JP2004021888A - Object extraction method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To extract an object from an image automatically and correctly. <P>SOLUTION: The method divides an image into blocks (S1), scans the image blocks beginning with the upper left corner of the image (S2), classifies the blocks into two types of edge blocks, that is, blocks with two-directional edges and blocks with one-directional edges (S4, S5, S6), and determines the block, which occurs first during block scanning, as a starting block (S7). The method evaluates one of the change points of the starting block and the remaining change points to determine the direction in which the blocks forming the outline of the object exist, and compares the difference of change points among neighboring blocks in the determined scan direction with the threshold value to determine the change point of the next block as the feature point of the object. By repeating the compairing process, the method extracts the object from the image. Whether to extract the object is also determined based on the motion vector of the feature points of the extracted object. <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a method for extracting an object from an image.
[0002]
[Prior art]
As a technology of image encoding, MPEG-4, which can handle video in content units, has been standardized so as to increase the degree of compression and support multimedia, as compared with MPEG-2, and has been defined therein. A method of encoding each object is an indispensable technique for obtaining a high compression rate. Therefore, it is important how accurately the object can be extracted from the image.
[0003]
[Problems to be solved by the invention]
At present, a method of extracting an object from an image is not specifically defined in MPEG-4, and is currently performed manually. That is, since object extraction is not automated, real-time processing becomes impossible, resulting in poor versatility.
[0004]
An object of the present invention is to provide an object extraction method that can automatically and accurately extract an object in an image.
[0005]
[Means for Solving the Problems]
In order to solve the above-described problem, the present invention detects an edge in an image, extracts an object by drawing a locus of the object, and determines whether or not the object can be extracted based on a motion vector. This is characterized by the following method.
[0006]
(1) A method for automatically extracting an object in an image,
Dividing the image into blocks of uniform size;
The divided blocks are scanned, and each block is classified into two types of edge blocks: a first block having edges in two directions and a second block having edges in only one direction. The process of
Determining the coordinates of three transition points for the first block and determining the coordinates of two transition points for the second block;
Determining the direction in which the block forming the contour of the object exists from the evaluation of one change point and the remaining change points of the start block, the block appearing earliest in the scan of the block;
Determining the change point of the next block as a feature point of the object by comparing the difference between the change point between adjacent blocks in the determined scan direction and the threshold value;
Is repeated to extract the object in the image.
[0007]
(2) The evaluation is
The coordinates of the upper left corner of the block and the coordinates of the change point in the block are acquired. If the value of the X axis of the coordinates of the change point matches the size of the block, the direction of the adjacent block is determined to be the X direction. For example, when the value of the Y axis matches the size of the block, the adjacent block direction is determined to be the Y direction.
[0008]
(3) A motion vector of each feature point of the extracted object is obtained, and whether or not the motion vectors of most of the feature points are in the same direction is used as a material for determining whether or not the object can be extracted. .
[0009]
(4) Among the motion vectors of the feature points, a motion vector of a feature point having a different direction is determined by changing a time span.
[0010]
BEST MODE FOR CARRYING OUT THE INVENTION
1 to 3 are object extraction procedure diagrams showing the embodiment of the present invention. The details will be described below with reference to FIGS.
[0011]
In FIG. 1, first, as shown in FIG. 4, an image BG obtained by dividing the image G into uniform blocks of size 16 × 16 is generated (S1). Each block constituting the image BG is scanned from the upper left, and classified into two representative types of edge blocks (S2).
[0012]
In this classification, as shown in FIG. 5, edge block 1 is a block in which one block contains edges in two directions, and edge block 2 is a block in which one block contains edges in only one direction. . Each edge block (three points for edge block 1 and two points for edge block 2) has a change point in luminance value.
[0013]
As shown in FIG. 6, for the edge block 1, the change points α (X1, Y1), β (X2, Y2), and γ (X3, Y3) of the luminance value are obtained, and the coordinate α is determined as the starting point coordinate A. I do.
[0014]
Next, the scan proceeds from the starting point coordinates A to the right block. For this block, the edge block 1 or the edge block 2 is identified (S3), and the coordinates of the luminance value change point are similarly calculated for each edge block (S4, S5). For example, coordinates δ (X3, Y3) and ε (X4, Y4) are set for the edge block 2 in FIG. Further, similarly, the scanning is sequentially advanced to calculate the coordinates of the changing point for each edge block (S6).
[0015]
As a result, the coordinates of the change points are given to the edge blocks until the scan is completed at the lower right of the image.
[0016]
Lastly, the block appearing earliest after the scanning is set as the starting block, and the starting coordinate A is determined (S7).
[0017]
Next, FIG. 2 shows a process of extracting feature points in an image. First, the change points β and γ with respect to the starting point coordinates A in the starting point block are evaluated to determine in which direction of the X or Y the edge block exists (S11). This evaluation method will be described later with reference to FIG.
[0018]
If the direction in which the edge block exists is the X direction, the trajectory of the edge block is obtained in the X direction from the starting block. In the starting block, the difference between β (X2, Y2) having the largest X coordinate and the coordinate δ (X4, Y4) having the smallest X coordinate of the next block is calculated, and it is confirmed that the difference is within a certain threshold value Th1. (S12). In the above case, it is expected that X2 + 1 = X4 and Y2 = Y4.
[0019]
Similarly, if the direction in which the edge block exists is the Y direction, the trajectory of the edge block is obtained in the Y direction from the starting block, and it is confirmed that the difference is within the threshold value Th1 (S13). .
[0020]
Next, the difference between the δ point and the ε point of the next block in the X or Y direction with respect to the starting block is determined, and the magnitude of this difference with a certain threshold Th2 is checked (S14). If it is determined in this check that the difference is equal to or smaller than the threshold value Th2, the point ε is set as coordinates for the next comparison (S16). If it is larger than the threshold value Th2, the δ point is held as the characteristic point B of the trajectory, and the ε point is set as the coordinates for the next comparison (S15).
[0021]
Similarly, in the next block, the difference between the coordinates of the δ point and the ε point in the X direction is obtained, and if it is smaller than the block size (here, 16), the scan direction is changed to the Y direction. If it is larger, it is scanned in the X direction (rightward). By repeating this, the characteristic points of each edge block can be obtained. An object can be extracted by connecting these feature points with a straight line. FIG. 7 shows an example of an object extracted as a trajectory of feature points A to P extracted by such a scan.
[0022]
An evaluation method for a change point of the image block will be described. In FIG. 3, the number of determinations i is initially set to 1 (S21), and the number of determinations i is smaller than 4, that is, the change point in the maximum of three directions is limited (S22). First, the upper left coordinates (XL, YL) of the block To obtain the coordinates (Xi, Yi) of the change point (S23).
[0023]
It is checked whether or not the X-axis value Xi = XL + 16 of the coordinates (Xi, Yi) (S24), and if they match, the edge block direction is set to the X direction (S25). If they do not match, it is checked whether the Y-axis value Yi = YL + 16 (S26). If they do not match, the process returns to the next determination (S27). If they match, the edge block direction is set to the Y direction (S28).
[0024]
Next, a motion vector is calculated for 16 feature points A to P obtained from the object extraction result in FIG. If this motion vector points in a completely random direction, the object extraction will fail. If one or two points are in different directions, the object can be extracted more accurately by separating blocks in the vicinity thereof more finely.
[0025]
The motion vector is as shown in FIG. 8, and some of the motion vectors are obtained. If a certain part of the motion vectors moves slightly differently from other parts, it is necessary to examine the motion vector of the part in detail. This portion (in the example of FIG. 8, the portion of the car tire) moves in a complicated manner as shown in FIG. 9 when viewed in a short time span, but moves in the same direction when viewed in a long time span. It can be seen that the motion vector is directed.
[0026]
In this way, by examining the motion vector, it can be confirmed that the portion where the object has been extracted based on the edge information has been correctly extracted.
[0027]
【The invention's effect】
As described above, according to the present invention, as a method of extracting an object of an image, an object is extracted using edge information, and it is determined whether or not the object can be extracted based on a motion vector of a feature point of the extracted object. As a result, there is an effect that the object can be automatically and accurately extracted, which has conventionally depended on manual labor.
[Brief description of the drawings]
FIG. 1 is an object extraction procedure (part 1) showing an embodiment of the present invention.
FIG. 2 is an object extraction procedure (part 2) showing the embodiment of the present invention.
FIG. 3 is an object extraction procedure (part 3) showing the embodiment of the present invention.
FIG. 4 is an example of dividing an image into 16 × 16 uniform blocks.
FIG. 5 is an explanatory diagram of an edge block.
FIG. 6 is an explanatory diagram of feature points of an edge block.
FIG. 7 is an example of object extraction.
FIG. 8 is an example of a direction of a motion vector of an object.
FIG. 9 shows an example of a motion vector having a long time and a short time span.

Claims (4)

A method for automatically extracting objects in an image,
Dividing the image into blocks of uniform size;
The divided blocks are scanned, and each block is classified into two types of edge blocks: a first block having edges in two directions and a second block having edges in only one direction. The process of
Determining the coordinates of three transition points for the first block and determining the coordinates of two transition points for the second block;
Determining the direction in which the block forming the contour of the object exists from the evaluation of one change point and the remaining change points of the start block, the block appearing earliest in the scan of the block;
Determining the change point of the next block as a feature point of the object by comparing the difference between the change point between adjacent blocks in the determined scan direction and the threshold value;
A method for extracting an object in an image by repeating the above-mentioned steps. The evaluation is
The coordinates of the upper left corner of the block and the coordinates of the change point in the block are acquired. If the value of the X axis of the coordinates of the change point matches the size of the block, the direction of the adjacent block is determined to be the X direction. 2. The method according to claim 1, wherein the direction of the adjacent block is determined to be the Y direction when the value of the Y axis matches the size of the block. 2. The method according to claim 1, wherein a motion vector of each feature point of the extracted object is obtained, and whether or not the motion vectors of most of the feature points are in the same direction is used as a material for determining whether or not the object can be extracted. Or the method for extracting an object according to 2. 4. The object extraction method according to claim 3, wherein, among the motion vectors of the feature points, a motion vector of a feature point having a different direction is determined by changing a time span.

Images