JPH0368085A - Method for detecting picture segmenting position and picture segmenting method using this method - Google Patents

Abstract

PURPOSE:To extract only a specific object by automatically determining edge parts of the gradually segmented specific object and finally obtaining all of the periphery of the outside line of the object. CONSTITUTION:A position Pi (i=1 to 4) where plural edge forms in the object are learnt are designated, and a smooth curve Co passing the designated position Pi for inference of the direction of an edge to be learnt it obtained, and a variable density waveform Wi on a straight line Vi (i=1 to 4) passing the designated position Pi in the normal direction of the designated position Pi of the curve Co is obtained. A variable density waveform W'i in the normal direction in another position Pi in the vicinity of the designated position Pi of the curve Co is obtained, and two variable density waveforms Wi and W'i are collated to obtain an extent Zi of positional deviation, and correction is performed, and the correction position is added to the designated position Pi to successively deform the curve to the outside line of the object. Finally, all of the periphery of the outside line of the object is obtained. Thus, only the specific object surrounded with the outside line is extracted.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention provides a method for detecting an image cutout point when only a specific object image is cut out from an image having gradation or a binary image, and a method using the detection method. The present invention relates to a cropping method for cropping an image, and particularly to a cropping method for separating and extracting only a specific region from a photographic image.

[Conventional technology]

The spread and progress of personal computers and word processors has been remarkable, and as the cost of the recording media used to record information on these devices has decreased, a large number of documents have been created and edited. Also, a method is often used in which images such as photographs having gradation are inserted into a document by combining each of the above devices with an image scanner. Furthermore, the creation of databases of these documents and the creation of electronic files of printed materials using optical disk files and microfilms have come to play an important role in OA (office automation). In particular, editing of image information in a printed document on a personal computer or word processor has become extremely important recently.

One of these editing functions is to automatically extract only a specific object within image information such as a photograph.

When manually cutting out a specific object from a grayscale image such as a photograph, even if some of the edges of the object are unclear, existing knowledge about the general shape and brightness of the object may be used. By using this, it is possible to infer edges, and as a result, it is possible to extract only that object.

In this way, human recognition extracts objects by considering all conditions and using common sense and knowledge to extract objects.

[Problem to be solved by the invention]

Conventionally, knowledge-driven methods that incorporate this human cognitive process, such as 'The ACRONYM model-
based vision 5ysten+''(P
roc, 6th TJCAI.

pp, 105~+13. ! An image recognition method described in 979) has been proposed.

However, even if we try to quantify the human cognitive process,
When automatically extracting any specific object, theoretically the genus is different due to the diversity of conditions and environments, such as the direction of the camera, the characteristics of the camera, the way of lighting, and the shape of the shadow. Even if it is possible, it is extremely difficult in practice and has not yet been put to practical use.

The purpose of the present invention is to solve such conventional problems and provide an image cutting point detection method that can automatically extract and cut out a specific object in gray scale image information such as a photograph, and an image using the method. The purpose is to provide an extraction method.

[Means to solve the problem]

In order to achieve the above object, the image cutting point detection method according to the present invention includes (a) a first step of determining a plurality of specified points on the boundary line between a specific object image and the background; a second step of finding a smooth curve that passes through the specified point, a third step of finding a shading waveform on a straight line passing through the specified point in the normal direction to the curve at each specified point, and each specified point. A fourth step in which a shading waveform on a straight line passing through the neighboring points in the normal direction to the curve at one point on the curve near the point is obtained for each point near the specified point, and for each specified point. The fifth step is to compare the grayscale image obtained at the specified location with the grayscale image obtained at the neighboring locations, calculate the amount of positional deviation in the normal direction of the neighboring locations, and correct the curve itself by correcting the position of each neighboring location. It is characterized by having the following steps. (b) Add each neighboring location whose position has been corrected to the designated location, and repeat the processes from the second step to the fifth step until the total number of designated locations reaches a predetermined number; and second
Another feature of the repeated processing after the first processing is that the gray scale waveforms are compared only for the neighboring locations whose positions have been corrected in the previous processing. Another feature is that (c) the curve is a spline curve that is a smooth curve that passes through a plurality of specified locations.

Furthermore, (d) when the direction of the boundary between a specific object image and the background at each designated location is clear, the curve is drawn at each designated location in the direction of the connection of edges with rapid changes in shading existing in the vicinity of each designated location. Another feature is that it finds a smooth curve that passes through multiple specified points at the time of initial input so that the tangential directions of the curves match. In addition, (1) the image cropping method of the present invention includes (e) cropping only a specific object image from a grayscale or binary image, displaying the cropped object image on a screen, and editing by combining the object image and other images; In the method of Step 2: Connecting adjacent cutout locations with a straight line or a clear curved line and using the curve as the outline of the cutout area; and Step 2: Extracting only an image either inside or outside the outline of the cutout area. It is characterized by having 3.

[For production]

In the present invention, a procedure for learning the diversity of one condition or environment is provided. In other words, (a) specify a location in the object where multiple edge shapes are to be learned, (b) find a smooth curve that passes through the designated location to estimate the direction of the edge to be learned, and (c) calculate the shape of that curve. Find the shading waveform on a straight line passing through the specified point in the normal direction of the specified point, (d) Next, find the shading waveform in the normal direction of the curve at a point near the specified point, and (e) 2 By comparing two shading waveforms, determining the amount of positional deviation, correcting it, and adding the corrected position to the specified location,
The program executes a procedure that sequentially transforms the curve into the outline of the object to be extracted.

In this way, by successively determining the edge portions of a specific object to be cut out, the entire circumference of the object's outline is finally determined. As a result, it becomes possible to extract objects enclosed within this outline.

[Example Go Hereinafter, embodiments of the present invention will be described in detail with reference to seven drawings.

FIG. 2 is an overall configuration diagram of an image editing system to which the present invention is applied.

In FIG. 2, 1 is a photograph serving as a manuscript, 2 is a first image memory or image memory area for storing a two-dimensional image, and 4 is a first image memory or image memory area for storing a two-dimensional image.
is a second image memory or image memory area for extracting and separating specific objects and storing them; 3 is a computer for detecting and cutting out image cutout points; 5 is a CRT display; 6 is a scanner for scanning photographs of original documents; is the coordinate input bag ra (mouse).

First, when you place the photo l on the scanner 6, the scanner 6
scans and reads the grayscale image information of the photograph, converts it into a digital signal 1s, and inputs it to the computer 3. Calculator 3 is
This digital signal Is is temporarily stored in the image memory 2 as a two-dimensional image.

Here, the computer 3 starts image cutout location detection processing. After separating a specific object from the photographic image in the image memory 2, the computer 3 stores the separated object image in the image memory 4. The CRT 5 is an image display monitor used by a person to output the coordinates of a specified location to the computer 3 using the coordinate input device 7 during editing.

FIG. 1(a) is an operation flowchart of a method for detecting an image cutout point showing an embodiment of the present invention, and FIG.
) is a diagram showing a full grayscale image A before being cut out and a specific image a cut out from the image A.

In addition, FIGS. 3(a) to (c) are diagrams showing designated locations and neighboring locations in the image cutout location detection method, and FIG.
a) to (d) are diagrams showing changes in each repetition of a spline curve.

The ultimate goal of image cutout location detection is to cut out a specific image a from the full grayscale image A, as shown in FIG. 1(b).

First, after displaying the full gray scale image A before being cut out as shown in FIG.
Specify and input the coordinates into the calculator 3. Here, the first step is to input i=1.2, 3°4 at four locations.

Next, as shown in FIG. 3(a), each specified location PI(i
= 1, 2, 3° 4) Find a spline curve C0 (second step), where the curve C0 is a curve that is first drawn to pass through a plurality of designated positions, and As will be explained later, the curve changes to a different shape while being sequentially corrected. In addition, in FIG. 3(a),
In order to clearly draw the curve C0, the original image A is expressed faintly. Next, V in FIG. 3(a).

Then, as shown by W in FIG. 3(b), a shading waveform Wi on a straight line Vi that is in the normal direction of the spline curve C0 and passes through each specified location Pl is determined (third step).

That is, in FIG. 3(a), four specified positions P, , P
,,P□ First, at the position of Pl in P4, set a straight line V in the normal direction of the curve C0, and set the gray waveform W on that V.
Measure l. As a result, as shown on the left side of Fig. 3(b),
The level (gradation value) on the right side is higher than on the left side.In other words, the boundary between the low level position on the left side and the high level position on the right side is the position where the boundary of the target object exists, and of course the position of Pl is the boundary of the object. , that is, at the cutting position.

Next, a point pJ on the curve C0 near the specified point Pl
1, and the gray waveform W'l on the straight line Vi' passing through the neighboring point Pi' in the normal direction of the curve C6 at that point.
is determined for each designated location Pi' (fourth step).
That is, in FIG. 3, a straight line V,/ is drawn in the normal direction of the curve C0 at a position P1' near the position P1, and a gradation waveform W1' on that V1' is measured. As a result, as shown on the right side of FIG. 3(b), the level (gradation value) on the right side is higher than on the left side, with the point to the left of coordinate P' on 1' being the boundary. In other words, the P' position is at a position shifted from the boundary of the target object.

In this way, in the third and fourth steps, P,
, P, , P, , P, gradation waveform at each position and p % , p, l , p, l in their vicinity
, P, and the shading waveform at each position are determined.

Here, the image density waveforms Wl on the left and right sides of FIG. 3(b)
Comparing W1' and W1', it can be seen that they exhibit similar waveforms because their positions are almost close to each other. However, since there is a slight deviation between the curve C0 and the actual outline formed by the edge portion, by shifting one of them by the amount of deviation Zl, a correct boundary line, that is, a cutting line can be obtained. Therefore, in order to correct the position of the neighboring point Pi', waveform comparison is performed in the next step, and the value of the deviation amount Zi is determined for each neighborhood. That is, the gray scale waveform Wi.

The neighboring locations are corrected using the deviation fiZi obtained by matching W5', and this corrected location is added to the designated location (fifth step). In waveform matching, the deviation amount Zi can be easily determined by using a well-known correlation method or the like. A point Pi' that is corrected by adding the obtained shift amount Zl to the neighboring point Pi' is expressed as a point Pi'.

Next, adding these correction points to the specified point Pi is set as a new specified point, and the processes from the second step to the fifth step are repeated (the sixth step, the second step to
5th step).

As the number of specified points on the edge increases, the curve C approaches the outer edge of the image you want to cut out. Therefore, set the number of specified points in advance, and when the number exceeds that number, , the image cropping point detection process is completed (sixth step).

FIG. 3(C)) is a diagram showing the newly determined curve C0.

In FIG. 3(c), p , l , p, 11 ,
p, IP,' are all at the position f! in FIG. 3(a).
! This is the position corrected by shifting P,',P,'p,l,p,I by the deviation amount of Zi. And, Figure 3 (C
) curve C1 corresponds to each of their corrected positions P1' to P
4' is the second correction curve drawn in addition to the previously specified positions P1 to P. This second correction curve has a shape closer to the boundary line of the target object than the first curve C. Then, the correction positions P' to P5 on this second curve
By executing the process shown in FIG. 1(a) with respect to ', the deviation at another position is measured again, and then a third curve is created by adding the corrected position.

As is clear from the curve aC in FIG. 3(C), as the number of specified points increases, it is possible to obtain a curve closer to the outline of the specific image to be extracted.

Therefore, the next nearby location can be easily detected. In addition, in the second and subsequent iterations, the image gradation waveforms W and Wi'
It is assumed that only the latest one (that is, the correction points p, # obtained in the previous iteration) among the specified points Pi is used as a reference point for cutting out.

FIGS. 4(a) to 4(d) are diagrams showing changes in each repetition of the spline curve in the present invention.

By determining the amount of positional deviation in the vicinity of the designated location and adding the corrected position based on this to the designated location, the spline curve C1 can be created from Figure 4 (a) to (b) and (b) to (C).
, changes from (C) to (d). Through repeated processing, the curve C0 gradually approaches the outline of the image you want to cut out.At this time, the number of specified points on the outline edge of the image you want to cut out gradually increases, so when it reaches a certain number, , ends the image cutout location detection.

Next, each coordinate Qi on the outline of the image to be cut out is determined by interpolating the image cutout points, which are a set of specified points, with a straight line or a smooth curve (Q, ,
Q,, Q,, Q,, -...--Find Qn)
. Then, based on these coordinates, the details of this clipping, which cuts out an image inside or outside the outline line, can be found, for example, in ``Polygon Scan Conversion (2)j (P
IX, EL, No. 16, pp.

This can be easily realized by using the method described in 136).

In addition, when the central position, maximum height, width, area, etc., which are the geometrical characteristics of the cut-out image, are required, an average calculation, maximum/minimum detection calculation, or It can be determined by addition operations, etc.

FIG. 5 is an explanatory diagram of an image cutout point detection method showing another embodiment of the present invention.

In the above example, a smooth curve passing through a designated location was simply defined as a spline curve, but if the direction of connection of edge portions at a designated location is known, the tangential direction of the curve at each designated location can be set in this direction. Let it be a conditional spline curve that matches . In other words, as shown in FIG. 5(a), if the tangent directions R, , R, , R can be clearly drawn with respect to the image cutout outline, the sixth-order curve C0 that takes these into account is first drawn. As described above, the process is repeated for multiple designated locations.

In addition, in the above embodiment, when setting the first neighborhood point, one neighborhood point is set for the specified point, but one neighborhood point is set on each side of the specified point, and processing is performed. Efficiency can be improved. That is, FIG. 5(b)
As shown in FIG.
′ and the right-hand neighborhood P1. ′, find the shading waveforms W, ′, W, , ′ in the normal direction, respectively.
Due to this.

Letting the corrected positions be p, #, p, . . . , the curve obtained by adding these positions to the designated position is C8'.

In this way, if the correction is performed simultaneously at positions near both sides of one specified position, the edge of the cut-out external portion can be approached more quickly.

[Effects of the Invention] As explained above, according to the present invention, since the edge portion of a specific object to be gradually cut out is automatically determined, it is possible to finally obtain the entire circumference of the object. , it is possible to extract only a specific object using this outline.

[Brief explanation of drawings]

FIG. 1 is a processing flowchart of a method for detecting an image cutout location according to an embodiment of the present invention and a diagram showing an example of an object to be cut out, FIG. 2 is an overall configuration diagram of an image editing system to which the present invention is applied, and FIG. The figure shows a specified point and a nearby point in the method of detecting an image cutout point in FIG. 1, FIG. 4 shows a change in each repetition of a spline curve, and FIG. 5 shows another embodiment of the present invention. FIG. 2 is an explanatory diagram of a method for detecting an image cutout location. l: photo, 3: computer, 2, 4: image memory, 5: CR
T display, 6: scanner, 7: coordinate input device, A
: Overall grayscale image, a: Specific partial image, Pill! Fixed location, PN2. Nearby location, C, Nispline curve. < 9 Figure (Part 1) Figure 3 (Part 1) (a) ■1 upper coordinate ■1' upper coordinate Figure Figure (2) (C) Figure (a) Spot

Claims (1)

[Claims] 1. In a method for detecting a cutout point by image processing for cutting out only a specific object image from a grayscale or binary image, a plurality of specified points are set on the boundary line between the specific object image and the background. a first step of determining, a second step of determining a smooth curve passing through the plurality of specified points, and a shading waveform on a straight line passing through the specified points in the normal direction to the curve at each of the specified points. A third step is to obtain for each designated location, and a shading waveform on a straight line passing through the nearby location in the normal direction to the curve at one location on the curve near each of the designated locations, A fourth step of determining the positional deviation in the normal direction of the nearby location by comparing the gray scale image determined at the designated location with the gray scale image determined at the nearby location for each designated location. a fifth step of correcting the curve itself by calculating the amount and correcting the position of each of the neighboring points. 2. Add each neighboring location whose position has been corrected above to the designated location, and repeat the processes from the second step to the fifth step until the total number of designated locations reaches a predetermined number, and 2. The image cutout point detection method according to claim 1, wherein in the second and subsequent repeating processes, gray scale waveforms are compared only for neighboring points whose positions have been corrected in the previous process. 3. The image cutting point detection method according to claim 1 or 2, wherein the curve is a spline curve that is a smooth curve passing through a plurality of specified points. 4. When the direction of the boundary between the specific object image and the background at each designated location is clear, the above curve is drawn in the direction of the connection of edges with rapid changes in shading existing in the vicinity of each of the designated locations. 3. The image cutting point detection method according to claim 1, wherein a smooth curve passing through a plurality of specified points at the time of initial input is obtained so that the tangential directions of the curves coincide with each other. 5. A method for cutting out only a specific object image from a gray scale or binary image by image processing, displaying the cut out object image on a screen, and editing the object image in combination with other images, as claimed in claim 1. After performing the processing from the first step to the fifth step shown in the above section 1 using one of the image cropping point detection methods from the sections 4 to 4, multiple designated points on the curve and a step 1 in which neighboring points are cut out points in a gray scale or binary image; a step 2 in which the plurality of adjacent cut out points are connected with a straight line or a smooth curve and the curve is set as an outline of the cut out region; An image cutting method comprising the step of extracting only an image either inside or outside the outline of the cutting area.