JP2000209431A - Method for extracting contour and storage medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accurately extract a contour of a thin and long object by detecting an edge strength from a received image, correcting the edge strength in response to a magnitude of a predetermined weight coefficient, setting a first target point of tracing the contour and selecting a vicinity pixel with a largest value of the corrected edge strength for a succeeding contour point. SOLUTION: First image data being a processing object are loaded to an image memory or the like, and an edge component is detected from the image data loaded in this image memory. Then pixels to start and to complete contour tracing are designated. The tracing start pixel is a first target pixel. A homogeneity coefficient decided depending on the distribution of a direction θ of the edge is multiplied with the edge strength E to calculated a corrected edge strength E'. After calculating the corrected strength E', a pixel whose corrected edge strength E' is maximum is retrieved for pixels in the vicinity of the target pixel. The retrieval can efficiently be conducted by employing a mask to limit the retrieval range.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a contour line extracting method and a storage medium, and more particularly, to extracting an object from an image and performing geometrical operations such as moving, inverting, rotating, scaling and deformation. Image editing and processing tools to perform processing, image synthesis tools to perform inset synthesis to other images, image processing tools to perform image processing such as color conversion and filtering, for example, limited to specified areas, and The present invention relates to a contour line extraction method and a storage medium used for computer vision or the like that divides an image into regions to understand a scene.

[0002]

2. Description of the Related Art Conventionally, a contour extraction method of the type for tracking the edge strength is as follows: (1) A local maximum edge pixel is tracked using a multi-valued edge image while referring to a past trace direction. There are an edge pixel tracking method and (2) a boundary point connection method of detecting a boundary point between an object and a background within a predetermined area and connecting the detected boundary points.

As described above, the edge pixel tracking method uses a multi-valued edge image to select a pixel having a maximum edge intensity in a pixel adjacent to a target pixel as a contour pixel. Tokiko 63-5
Japanese Patent No. 3587 discloses a configuration in which the area for searching for the next contour point candidate is limited to a range of 180 degrees with respect to the immediately preceding contour tracking direction to improve tracking efficiency. Japanese Patent No. 2,645,995 discloses that, based on the premise that it is natural to perform the next tracking in the same direction as the past tracking direction, the weighting becomes stronger as the pixel edge strength in the direction along the past tracking direction increases. A method of tracing an edge is described.

In the boundary point connection method, as described above, a contour line is extracted only for a band-like region set to include the contour line. In the method described in JP-B-63-55101,
The operator first traces the target image roughly along the contour. A band-like region having a predetermined width in the trace of the trace is selected, and the outline of the image is determined from the difference in the color tone value in the band-like region. On the other hand, US Pat.
In the method described in Japanese Patent Application Laid-Open No. 7-92650, a process of forming a band-like region by connecting at least two designated points on a contour and detecting a contour point in the width direction of the band-like region is performed. The contour is extracted by repeating while moving in the longitudinal direction of the band-shaped region.

[0005]

In the prior art, since the contour is extracted only based on the magnitude of the change in the image data value, that is, the edge strength, the edge strength of the contour is equal to the edge strength near the contour. If it is not sufficiently large, there is a disadvantage that the contour cannot be extracted accurately.

Specifically, first, when there is a pixel having a large edge strength near a contour, a pixel near the contour is selected as a contour point. Second, when the contrast of the contour line is extremely low, the path of the contour tracking cannot be determined only by the edge strength information. As a result, a fine stray is repeated, so that an accurate contour cannot be extracted. Third
On the other hand, when an object having an elongated shape is extracted, the outlines on both sides of the object are close to each other, and the outline having the smaller edge strength cannot be tracked.

An object of the present invention is to provide a contour extraction method and a storage medium which do not cause such inconvenience.

That is, an object of the present invention is to provide a contour line extracting method and a storage medium capable of extracting a contour line without being disturbed by the granular noise even if there is a granular noise having random direction in the vicinity of the contour line. And

Another object of the present invention is to provide a contour extraction method and a storage medium which can accurately extract the contour of an elongated object.

[0010]

A contour line extracting method according to the present invention includes an edge detecting step of detecting an edge intensity from an input image, and an edge intensity detecting method for detecting the edge intensity according to a magnitude of a weight coefficient determined based on predetermined information. And calculating a corrected edge strength, calculating a corrected edge strength, a starting point setting step for setting a first point of interest for contour tracking, and selecting a neighboring pixel having the largest corrected edge strength value as a next contour point. An extraction step of extracting a contour line by repeating contour tracing.

[0011] The storage medium according to the present invention stores the program software of the above-described contour line extraction method.

[0012]

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

FIG. 1 shows a flowchart of one embodiment of the present invention. First, image data to be processed is loaded into an image memory or the like (S1). An edge component is detected from the image data loaded in the image memory (S2).
As an edge detection method, for example, a first derivative operator such as a Sobel operator shown in FIG. 2 is used. FIG. 2A shows a small image area composed of 3 × 3 pixels, and A to I are image values at each pixel. FIG. 2 (b)
2 is a mask for detecting a horizontal edge component Gx, FIG.
(C) shows a mask for detecting a vertical edge component Gy. The edge components Gx and Gy are respectively calculated according to the following equations. That is, Gx = (A + 2B + C) tens (G + 2H + I) (1) Gy = (A + 2D + G) tens (C + 2F + I) (2) The product-sum operation of the equations (1) and (2) is performed on the entire image or the area where the contour line is extracted. Limited execution. The edge strength E and the edge direction θ are defined by the following equations using the edge components Gx and Gy. That is, E = (Gx × Gx + Gy × Gy) ^1/2 (3) θ = arctan (Gy / Gx) (4) Equation (3) may be simplified as follows. That is, E = | Gx | + | Gy | (5) The operation of Expression (4) can be accelerated by utilizing a read-only memory (ROM).

As shown in FIG. 3, an edge may be detected by using a mask group obtained by rotating the coefficient of the first-order differential operator. The eight masks (Robinson operators) shown in FIGS. 3A to 3H detect edges with different directivities of 45 degrees each. These eight masks are equally convolved with image data, and the edge strength obtained from the mask having the largest output value and the direction of the edge indicated by the mask are defined as the edge strength and direction of the pixel of interest. In this case, the calculation of Expression (4) can be omitted.

Next, a pixel (control point) for starting and ending contour line tracking is designated (S3). The start pixel is the first target pixel. The contour is tracked from the start pixel to the end pixel. If the start pixel and the end pixel match, the contour is determined as a closed curve. The start and end pixels for contour tracing are usually specified manually by the operator, but for simple tasks such as contour tracing of binary images, the designation of start and end pixels can be automated.

A corrected edge intensity is calculated by multiplying the edge intensity E by a homogeneity coefficient determined based on the distribution of the edge direction θ (S4). FIG. 4 shows a detailed flowchart of step S4.

FIG. 5 is an input image of a composition in which a camera is placed on a carpet. FIG. 6 shows the distribution of the intensity E of the edge in the rectangular frame in FIG. In FIG. 6, the vertical axis indicates the edge strength, and the horizontal axis indicates the horizontal position in the rectangular frame. FIG.
In US Pat. No. 6,073,072, the edge strength is detected using a Sobel operator. The edge strength of the background portion is large, and the outline of the camera is lost in the background edge. FIG. 7 shows the distribution of the edge direction θ in the rectangular frame in FIG. In FIG. 7, the distribution in the edge direction θ is represented by a vector having a constant length.
The vectors in the background are lacking in unity and are inhomogeneous, but the vectors in the outline are aligned in a certain direction, and the homogeneity is high.

In FIG. 4, first, the distribution in the edge direction θ is smoothed (S11). Since the value of the edge direction θ is discontinuous at both ends of the existence definition range (0 to 2π or −π to + π, etc.), the cosine value cos θ and the sine value sin θ of θ are used during smoothing.
First, for each of the cosine and sine values,
For example, a convolution operation of a mask of 3 × 3 pixels having an equal weighting coefficient is performed. FIG. 8 shows an example of the result of the smoothing process. In the background part where the edge direction θ was different,
Both the cosine value and the sine value cancel each other out with the surrounding cosine value and the sine value and become small values. Conversely, in a contour portion in which the directions of edges are aligned, this is equivalent to a simple addition process, and the result is a relatively large value.

Next, for each pixel, the degree of uniformity of the direction of the edge and the edge of the neighboring pixel, that is, the homogeneity of the edge direction is determined by a cosine value (average value of neighboring pixels) and a sine value after the smoothing process. Evaluation is performed based on the magnitude of (average value of neighboring pixels), and the homogeneity coefficient is determined (S1).
2). The most basic homogeneity coefficient h is obtained by the following equation. That is, h = (average value of cos θ) ² + (average value of sin θ) ² (6) The homogeneity coefficient h indicates the vector length after the smoothing process shown in FIG.

FIG. 9 shows that the homogeneity coefficient h is 0 to 255.
Is scaled to the range of and graphed. The vertical axis indicates the homogeneity coefficient, and the horizontal axis indicates the position in the horizontal direction. A pixel having higher homogeneity in the edge direction has a larger homogeneity coefficient, and a non-uniform pixel has a smaller homogeneity coefficient.

The corrected edge strength E is obtained by multiplying the obtained homogeneity coefficient h at least once by the previously obtained edge strength E (S13). In this embodiment, the edge strength E is multiplied by the cube of the homogeneity coefficient h to obtain a corrected edge strength E ′. That is, E ′ = h ³ × E (7) An example of the corrected edge strength E ′ is shown in FIG. The vertical axis indicates the corrected edge intensity E ′, and the horizontal axis indicates the position in the horizontal direction.
In FIG. 10, the contour lines clearly stand out.

Similar processing examples are shown in FIGS. FIG. 11 is a low-contrast input image of a composition in which a person stands against a white wall. FIG. 12 shows an edge intensity distribution in the rectangular frame in FIG. FIG. 13 shows the distribution of the homogeneity coefficient. FIG. 14 shows an edge intensity distribution corrected by the homogeneity coefficient. 12, 13, and 1
4 corresponds to FIGS. 6, 9 and 10, respectively. In FIG. 14, a low-contrast outline is prominent.

The homogeneity coefficient h can of course be used not only for multiplication but also for addition or non-linear conversion according to the magnitude of h.

Returning to FIG. After calculating the corrected edge strength E 'in this way, a search is made for a pixel having the maximum corrected edge strength E' near the pixel of interest (S5). The search can be made more efficient by applying a mask that limits the search range. A pixel having the maximum corrected edge intensity E 'is set as the next target pixel (S6). It is evaluated whether or not the new target pixel satisfies the contour ending condition (S7). If the ending condition is not satisfied, the contour tracing is continued (S5). If the ending condition is satisfied, the contour tracing is terminated.

This embodiment has the following effects. That is, it is possible to effectively suppress a strong edge component having randomness in the vicinity of the contour. Extremely low-contrast outlines that blend into a flat background can be enhanced. Since all the corrected edge intensities have been calculated before the contour tracking step (from step S5), efficient and high-speed contour extraction can be realized.

In the above embodiment, the contour line is separated from its neighboring region, paying attention to the homogeneity in the edge direction. In the second embodiment, information on the edge direction in each pixel is more directly utilized. Specifically, the contour is tracked with a pixel in the vicinity of the target pixel, which has a high edge strength and whose edge direction is the same as the edge direction of the target pixel, as the next target point.

FIG. 15 shows a small image area composed of 3 × 3 pixels. In this small image area, a hatched line 5 is written.
The pixels are limited in the field of view, and serve as an inaccessible area for contour tracking. The vectors v ₀ , v ₁ , v ₂ , and v ₃ drawn on the remaining four pixels are represented by the horizontal edge component Gx shown by the equation (1), that is, the first-order differential value in the vertical direction, and the equation (2) ) Is a gradient vector determined by the vertical edge component Gy, ie, the first-order differential value in the horizontal direction. The direction of the gradient vector and the direction of the edge are orthogonal. The center vector v ₀ is a gradient vector at the pixel of interest. The angles α, β, and γ are angles formed by the gradient vector between the target pixel and each of the neighboring pixels.

In this embodiment, as shown by the following equation (8), a pixel having a large projected component of the gradient vector of the neighboring pixel onto the gradient vector of the target pixel is selected as the next target point. That is, max (| v ₁ | cos α, | v ₂ | cos β, | v ₃ | cos γ) (8) Using the equation (3), the equation (8) is changed as follows. That is, max (E ₁ cos α, E ₂ cos β, E ₃ cos γ) (9) According to the definition of the inner product, the operation of equation (9) is performed by using the gradient vector v ₀ of the target pixel and the gradient vector of the neighboring pixel.
This is realized by the inner product operation of the vectors v ₁ , v ₂ , v ₃ . The inner product operation is performed by the following expression (10) instead of the operation of the expression (9).
Is equivalent to performing the operation of That is, max (E ₀ E ₁ cos α, E ₀ E ₂ cos β, E ₀ E ₃ cos γ) (10).

[0029] Since the edge intensity value E ₀ equally target pixel in all the compared elements are applied, there is no effect on the result of the magnitude comparison.

The features of the second embodiment will be described. FIG.
(A) shows a cross section of an elongated object of one pixel width having a flat background. FIG. 16B shows that the object having one pixel width is 1
3 shows edges detected by a second derivative type operator. The edge detected at the rising edge of the image density and the edge detected at the falling edge of the image density are distinguished by a difference in sign. FIG. 16C shows the absolute value of the detected edge. Since normal contour tracing is based on edge strength, two close contours as shown in FIG. 16C are often mixed during tracing.

FIG. 17 is a schematic diagram showing an enlarged view of the above-mentioned one-pixel-width object and its surrounding gradient vectors in a small area of an image composed of 5 × 5 pixels. As described above, since the contour is tracked by comparing the inner product operation values of the gradient vectors with each other, for example, when the contour on the left side of the one-pixel width object shown in FIG. The gradient vector of the right contour of the object is in the opposite direction,
The cosine value cosθ between the two vectors is the minimum value -1 or -1
It is a value close to. Therefore, it is unlikely that the right contour is started to be traced while the left contour is being traced.

FIG. 18 shows an example of a gradient vector for an image in which a left half shows an object and a right half shows a background in a small area of an image composed of 5 × 5 pixels. There is a strong linear edge in the background, orthogonal to the boundary between the object and the background. Now, consider tracing the contour in the order of pixel numbers 1, 2, 3, 4, and 5. At the location of pixel number 2, the course is either 3 or 3 '. If the contour is traced only depending on the edge strength as in the conventional example, the background edge having a higher edge strength is drawn, and the process proceeds to the pixel number 3 '. However, in the present embodiment, the magnitudes of the inner product values of the gradient vectors are compared as described above. The gradient vector at the pixel number 2 and the gradient vector at the pixel number 3 'have an orthogonal relationship, and the cosine value is 0 or a value close to 0. Therefore, in the present embodiment, the inner product value also becomes a small value, and the pixel number 3 ′ is not erroneously selected as the next point of interest.

The calculation method of the gradient vector is as follows.
Any method based on the difference between neighboring pixels may be used,
The invention is not limited to the Sobel operator described in the first embodiment.

The second embodiment has the following effects.
That is, the direction of the gradient vector is taken into consideration, and two adjacent contour lines can be clearly distinguished, so that a cape or a cove-like elongated contour whose contour lines protrude at an acute angle can be accurately extracted. Since it has a straight-line contour tracking characteristic, it is strong in extracting a straight contour, and is hardly affected by strong edges that intersect with the contour. Since the first derivative in the horizontal and vertical directions of the image density obtained in the process of calculating the edge strength is the gradient vector itself, the conventional edge strength-based contour tracing or the first embodiment can be performed without additional calculation. It is easy to combine.

Next, a third embodiment of the present invention will be described. In the second embodiment, when calculating the inner product of the gradient vector, the positional relationship between the target pixel and its neighboring pixels is not taken into account. However, in the third embodiment, the positional relationship between the target pixel and its neighboring pixels is determined. Also consider.

FIG. 19 shows a small image area composed of 3 × 3 pixels. A hatched portion 5 is written in this small image area.
The pixel is defined as an entry prohibited area for contour tracking. Vectors v ₀ , v ₁ , v ₂ , v drawn by broken lines in the remaining four pixels
₃ indicates a gradient vector at each pixel. These gradient vectors v ₀ , v ₁ , v
Vectors e ₀ , e ₁ , e ₂ , e orthogonal to ₂ and v ₃
Reference numeral ₃ denotes an edge vector determined by the horizontal edge component Gx represented by Expression (1) and the vertical edge component Gy represented by Expression (2). The gradient vector and the edge vector have the same length and are orthogonal.

The center vector e ₀ indicates an edge vector at the pixel of interest. The angles α, β, and γ indicate the angles formed by the edge vector e ₀ of the target pixel and the displacement vectors A, B, and C connecting the target pixel and the center positions of the neighboring pixels. The displacement vectors A, B, and C are (0, 1), respectively.
(1,1) and (1,0). In FIG. 19, 4
The three edge vectors e ₀ , e ₁ , e ₂ , and e ₃ are all aligned in substantially the same direction, and the vector lengths, that is, the edge intensities are also approximately equal. In this case, only the method of conventional edge strength based contour tracing method or the second embodiment, it is not possible to obtain a certain course, here, the outline tracing along the edge direction e ₀ in the target pixel Considering the most natural, the edge vector e ₂ is replaced by the edge vector e ₁ ,
compared with e ₃ performs weighting such that priority.

More specifically, according to the following equation (11) for edge intensity-based contour tracing, and to the following equation (12) for inner product value-based contour tracing, angles α, Weighting processing is performed according to β and γ. That is, max (| e ₁ | cos α, | e ₂ | cos β, | e ₃ | cos γ) (11) max (e ₀ e ₁ cos α, e ₀ e ₂ cos β, e ₀ e ₃ cos γ) (12) .

The calculation of the cosine value in the equations (11) and (12) is realized by the inner product calculation as in the second embodiment. According to the definition of the inner product, each cosine value in the equations (11) and (12) can be changed as in the following equation (13). _{That, cosα = (A · e 0} ) / (| A || e 0 |) = e y / | e 0 | cosβ = (B · e 0) / (| B || e 0 |) = (e x ^{_{+ e y) / (2 1/2}} | e 0 |) cosγ = (C · e 0) / (| C || e 0 |) = e x / | e 0 | (13) _where, e 0 = (e _x , _ey ).

All the elements to be compared are the edges of the pixel of interest.
Strength value | e₀| Is divided equally, so the edge strength
Value | e₀| Does not affect magnitude comparison. Therefore, the actual performance
The calculation is performed at the pixel of interest according to the positional relationship with the pixel of interest.
Component e of the edge vector _x, E_yThe corresponding neighborhood
Just multiply by the raw edge vector. This is a note
The displacement vector A of the edge vector at the eye pixel,
Equivalent to using the amount of the projected component to B and C as the weighting factor
No. For example, if there is a strong edge in the horizontal direction,
Horizontal component e of the elementary edge vector_xIs vertical
Minute e_yFrom equation (13), e_x/ | E
₀Is relatively large, and the edge vector
Pixel to the right along the direction (horizontal direction)
Weighting is performed. Similarly, strong edges in the vertical direction
If there is, the vertical component of the edge vector of the pixel of interest
e_yIs the horizontal component e_xEquation (13)
By e_y/ | E₀Becomes relatively large,
The upper pixel along the direction of the vector (vertical direction)
Weighting is performed so that priority is given. Strong in diagonal direction
The same applies when there is an edge. Attention pixel of contour tracking
In order to change the sensitivity to the edge direction in the
Component e of_x, E_yIs changed by, for example, 1/2 power or square power.
You may make it.

In this manner, the priority is given to the pixel in the direction along the edge direction in the target pixel, so that smooth contour tracking with good connection can be realized.

The present invention may be applied to a configuration including a plurality of devices, or may be applied to a configuration including a single device.

Further, in order to operate various devices so as to realize the functions of the above-described embodiment, software for realizing the functions of the above-described embodiments is installed in a computer in an apparatus or a system connected to the various devices. The present invention also includes a case in which a program (code) is supplied and a computer (CPU or MPU) of the apparatus or system is operated by operating the various devices according to stored programs.

In this case, the program code itself of the software realizes the function of the above-described embodiment, and the program code itself and means for supplying the program code to the computer, for example,
A storage medium storing such a program code constitutes the present invention. As a storage medium for storing such a program code, for example, a floppy disk, hard disk, optical disk, magneto-optical disk, CD-ROM, magnetic tape, nonvolatile memory card, ROM, or the like can be used.

When the computer executes the supplied program code, not only the functions of the above-described embodiment are realized, but also the OS (operating system) or other operating system running on the computer. Even when the functions of the above-described embodiments are realized in cooperation with application software and the like, it goes without saying that such program codes are included in the embodiments of the invention according to the present application.

Further, the supplied program code is:
After being stored in the memory provided in the function expansion board of the computer or the function expansion unit connected to the computer, the CPU or the like provided in the function expansion board or the function expansion unit executes one of the actual processing based on the instruction of the program code. It is needless to say that a case where the functions of the above-described embodiments are realized by performing all or part of the processes and executing the processing is also included in the invention according to the present application.

[0047]

As can be easily understood from the above description, according to the present invention, a corrected edge strength is calculated by multiplying an edge strength by a weight coefficient determined based on information on an edge direction detected from an image. , Because the next pixel with the highest corrected edge strength value is selected as the next contour point, the effect of strong edge components other than the contour near the contour can be suppressed, and as a result, the contour extraction that is strong against noise and low contrast Can be realized.

[Brief description of the drawings]

FIG. 1 is a flowchart of a first embodiment of the present invention.

FIG. 2 is a diagram showing a sobel operator.

FIG. 3 is a diagram showing a Robinson operator.

FIG. 4 is a flowchart illustrating a procedure for calculating a corrected edge strength.

FIG. 5 is an example of an input image for calculating a corrected edge strength.

6 is a schematic diagram of an example of a lateral edge intensity distribution by a Sobel operator in a rectangular frame of the input image shown in FIG. 5;

FIG. 7 is a schematic diagram of a distribution of an edge direction θ in a rectangular frame in FIG. 5;

FIG. 8 is an example of a result of a smoothing process.

FIG. 9 is a schematic diagram illustrating a distribution example of a homogeneity coefficient h.

FIG. 10 is a distribution example of a corrected edge intensity E ′.

FIG. 11 is another example of an input image.

FIG. 12 is a schematic diagram showing an edge intensity distribution in a rectangular frame of the input image shown in FIG. 11;

13 is a schematic diagram showing a distribution of homogeneity coefficients in a rectangular frame of the input image shown in FIG.

FIG. 14 is a schematic diagram showing an edge intensity distribution corrected by a homogeneity coefficient.

FIG. 15 is a schematic diagram showing a small image area composed of 3 × 3 pixels.

FIG. 16 is an explanatory example of the second embodiment.

FIG. 17 is an enlarged schematic view of an object having a width of 1 pixel and a gradient vector around the object in a small area of an image composed of 5 × 5 pixels.

FIG. 18 is an example of a gradient vector for an image in which a left half shows an object and a right half shows a background in a small area of an image composed of 5 × 5 pixels.

FIG. 19 is an example of a small image composed of 3 × 3 pixels for explaining the third embodiment.

 ────────────────────────────────────────────────── ─── Continued on the front page (72) Inventor Fumiaki Takahashi 3-30-2 Shimomaruko, Ota-ku, Tokyo F-term (reference) in Canon Inc. 5B057 AA11 BA25 CA08 CA12 CB08 CB12 CD03 CE02 CE03 CE06 CH09 CH11 DA13 DB02 DC16 5C076 AA13 AA24 AA31 BA04 BB04 BB25 5C077 LL02 PP03 PP15 PP22 PP23 PP47 PP54 PQ12 PQ22 SS02 5L096 AA06 BA07 DA02 EA05 EA23 EA39 FA06 FA45 FA54 LA05

Claims (7)

[Claims] An edge detecting step of detecting an edge intensity from an input image; and a corrected edge intensity calculating step of calculating a corrected edge intensity obtained by correcting the edge intensity in accordance with a magnitude of a weight coefficient determined based on predetermined information. An extraction step of extracting a contour line by repeating a contour tracing step of selecting a neighboring pixel having the largest value of the corrected edge strength as a next contour point; And a contour line extracting method. 2. The method according to claim 1, further comprising an edge direction detecting step of detecting an edge direction from the input image, wherein the correcting edge strength calculating step evaluates the homogeneity of the distribution of the edge direction, and based on the homogeneity. 2. The contour line extracting method according to claim 1, wherein the correction edge strength is calculated according to the magnitude of the determined homogeneity coefficient. 3. The method according to claim 1, further comprising a gradient vector detecting step of obtaining a density gradient distribution from the input image, wherein the correcting edge intensity calculating step includes the step of calculating a gradient vector of the pixel of interest and each gradient of pixels in the vicinity of the pixel of interest. Cosine value co of angle θ between vector
2. The contour extraction method according to claim 1, wherein sθ is calculated, and a corresponding edge strength of a pixel adjacent to the pixel of interest is corrected according to the obtained cosine value to calculate a corrected edge strength. 4. The method according to claim 3, wherein the correction of the edge intensity according to the cosine value cosθ is realized by an inner product operation of a gradient vector of the pixel of interest and a gradient vector of a pixel near the pixel of interest. Contour extraction method. 5. An edge direction detecting step for detecting an edge direction from the input image, wherein the correcting edge strength calculating step includes determining a positional relationship between the edge direction and a neighboring pixel of the target pixel in the target pixel. 2. The contour line extracting method according to claim 1, wherein the edge strength of the neighboring pixel is corrected according to a weight coefficient determined based on the calculated edge strength. 6. The contour line extracting method according to claim 5, wherein a weighting coefficient is larger for a pixel closer to the target pixel located in a direction along the edge direction of the target pixel. 7. A storage medium storing program software for the contour line extraction method according to claim 1.