIG. 4 regarding the section determination processing (step 33) for determining whether the section 48 of level 1 is a human area. .
In the section determination processing here, the section determination processing is performed for each section of level 1 divided in the boundary candidate point selection processing (step 32) in order from the section above the image.

In FIG. 4, first, it is determined whether the section length of level 1 is larger than the specified value LEN. If it is determined to be "NO" in this determination, that is, if the section length of level 1 is less than LEN, the section determination process (step 50) is performed on the section.

This section determination process (step 50) is shown in FIG.
The determination process shown in the flowchart of FIG. That is,
If the evaluation value of the section, the brightness in the section and the average value of the edges in this embodiment satisfy the following conditions, the section is determined as the background section. The condition is the flow of FIG.
As shown in the chart, the average brightness value is within the predetermined allowable range centered on the reference brightness value, and the average edge value is less than the second threshold value TH2 set in advance. There is. Here, the initial value of the reference luminance value is set so as to be the average luminance of the background upper portion without fail.

If the applicable section is other than the above conditions, it is output that it is a person section. Returning to the flow chart of FIG. 4, it is determined whether the section determination in the section determination processing (step 50) is a background section, and if it is determined that it is a background section, the level 1 section reference The brightness is updated (step 50a). This reference luminance is the flow of FIG.
Used for chart determination. In this update process,
Each time the interval is determined to be the background interval, the standard brightness value is equal to or less than the predetermined standard deviation of the brightness of the interval,
That is, when the background portion is smooth, it is determined that the average luminance value is reliable as the reference luminance value, and the reference luminance value is updated to the average luminance value.

By thus updating the reference luminance value, accurate determination can be performed even if the characteristic amount of the background portion, that is, in the case of the present embodiment, the luminance is gradually changing.

As described above, it is determined whether or not the level 1 section is the background section, and the reference value used in the level 1 section determination is updated. Level 1 mentioned above
If the section length of is greater than the specified value LEN, then "YE
S ”, that is, when the section length in the section of level 1 is equal to or greater than the predetermined value LEN, the length is reduced to almost the specified value LEN as shown by 47 in FIG. 3B. Processing is performed by dividing the sections into sections and setting these sections as Level 2 sections.

Then, a section determination process (step 51) is performed for each level 2 section. The determination processing of this section determination processing (step 51) is basically the same as the above-described flowchart determination processing of FIG. This section determination processing (step 51) is basically the same as the section determination processing 50 for the section of level 1 described above, but the reference luminance value to be used is used only during section determination of level 2.

Then, it is judged whether or not the section judgment processing has been performed for all the level 2 sections (step 5).
1a), if it is determined that there is a level 2 section for which the section determination processing has not been performed, it is determined that the level 2 section is a background section, and the standard deviation of the luminance in that section is a certain value or less. If it is determined that the reference luminance value used in the level 2 section determination is updated (step 51b).

When the section determination processing (step 51) is performed for all the level 2 sections, the number of small sections of level 2 which is determined to be "NO" in step 51a and which is determined to be the background is constant. It is determined whether the ratio is exceeded.

When it is determined in step 51a that the number of small sections of level 2 which are determined to be background exceeds a certain ratio, the corresponding section of level 1 is set as the background section and The value is updated to the reference value used in the section determination (step 50a).

As the value to be updated, the average brightness level of the penultimate small section is used. This is level 1
This is because the luminance change at the boundary of the section is not always clear, and the luminance value of the last small section may include a portion where the luminance value of the adjacent level 1 section changes. .

In this way, the background / person boundary position detection unit 2
In No. 5, since the reference value used when determining whether it is the background section is updated, it seems that the background brightness is not uniform and changes smoothly due to shading depending on the condition of the light rays. Even in this case, the boundary between the background and the object can be detected. Therefore, in the system as shown in the embodiment, the degree of freedom of the illumination condition when capturing the input image is increased. Generally, in order to perform uniform illumination, it is necessary to devise such as using many lights, but the boundary position between the background and the object can be stably detected even when the image is taken under simple illumination. In addition, since the detected position of the boundary is highly reliable, the burden on the subsequent processing is reduced when the position is used as a reference point for detecting the position of the person in the image.

Next, the processing for detecting the top of the head performed by the top position detecting section 26 will be described. The parietal position detecting unit 26 detects the Y coordinate of the parietal region, but the band-shaped region is arranged at approximately the center of the image as shown by 43a in FIG. It can be detected by performing a process substantially similar to that of the person boundary position detection unit.

Next, the processing in the initial contour setting section 27 will be described. The initial contour setting unit 27 sets an initial contour for searching a human contour by a human contour searching unit 28, which will be described later. The initial contour is drawn outside the person area according to the boundary position between the background and the person area on both sides of the image obtained by the background / person boundary position detecting section 25 and the position of the crown obtained by the crown position detecting section 26. Set as shown in 6. In the initial contour, there are a type 70 like A in FIG. 6 and a type 71 like B. Here, the type A is determined by the position of the crown and the boundary positions of the background and the person on both sides of the image, and the type B is determined by the boundary positions.

Next, the processing of the person contour searching unit 28 will be described with reference to the flowchart of FIG. First, move the initial contour by moving each control point,
The convergence judgment is performed according to the judgment logic shown in the flowchart of FIG. 8 (step 80).

The determination of the convergence (step 80) is made by checking the change state of the evaluation value of the contour at each time point as shown in the flowchart of FIG. That is, when the evaluation value is obtained from the edge strength on the contour and the difference from the evaluation value at the time of the previous determination is equal to or less than a certain value and the change amount is less than the certain value for a predetermined number of times, the contour converges. To output that.

Then, it is determined whether or not the result of the convergence determination process shown in FIG. 7 has converged. If it is determined that the result has not converged, a process for updating the control point is performed (step 8).
1). Then, the destination of the control point is searched for (step 82), the control point is moved, and the process returns to the beginning and interpolation between the control points is performed. Such contour search is repeatedly performed until the contour that moves along with the search matches the contour of the human region and converges.

Next, the search for the control point movement destination of the flow chart of FIG. 7 (step 82) will be described. First, a contour moving method will be described with reference to FIG. The contour is moved by moving the control points forming the contour. The control points at both ends of the contour are not moved. The destinations of the control points k other than the both ends of the contour are determined by three types of vectors. The first vector 100 is a vector toward a point having the maximum edge strength near the control point. The second vector 101 is a vector that goes to the equilibrium position of the control points when it is considered that the control points receive the tension from the neighboring control points. The third vector 102 forces the control point when the control point is far away from the human contour and the size of the first vector 100 is zero and when there is an unnecessary edge that is not caused by the human contour. It is a vector that moves toward the person because it moves in the direction toward the person outline. These 3
The destination of the control point is determined by the vector weighted sum of the seeds.

By the way, when the position of the end point of the initial contour is not on the boundary between the background and the person, the contour may move as shown in FIG. 10 and the contour may try to extend outside the image. In order to prevent this, the edge image used to obtain the first vector 100 has an edge value set in advance to a sufficiently strong edge value on a straight line from the end point of the contour to the bottom of the image as shown in FIG. You may leave it. This is called a barrier 110. In this case, since the edge is strong after the contour reaches the straight line, the contour stays in the vicinity of the barrier and does not extend outside the image. in this case,
The contour to be extracted includes not only the person contour but also a portion along the barrier, but a desired person contour can be extracted by using a portion other than the barrier.

Next, referring to the flow chart of FIG. 12, the processing for determining the destination of the control point will be described. The weighting coefficient changes depending on the edge average value of pixels near the control point. When the edge average value increases, the weighting coefficient of the third vector decreases. By setting the range of this neighborhood area to an appropriate size, the weight of the first vector is increased when the control point is sufficiently close to the human contour, and it is caused by the local luminance change in the background. The weight of the third vector can be increased at the edge portion. With such a mechanism, it is possible to quickly converge the vicinity of the human contour without being caught by an unnecessary edge of the background portion.

Next, the control point updating process (step 81) will be described with reference to FIG. In this process, in order to avoid trapping the control points at the edges caused by unnecessary object images existing in the background part, in order to avoid the position being the same as the previous time for each search, except for the vicinity of the end points of the contour. A new control point sequence is selected by selecting points from the contour at approximately equal intervals. The control point before the Nth update is 13 in FIG.
When it is as shown in 0, the control point is selected so as to be at a different position on the contour like 131 in the Nth update. Since the intervals between the control points are selected to be substantially equal, the number of control points changes depending on the length of the contour. By this processing, even if there are some edges in the background portion, it is possible to move to the person outline without the outline being caught by these edges.

As described above, the contour moves from the initial contour x1 as shown in FIG. 14, converges to the person contour y through the contour x2 being searched / moved, and extracts the person's contour. You can

Next, the extracted contour output unit 29 will be described. Here, processing is performed in which the person contour y searched by the person contour searching unit 28 is superimposed on the original image in a color that can be distinguished from the original image and displayed on a display (not shown).

As described above, the edge search and movement are repeated based on the automatically set initial contour to extract the human contour, so that the operator does not have to perform work such as specifying a rough position. Moreover, the outline of the person can be automatically extracted. When it is necessary to process a large number of input images, even a slight amount of processing by an operator is very troublesome and impractical. In this respect,
Since this apparatus can automatically extract a person's contour, various systems using the extracted contour can be configured.

The background / person boundary position detecting unit 25
Although the luminance image is used for the processing for deriving the feature for identifying whether each divided section performed in step 1 is a background or a person, each component image of, for example, a color image may be used instead of this luminance image. Furthermore, the strip regions R42 and L43
Is divided into a plurality of sections in the Y direction 41, the edge Y
Although the projection waveform of the image of the direction component is obtained, the X direction component image of the edge, or the sum of the X direction component and the Y direction component may be used.

In the above embodiment, in the strip area,
The process of detecting the boundary position between the object and the background was performed in order to find the boundary position between the background and the shoulders of the person near both sides of the image. It goes without saying that it can be generally applied to detect the boundary position between an object having different brightness and the background.

Further, in the above embodiment, the human contour output unit 2
In addition to the display, the output method of 9 can be output in various ways such as output in the form of paper or film with a printer, output in the form of computer data file, output in the form of video signal, etc. Is.

Furthermore, in the above embodiment, the upper half of the body of a person is photographed using a CCD camera under a substantially uniform background, and the contour line of the person region in the photographed image is automatically extracted. It is also possible to extract the contour line of the above from the image showing the upper half of the person, and use the extracted contour to synthesize and display only the person portion in the separately prepared background image.

Various applications are conceivable as application examples using the human contour obtained by the system of the present embodiment. For example, a system that cuts out a person area from the original image using the extracted person outline and inserts it into a separately prepared landscape image, and also combines it with a resin or metal plate processing device to form a keychain in the shape of the upper body of the person. Various application forms are possible such as a system for automatically creating.

[0059]

As described above in detail, according to the present invention, the outline of the person is extracted by repeating the search and movement of the edge based on the automatically set initial outline, and thus the outline position by the operator is extracted. The outline of a person can be automatically extracted without performing work such as specifying.

Further, since the reference value for judging whether the background or the person is updated, even when the background brightness is not uniform and changes smoothly due to the shading of the light rays, The boundary between the background and the object can be detected. Furthermore, since the control point is moved by the weighted sum of weighting of the three vectors that changes when moving, the contour can be converged quickly.

[Brief description of drawings]

FIG. 1 is a block diagram of an object contour extraction apparatus according to an embodiment of the present invention.

FIG. 2 is a flowchart for explaining the operation of a background / person boundary position detection unit 25.

FIG. 3 is an explanatory diagram for explaining processing of a background / person boundary position detection unit 25.

FIG. 4 is a flowchart for explaining processing of a background / person boundary position detection unit 25.

FIG. 5 is a flowchart illustrating a section determination process of a background / person boundary position detection unit 25.

FIG. 6 is an explanatory diagram of an initial contour setting unit 27.

FIG. 7 is a flowchart for explaining the operation of the person contour search unit 28.

FIG. 8 is a flowchart for explaining a convergence determination process of a person contour search unit 28.

FIG. 9 is an explanatory diagram of a process of determining a movement amount of a control point of the human contour search unit 28.

FIG. 10 is an explanatory diagram of a process of preventing the contour of the person contour search unit 28 from protruding outside the image.

FIG. 11 is an explanatory diagram of a process of preventing the contour of the person contour search unit 28 from protruding outside the image.

FIG. 12 is a flowchart of a process for obtaining a movement amount of a control point of the human contour search unit 28.

FIG. 13 is an explanatory diagram of control point update processing of the human contour search unit 28.

FIG. 14 is an explanatory diagram of a process of extracting a person outline by the person outline search unit 28.

FIG. 15 is an explanatory diagram of a conventional method.

FIG. 16 is an explanatory diagram of a conventional method.

[Explanation of symbols]

20 ... Control unit, 21 ... CCD camera, 22 ... A / D conversion unit, 23 ... Image storage unit, 24 ... Edge image creation unit, 25
... background / person boundary position detection unit, 26 ... crown position detection unit, 27 ... initial contour setting unit, 28 ... person contour search unit, 2
9 ... Human contour output unit.

Claims (6)

[Claims] 1. An edge value projection waveform calculation step of calculating a projection waveform of an edge value for a pixel in a band-shaped area set so as to cross a boundary line between a background and an object area, and calculation in the edge value waveform calculation step. A dividing step of dividing the band-shaped area into a plurality of sections with a maximum point of the waveform of the projected waveform of the edge value determined by the first reference value as a boundary; A background / object region boundary detection method comprising: a background / object region determination step of determining whether the object region is an object region. 2. The background / object region determination step is performed when the average brightness in each section is within an allowable range of reference brightness and the average edge value in the section is smaller than a second reference value. 2. The method for detecting the boundary between a background and an object area according to claim 1, wherein the method determines that it is a background, and otherwise determines that it is an object area. 3. When the background / object region determination step determines that the background is the background, the average luminance of the determined section is updated as the reference luminance of the next section. Boundary detection method between background and object area. 4. An evaluation value calculating step of calculating an evaluation value from an edge strength on an initial contour set outside the object area in advance, and a contour moving step of moving the initial contour in a direction narrowing the object area direction. An evaluation value calculation process that calculates the evaluation value of the contour from the edge strengths of the control points that are arranged at approximately equal intervals on the contour obtained in this contour movement process, the previous evaluation value and the current evaluation value And the convergence determination step for determining whether the contour has converged, and if the convergence determination step determines that the contour has not converged, move the contour further in the direction of narrowing the object region. A contour extraction method characterized in that the evaluation value of the contour is calculated again by the evaluation value calculation step, and repeated until the contour converges in the convergence determination step. 5. The contour moving step moves to a balanced position of control points when it is considered that a first vector is directed to a maximum edge strength position near a control point on the contour and tension is applied from the control point. 5. The contour extraction method according to claim 4, wherein the destination of the control point is determined by a weighted sum of the second vector and a third vector for forcibly moving the control point toward the person. 6. An image pickup means for picking up an image of an object against a background, a storage means for storing image data of the object taken by the image pickup means, and a background and an object stored in the storage means. The background object boundary detection unit that calculates the projection waveform of the edge value for the pixels in the band-shaped area set to cross the boundary with the area and detects the boundary between the background and the object, and this background object boundary detection unit First contour setting means for setting an initial contour of the object based on the detected boundary, and first initial contour of the object set by the initial contour setting means to the edge strength maximum position near the control point on the control point
Of the control point and a second vector heading toward the equilibrium position of the control point when it is considered that the tension is applied from the control point,
An object contour search means for determining the destination of the control point by the weighted sum of the third vector for forcibly moving the control point toward the person, and an outline of the contour based on the evaluation value of the control point of the contour searched by the object contour search means. An object contour extracting device, comprising: convergence determining means for determining convergence.