JPH05210736A - Method for extracting contour of image - Google Patents

Abstract

PURPOSE:To recognize a contour position highly accurately coincident with a practical contour position by registering edge components on picture element positions existing in a fixed direction in respective areas of picture element boundaries having density differences. CONSTITUTION:An image to be processed is divided into several areas and an operator appropriate for the status of an image to be processed in each divided area is applied to the area. Namely when a personal image is divided into right and left parts through its center, edge components detected on the right, lower and lower right picture element positions of respective image boundaries having density differences are registered in the left half image and edge components detected on the left, lower and lower left picture element positions of respective boundaries having density differences are registered in the right half image. In this case, a difference of density values between a remarked picture element and its adjacent picture element is found out, and when the absolute value of the difference exceeds a prescribed value, the remarked picture element is recognized as an edge component and an edge mark is registered in the remarked picture element.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for extracting an edge component of an image, and more particularly to an image contour extracting method useful for recognizing a human image, emphasizing an image and other image recognition.

[0002]

2. Description of the Related Art In extracting an edge of an image, the edge component of the image to be extracted is extracted. As a method of extracting the edge component of the image, see "Digital Image Signal Processing" page 106 published by Modern Science Co., Ltd. As shown in FIG. 10, a directional differentiation operator and a Laplacian (second derivative) operator, which is an isotropic differentiation operator, are disclosed.

Now, the image on the discrete plane is f (x, y)
(Where x and y are integers), the edge strength of the image can be calculated by adding the weights corresponding to the density values of the pixel (x, y) and the pixels in the vicinity thereof and adding them together. .. An array of weights is called an operator.

The operator first introduced into the method of extracting the edge component is Roberts 'operator shown in Equation 1 and Equation 2 or Equation 1'and Equation 2', and this operator has a size of 2 × 2. In the diagonal direction (45 °, 1
The density change Δ of 35 °) is detected. Δ45 = f (x + 1, y + 1)-(x, y) (1) Δ135 = f (x, y + 1)-(x + 1, y) (2) or

[0005]

[Equation 1]

As shown in FIG. 10, the directional differential operator detects edges in eight directions by eight operators provided corresponding to the direction of the edge, and the direction having the highest edge strength is the edge direction of that point. The edge strength at that point is defined as the edge strength at that point. The Laplacian (second derivative) operator is an isotropic operator.

[0007]

According to the conventional edge extraction method, an edge component is generated on one side or both sides fixed on the dark side or the light side of the actual contour (density boundary) depending on the calculation method. It is considered that the actual contour exists not between pixels but between two adjacent pixels.

Therefore, when there is an object having different shades in the background of which density is almost constant, according to the conventional method of calculating the edge component, in a part, the inside of the actual contour of the object is detected. In addition, an edge component is generated outside the actual contour of the object in a certain portion.

For example, when an image having a density distribution of each pixel as shown in FIG. 11 is considered, if a boundary between a pixel area having a certain density and a pixel area having another density is an outline portion, a certain calculation method is used. Then, since the edge component is registered in the pixel with the higher density, the edge component is registered in the pixel at the pixel position marked with a white circle in FIG. 11, and as a result, the pixel density of the object is In the portion lighter than the background density, the edge component is registered in the pixel in the background portion, not in the pixel on the object side.

If image recognition is performed based on this edge component and, for example, the background portion is filled with an arbitrary color, a gap is created between the object and the filled portion at a part of the contour. May occur (Fig. 12).

That is, since the density of the pixel of the object is lighter than the density of the background pixel even though it is the pixel of the area that is originally the background portion, the edge component is registered and the contour is registered. Since it is processed as a line pixel, the pixel in this portion is not filled and remains in the original color.

As described above, in the case of automatically recognizing the entire area other than the object, that is, the background portion, if the pixels in which the edge component is registered are not constant inside or outside the object, Depending on the situation, there is a possibility that a part having a width of about one pixel may be left behind. Then, when such a situation occurs, the background becomes unevenly filled, and the quality of the obtained image is impaired.

In particular, since it is desired that a person image used for an ID card or the like has a high image quality, it may be a problem that when the background portion is automatically recognized, uneven filling occurs between the person image and the background portion. Become.

Therefore, the object of the present invention is to
Depending on the situation of the image, it is possible to prevent the registration of edge components from becoming irregular inside or outside the actual contour in some places, and only inside the actual contour of the object. Provide an image contour extraction method that enables edge components to be registered either on the outer side or on the outer side only so that the contour position that matches the actual contour position can be recognized with high accuracy. To do.

[0015]

In order to achieve the above object, the present invention is as follows. That is, as a method of extracting an image contour by performing data processing on an edge component extraction operator for a digitized two-dimensional image, a dividing step of dividing an image to be processed into a plurality of rectangular areas and this dividing step. The edge component extraction operator for extracting the edge component from the difference value between the pixel of interest and the density value of each pixel around the pixel of interest is set in advance for each divided area to obtain the edge component. An edge component extracting step of extracting an edge by registering an edge in the pixel of interest.

The edge extraction operator targets 2 × 2 pixels, one of which is the target pixel, and the maximum absolute value of the difference between the target pixel and each of the three pixels adjacent to the target pixel is the maximum value. An edge component that exceeds a predetermined value is returned.

[0017]

According to the present invention, the image to be processed is divided into several areas, and an operator suitable for the situation of the image to be processed in the area is applied to each of the divided areas, so that the actual object is processed. The calculation method is adopted so that the edge component occurs in the pixel immediately inside the contour.
For example, since the human image is substantially bilaterally symmetric, when the human image is divided into the left and right from the center, the right side of the edge component is the left image and the left side of the edge component is the right image. I can think.

Therefore, as a method of calculating the edge component, regardless of which side of the edge has a high density and which side has a low density, the edge component is always generated in a fixed direction (lower left direction, lower right direction). The method of letting is adopted.

That is, when the human image is divided into the left and right from the center, the edge components detected are registered at the right, lower, or lower right pixel positions with respect to the image boundary where there is a density difference in the left half image. In the right half image, the edge components detected are registered at the pixel positions on the left, the bottom, or the bottom left with respect to the pixel boundary having the density difference.

Here, the difference in density value between the pixel of interest and each of its adjacent pixels is obtained, and if the absolute value of the pixel is maximum and exceeds a predetermined value, it is recognized as an edge component and the edge is registered in the pixel of interest. ..

As described above, the edge components are registered on both sides of the actual contour by registering the edge components at the pixel positions in a constant direction in each area with respect to the pixel boundary having the density difference. There is an action to prevent that. As a result,
It is possible to obtain a sharp contour that matches the actual edge position with high accuracy.

[0022]

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

(Embodiment 1) FIG. 1 shows a digital image of a person created for use in an ID card or the like. This digital image can be obtained by digitizing a video signal obtained by photographing a person with a video camera with an A / D converter or reading a person photograph with an image scanner and digitizing the photograph. Since it is desired that a person image used for an ID card or the like is of high quality, it is a problem that unevenness of illumination or stains on the wall are reflected in the background portion of the person image.

Various image processing can be performed on the digital image by using a computer. An image having a problem in the background portion as described above can also be used if it is corrected by image processing. In the case of an ID image, the background portion may be filled with a single color or replaced with an arbitrary pattern.

For that purpose, first, the background portion must be recognized. Here, it is important to match the recognized background range with the actual background area with high accuracy.
Otherwise, the vicinity of the boundary of the object (target image) in the corrected image will be unnatural. Although there are various image recognition techniques, the method of performing edge extraction is generally used.

In this embodiment, two operators are used to extract the edge component. One is for the left half of the person image, and the other is for the right half. Therefore, the operator for the left side and the operator for the right side will be respectively referred to hereinafter. Here, the left half will be described.

The left operator is shown in FIG. The meaning of each character in the figure is different from the above-mentioned operator, etc., and the density data of each pixel corresponding to the operator when superimposed on the image.
Represents the data.

Of the four pixel positions P1 to P4, P
The pixel position of 4 is the pixel for which the edge component calculation is performed. That is, the density data of the pixel at the P4 position to be calculated is "x", the density data of the pixel at the upper left pixel position P1 is "a", and the density data of the pixel immediately above the pixel position P2 is "b". ”Means that the density data of the pixel at the left pixel position P3 is represented as“ c ”. The edge component E of the pixel at the P4 position is calculated as follows.

[0029]

[Equation 2]

That is, when the pixel of interest and the three pixels (P1 to P3) on the upper left, left and upper thereof are considered as one block,
Check the density difference of each of the three pixels with respect to the pixel of interest,
The edge strength of the pixel of interest (edge-likeness, density-like difference portion likeness) indicates the largest density difference.

If the pixel is recognized as an edge, that is, if the pixel has an edge strength of a predetermined value or more, the edge component is registered in the target pixel. The position of the pixel of interest is moved by one pixel at a time, and the edge strength is checked for all the pixels on the left half of the image. Note that the edge registration is performed for an edge registration memory of the same size as the image.

As a result, the edge extending horizontally is located below the edge of the edge, the edge extending vertically is located to the right of the edge, and the edge of the diagonal line rising to the right is located below the edge. It is registered in the lower right pixel (see FIG. 3).

Since the maximum absolute difference is used, the operator is sensitive to edges. It is also a relatively high-speed operator because it is not necessary to check 8 directions but 3 directions.

On the other hand, the person who is the object in the person image is generally located at the center of the screen. Therefore, when the screen is divided from the center to the left and right, the target object is closer to the lower right in the image on the left. The lower right side of the outline of the object is the object, and the upper left side is the background portion. When the aforementioned left-side operator is applied to such a left-side image to extract the edge component, the edge component is registered almost only inside the object (see FIG. 5).

By recognizing the range from the upper left portion of the left side image to the pixel before the first edge component as the background portion, the object and the background portion can be vividly or clearly separated ( (See FIG. 6).

Then, by filling the part recognized as the background with a uniform color or replacing it with an arbitrary pattern, it is possible to eliminate problems such as uneven illumination existing in the background part. Moreover, since the recognized background range and the actual background range are in good agreement, it is possible to make a correction with less discomfort.

The same applies to the right half, but the operator for the right side shown in FIG. 2B is used as the operator. In this case, the position of the pixel of interest is 2 ×
Of the pixels in the block having the two-pixel matrix structure, the description is the same as the above description for the left-side operator except that it is at the P3 position. However, since the pixel of interest is at the P3 position, the vertically extending edge is registered in the pixel to the left of that edge (see FIG. 4). for that reason,
In the contour on the right side of the object, edges are almost never registered in pixels located outside the object except for the lower contour.

Therefore, when the background portion of the image is filled in such a manner that the person image as the object is located at the center of the image and the surroundings become a monotonous background portion, such as the person image on the ID card, For the right half area of the image, the operator for the right side was used, and for the left half area of the image, the operator for the left side was used. It becomes possible to make it difficult to register the edge, and it becomes possible to minimize the unpainted portion of the background portion.

(Embodiment 2) In Embodiment 1 described above, the image is simply divided into the left and right regions, so that there is an advantage that the processing is simple, but the effect of edge extraction is fully achieved in a part of the image. It cannot be said that That is, there is a possibility that the edge registration may be performed on the pixel on the background side in a part of the image for the following reason.

That is, in the person image, the edge direction in the portion mainly from the ear to the neck is approximately right downward in the left image and approximately left downward in the right image. However, these cannot be sufficiently effectively processed by the left operator and the right operator described in the first embodiment.
Because, the operators desired in these parts are the operators who register the edges to the upper right and the upper left, respectively, in reality, but in reality, they move to the lower right,
This is because the edge component is registered in the lower left direction. In this embodiment, as a measure against this problem, the image is divided into six areas, and an operator suitable for each is used.

First, an edge is extracted by a general edge extraction method as shown in FIG. 7, and the widest portion W1 of the person's head and the narrowest portion W2 of the neck are recognized. The image is first vertically divided into three parts by horizontally dividing the y parts of the two parts W1 and W2. Further, the image is vertically divided at the center to be horizontally divided into two.

As a result, as shown in FIG. 7, the image is divided into six areas and divided into areas A1 to A6. The areas A1 to A6 divided in this way will be referred to as first to sixth areas. The operators used in the respective areas A1 to A6 are shown in FIGS. Calculation formula is second
It is the same as Formula 1.

The operators shown in FIGS. 9A and 9B are the same as the operators used in the first embodiment and have the above-mentioned characteristics. The operators shown in (c) and (d) of FIG. 9 are operators newly added in this embodiment, and (c) shows P among blocks having a configuration of 2 × 2 pixels.
The pixel at the two positions is set as the pixel of interest, and P1, P3, P4
The density difference with the pixel at each pixel position is calculated, and the one with the maximum density difference is calculated as the edge strength, and (d)
Is a block of 2 × 2 pixels, the pixel at the P1 position is the pixel of interest, the density difference between this pixel and the pixels at the pixel positions of P2, P3, and P4 is calculated, and the pixel with the maximum density difference is the edge. Is an operator for performing a process of registering an edge in a pixel of interest when the edge component is recognized as an edge component.

If the method of the first embodiment is adopted, the operator in (c) has a characteristic that the edge component is registered in the upper right part of the edge, and the operator in (d) is registered in the upper left part of the edge. It has the characteristics to say.

As a result, by adopting the operator of FIG. 9, the operator of (a) can register the edge component in the lower right part of the edge, and the operator of (b) can register the edge component in the lower left part of the edge. Can be registered, the operator of (c) can register the edge component to the upper right of the edge, and the operator of (d) can register the edge component to the upper left of the edge.

From the characteristics of the person image of the ID card, the first area (A1), the third area (A3); The edge component is registered near the lower right of the edge. Second area (A2): An edge component is registered near the upper right of the edge. Fourth area (A4), sixth area (A6): An edge component is registered near the lower left of the edge. Fifth area (A5): An edge component is registered near the upper left of the edge. You can do what you said.

Therefore, for the first area (A1), the operator of FIG. 9 (a), for the second area (A2), the operator of FIG. 9 (c), and for the third area (A3). 9) for the fourth area (A4), (d) in FIG. 9 for the fifth area (A5). ) Are processed using the operators in FIG. 9B for the sixth area (A6), respectively, so that the edge components are registered on the inner side of the person who is the object as a whole. Will be able to. The rest of the procedure is the same as in the first embodiment.

Since the four types of operators used here are designed to extract edge components in three directions centered on the diagonal direction, they are said to be suitable for processing the regions separately vertically and horizontally. There are also advantages. If the operator extracts edges in only three directions (for example, left, upper left, and lower left) centering on the vertical and horizontal directions, the area must be divided by 45-degree diagonal lines. .. Otherwise, there may be a portion where the sensitivity to the edges in the vertical direction and the horizontal direction becomes weak.

(Embodiment 3) In the second embodiment, first, a locally wide portion and a narrow portion of the width of the person in the image are detected, and the image is vertically divided into three at this position. This is not a problem if the person is in line symmetry, but if the person is asymmetric, there is a concern that an edge may be registered in the background side pixel depending on the situation. In this embodiment, the extraction of the geometrical feature of the image is further advanced by one step, and the following division is performed, so that more appropriate processing can be performed.

The third embodiment will be described with reference to FIG. First, the pixel P at the highest position on the top of the contour of the person
find a. This may be found by a method similar to that adopted in the second embodiment.

If found, the image is divided into two by dividing it by a vertical boundary line LY adjacent to the pixel Pa. Next, find the pixels Pb to Pe having the local maximum and minimum x-coordinates near the ears and near the neck. If found, the horizontal boundary line L adjacent above or below these pixels
Partition at X1 to LX4 and divide vertically. The dividing line at this time is stopped when the vertical boundary line LY is reached. As a result, the image is divided into six areas A1 'to A6'.
The following procedure is the same as in the second embodiment.

According to this division method, the area A shown in FIG.
Since the edges 1 ′, A3 ′, and A5 ′ are rising to the right and the areas A2 ′, A4 ′, and A6 ′ are descending to the right, the four types of operators shown in the second embodiment (FIG. 15) are shown. Can effectively detect edges.

As described above, according to the present invention, the conventional edge formation registered on both sides of the contour line is registered only on one side of the contour line. It is possible to obtain the advantage that it is possible to recognize the contour position that is in good agreement with. Further, since the operator used for edge extraction has limited edge directions to be detected, there is an effect that it is possible to perform the search in a shorter time than an operator who searches all directions. Further, since the maximum edge component is detected among the vertical, horizontal, and diagonal directions, there is also an effect that edge detection with high sensitivity can be performed.

It is needless to say that the present invention is not limited to the above-described embodiment, and can be appropriately modified and implemented without changing the gist of the present invention. For the right side is for the right side, and
The left side was processed by the operator for the left side. The right side of the image divided into two is the operator for the fifth area in FIG. 9, and the left side is the operator for the second area in FIG. By performing the processing, the edge component can be registered only outside the actual contour of the object.

[0055]

As described in detail above, according to the present invention,
Depending on the situation of the image, it is possible to prevent the registration of the edge component from becoming an irregular situation such as being inside or outside the actual contour depending on the situation. An image contour extraction method capable of registering an edge component only on the inner side or only on the outer side so that a contour position that matches an actual contour position can be recognized with high accuracy. Is obtained.

[Brief description of drawings]

FIG. 1 is a diagram showing an example of a processing target image for contour extraction.

FIG. 2 is a diagram showing an example of an operator used in Example 1 of the present invention, in which (a) is an operator for the left side,
FIG. 6B is a diagram showing a right operator.

FIG. 3 is a diagram for explaining an edge registration pixel position detected by an operator for the left side of FIG.

FIG. 4 is a diagram for explaining an edge registration pixel position detected by an operator for the right side of FIG.

FIG. 5 is a diagram showing an example of an edge component registration state when a left-side operator is applied to a left-side image of a human image on a plain background.

6 is an image after the edge component registration shown in FIG.
The figure which showed visually the case where the background part was isolate | separated using the edge component registration information in an easy-to-understand manner.

FIG. 7 is a diagram showing an example of division of a contour extraction processing target image adopted in the second embodiment of the present invention.

FIG. 8 is a diagram showing an example of division of a contour extraction processing target image adopted in the third embodiment of the present invention.

FIG. 9 is a diagram showing an example of area-specific operators used in the second and third embodiments of the present invention.

FIG. 10 is a diagram showing examples of various operators used in the conventionally used contour extraction processing.

FIG. 11 is a diagram showing an example of edge component registration when a conventional method is used.

FIG. 12 is a diagram for explaining a background filling result in the case of an image in which edge components are registered as in FIG. 11.

[Explanation of symbols]

a, b, c ... Pixel density value (data), x ... Density value of target pixel, P1 to P4 ... Pixel position.

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] April 30, 1993

[Procedure Amendment 1]

[Document name to be amended] Statement

[Correction target item name] 0001

[Correction method] Change

[Correction content]

[0001]

The present invention relates is related to method for extracting an edge component of an image, and more particularly to a useful image contour extraction method in addition to the image recognition of the recognition eighty human image.

[Procedure Amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0007

[Correction method] Change

[Correction content]

[0007]

The present invention is to provide a digitized image
In the image, the position of the boundary line between the pixels with the density difference
It will be called the actual contour position. Also usually well done
One of the recognized image recognition methods is to register contour components.
By considering the inside of these including the pixel position as an object
is there. In this way, only the background part will be painted in a different color.
When performing processing such as crushing, you can start from any position on the background.
If you fill up just before the pixel where the edge component is registered,
It is convenient because it means that the scenery is filled. The conventional edge extraction method the calculation method, versus the contour of the actual
Detecting the edge component in the pixel-enriched side or light side Neu displacement or one or both sides of the pixel positions, and.

[Procedure 3]

[Document name to be amended] Statement

[Correction target item name] 0008

[Correction method] Change

[Correction content]

[0008] Thus, in the almost constant background concentration, back
There is an object that has both dark and light parts of the scene
Considering such an image , conventional edge extraction methods register edge components at pixel positions on both sides of the actual contour of the object.
Therefore, the edge component is registered from any position on the background.
Even if you fill the area just before the pixel
It doesn't mean to fill it.

[Procedure amendment 4]

[Document name to be amended] Statement

[Correction target item name] 0009

[Correction method] Change

[Correction content]

[0009] us consider an image exhibiting concentration distribution as shown in FIG. 11 for example. Using a contour component extraction operator
Doing calculations by pixel to the edge component of the high density side is registered, the pixel position location marked in white circle in FIG. 11
Edge component Ru is registered in. As a result, in the part where the pixel density of the object is lighter than the background density ,
Then , the edge component is registered not in the pixel on the object side but in the pixel in the background portion.

[Procedure Amendment 5]

[Document name to be amended] Statement

[Correction target item name] 0010

[Correction method] Change

[Correction content]

When the background portion is filled with an arbitrary color based on this edge component, it is shown in FIG.
Uni portion coated Ri crushed part of the contour, the actual contour of the position
Since they not reach the location, between the filled background portion and the object, Ru danger resulting gaps.

[Procedure correction 6]

[Document name to be amended] Statement

[Correction target item name] 0011

[Correction method] Change

[Correction content]

[0011] That is, without straw also written want to separate the background portion and the object at the original position of the actual edge, edge components
Is registered at the pixel positions on both sides of the actual contour.
In addition, it was not possible to divide the actual contour position accurately .

[Procedure Amendment 7]

[Document name to be amended] Statement

[Name of item to be corrected] 0037

[Correction method] Change

[Correction content]

The same applies to the right half, but the operator for the right side shown in FIG. 2B is used as the operator. In this case, the position of the pixel of interest is 2 ×
Of the pixels in the block having the two-pixel matrix structure, the description is the same as the above description for the left-side operator except that it is at the P3 position. However, since the pixel of interest is at the P3 position, the vertically extending edge is the pixel to the left of that edge, and the edge of the diagonal line rising to the left is its left.
An edge is registered in the lower pixel (see FIG. 4).

[Procedure Amendment 8]

[Document name to be amended] Statement

[Correction target item name] 0048

[Correction method] Delete

Claims (1)

[Claims] 1. A dividing step of dividing an image to be processed into a plurality of rectangular areas as an image contour extracting method for extracting an image contour by data processing of a digitized two-dimensional image by an edge component extracting operator. The edge component extraction operator for extracting the edge component from the difference value between the density value of the pixel of interest and the density value of each pixel around the pixel of interest is set in advance for each of the divided areas divided by this dividing step. And an edge component extracting step of extracting an edge by registering the edge in the pixel of interest.