JPH0769967B2 - Shape description method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a shape description method, and more particularly to a shape description method for hierarchically describing shapes.

[Prior Art] It has been pointed out that the structure description method is important for shape matching and classification.

Conventionally, for example, as a hierarchical representation method of a figure showing a contour line, there is a scale space representation focusing on the curvature of the contour line. This method smooths the curvature function along the contour line with a plurality of Gaussian functions having different standard deviations, and traces a feature point (for example, a point on the contour line where the curvature is 0) to determine the feature of the figure. Is related to the change in resolution to describe the structure of the figure.

[Problems to be Solved by the Invention] However, in this method, the change of the figure feature points along the resolution axis is unstable, and if this is structured by a tree structure or the like, a subtle change of the feature may occur. However, there is a problem that the contour of the shape cannot be accurately structured.

In addition, a concrete method that clearly symbolizes and expresses the feature amount of the shape and the relation description between each level by different resolution levels has not yet been established, and therefore recognition of shape matching and classification is not performed. Not applied for.

The present invention has been made in order to solve the above problems, and an object thereof is to obtain a shape description method capable of accurately structuring and describing a shape hierarchically.

[Means for Solving the Problem] The present invention is a shape description method, which shows a contour of a shape.
3D image signal, the contour of the shape is smoothed for each smoothing level indicating the degree of smoothing, a plurality of smoothed contour signals are output, and the smoothed contour signal is received and smoothed. For the contours, the skeletons of the contours are extracted in the form of connecting line segments, and the skeleton is approximated to a polygonal line to obtain the polygonal line signal and the intersection signal of the polygonal line, and the shape including the total length of the polygonal line and the length and direction of each polygonal line. To represent the approximated skeleton as a tree, perform labeling on the end points and paths for each tree, and label the extracted features. The tree is added to the created tree.

[Operation] In the shape description method according to the present invention, a two-dimensional image signal is smoothed for each smoothing level indicating a degree of smoothing, a contour signal is output, and a skeleton of the contour is extracted in the form of a connected component. Approximate the skeleton to a polygonal line, extract the feature amount for expressing the shape based on the intersection signal of the polygonal line signal and the polygonal line signal, and express the approximate skeleton as a tree based on the polygonal line signal and the intersection signal, and make an end point for each tree. And the path is labeled, and the extracted feature amount is added to the labeled tree, so that the shape is normally described.

[Embodiment of the Invention] FIG. 4 is a block diagram showing a configuration of hardware around a shape description device of the present invention.

Referring to FIG. 4, an original image whose shape is to be described is input by a television camera 61 or an image scanner 63. The image signal obtained by the television camera 61 is A / D converted by the A / D converter 62, and the image data is stored in the image data storage device 64. On the other hand, when the image scanner 63 is used, the image data obtained thereby is stored in the image data storage device 64.

The shape description device 70 obtains a two-dimensional image pattern from the image data storage device 64. The shape description device 70 makes a hierarchical structure description for this image pattern and outputs it as shape data. The obtained shape data is stored in the shape data storage device 65 as needed, and serves as a storage model when shape matching or classification is performed later.

FIG. 1 is a flow chart showing an embodiment of the shape description device according to the present invention.

Referring to FIG. 1, in step 1, a two-dimensional image pattern showing a certain shape is input. In step 2, the contour line is traced for this image pattern, and the contour line is extracted. In step 3,
First, the coordinate series of the extracted contour line is smoothed by a Gaussian function having a sufficiently large standard deviation σ ₁ . This is called level 1 smoothing, and the largest smoothing is performed.

Next, in step 4, the skeleton is extracted in the form of connecting line segments from the level 1 smoothed contour figure. In step 5, the skeleton is symbolized to describe the feature amount of the figure. As a method of symbolizing the skeleton, the polygonal line approximation of the skeleton is performed. The line approximation yields nodes and branches.

In step 6, a feature amount for expressing the shape of the node and the branch is added. The feature amount at the node and the branch is, for example, as shown in the following Table 1 at each level.

In parallel with this, in step 7, in order to describe the relationship between each level, the approximate skeleton obtained by the polygonal line approximation is expressed as a tree, and nodes (end points) and branches (paths) about this tree are expressed. Label it. Labeling here involves associating nodes and branches with reference to the previous level tree. (However, labeling is performed independently for level 1.) In step 8, the feature amount extracted in step 6 is added to the labeled tree obtained in step 7. This provided a structural description for one level of smoothing.

The structure description is similarly made for Level 2 and below, but the standard deviation σ _i (i = 2,3, ...) for smoothing is set gradually smaller, which allows the structure description to be made hierarchically. it can. That is, a tree including feature quantities is obtained for many levels from the coarse level to the fine level.

As described above, the shape can be accurately structured and described hierarchically.

Next, the algorithm will be described using a camel as a specific example.

FIG. 2 (a) shows the contour of the camel extracted from the input image pattern.

FIG. 2B shows that level 1 (coarse level) is smoothed in step 3 and skeleton polygonal line approximation is performed in step 5. In this figure,
There is one node N1 and three branches 1,2,3.

FIG. 2 (c) shows that an approximate skeleton is created by further performing the smoothing of the dense level. The approximate skeleton is determined for each line segment of the skeleton by examining the k-curvature where k is a value obtained from the line segment length.

By comparing FIGS. 2 (b) and 2 (c), it is found that the coarse-level and fine-level approximate skeletons show global features and detailed features indicated by the shapes of the input image patterns, respectively. Recognize. In this example,
Although only two representative levels are shown, in practice, the intermediate level between them is also processed.

FIGS. 3 (a) to 3 (e) are trees showing the relational description of each level formed in step 7 of FIG. In these figures, symbols N1, N2, ... Show node names, and numbers show branch names.

FIG. 3 (a) shows the coarsest level tree. Based on this tree, as shown in FIGS. 3 (b) to 3 (e), a tree of higher level is sequentially created, and the respective nodes and branches are labeled to describe the relation.

In the labeling method in this example, for example, as shown in FIG. 3 (b), a node in which a new node appears refers to the previous level, and the node N2 appears in the branch 3. Split into one branch 31, 32 and 33. Similarly, at the next level shown in FIG. 3C, since the node N3 appears in the branch 31 of the previous level, the branch 31 is divided into branches 311, 312 and 313.
Is divided into In the same manner, by repeating this operation until the dense level shown in FIG. 3 (e), the relational description is obtained. Here, in the final dense level tree shown in FIG. 3E, the relationships of all levels are described.

Note that FIG. 3 (a) is a tree in which relational description is made corresponding to FIG. 2 (b), and FIG. 3 (e) is FIG. 2 (c).
Is a tree in which the relational description is made in correspondence with.

Next, using the hierarchical structure description obtained as described above, a method of utilizing shape matching or classification will be briefly described.

FIG. 5 is a schematic flow chart showing the procedure of shape matching and classification.

With reference to FIG. 5, first, in step 51, the shape description device of the present invention describes the hierarchical structure of the input image pattern to obtain a multi-level tree representation. Next, in step 52, matching or classification is performed by comparing this tree representation with a prestored storage model 54. This storage model 54 is a database of shapes, which are represented in advance by a plurality of levels as a tree for many typical or typical shapes by the shape description device.

Here, matching and classification can be performed at a desired level, and efficient use can be achieved by hierarchically using the feature amount and the relational description. Further, by giving a certain degree of freedom to the feature amounts shown in Tables 1 and 2, for example, flexible matching can be performed even for shapes that may change, such as living things. Is possible.

[Effects of the Invention] As described above, according to the present invention, the contour of the shape 2
3D image signal, smoothes each smoothing level indicating the degree of smoothing, outputs a contour signal, extracts the skeleton of the contour in the form of connecting line segments, approximates the skeleton with a polygonal line, and then By extracting the feature quantity for expressing the tree, expressing the approximate skeleton as a tree, labeling the end points and paths of each tree, and assigning the feature quantity to the labeled tree Can be hierarchically structured and described.

[Brief description of drawings]

FIG. 1 is a flow chart showing an embodiment of the shape description device according to the present invention. 2 (a) to (c) show the first
It is a figure which shows a contour line and the approximate skeleton about the example of the camel applied to a figure. Figure 3 (a) to (e)
FIG. 3 shows a labeled tree for the camel example applied to FIG. 1. FIG. 4 is a block diagram showing the configuration of hardware around the shape description device of the present invention. FIG. 5 is a schematic flow chart showing a procedure for performing shape matching and classification using the shape description device of the present invention. In FIG. 3, N1 to N5 are node names, 1 to 1000
3 indicates the branch name.

Claims (1)

[Claims] 1. A two-dimensional image signal indicating a contour of a shape is received, the contour of the shape is smoothed for each smoothing level indicating a degree of smoothing, and a plurality of smoothed contour signals are output. Receiving the smoothed contour signal, for the smoothed contour, the skeleton of the contour is extracted in the form of a connecting line segment, and the skeleton is approximated by a polygonal line to obtain a polygonal line signal and an intersection signal of the polygonal line, and a polygonal line. Extract the feature amount for expressing the shape including the total length and the length and direction of each polygonal line, receive the polygonal line signal and the intersection signal, represent the approximate skeleton as a tree, and for the end points and paths for each tree A shape description method comprising performing labeling and adding the extracted feature quantity to the labeled tree.