JP2000082148A - System for extracting shape feature of object image and system for recognizing image object and device therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To extract the shape features of an object image by hierarchically describing the shape features from the outline of the image of an object inputted from an image inputting equipment such as a scanner or a digital camera, for example, the leaf of a plant. SOLUTION: The basic structure of a leaf 1 is constituted of triangular fragments 11-15 jumping to the four directions and a vertex A of each fragment. The leaf can be identified by the structure constituted of the fragments and the shapes of the fragments. A hierarchical procedure is used for a recognition method, and the perspective structure of the outline of the leaf and the rough shapes of the fragments constituting the leaf are captured, and classification is performed, based on obtained featured values. Moreover, the further detailed shapes of the fragements are captured for the kind of a leaf which can not classified in this stage, and classification is performed, based on the obtained featured values.

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 a shape feature of an image object, an object recognition method, a plant classification method, and more specifically, to a method for recognizing a plant seed based on a leaf shape.

[0002]

2. Description of the Related Art Plants are basically classified by the shape and structure of their flowers and reproductive organs (TE Weier, CR Stockin
g, MG Barbour, and TL Rost.Botany: an introdu
ctionto plant biology 6th ed. John Wiley & Sons, 1
982.). However, it is difficult to input the shape and structure of a flower or a reproductive organ due to its complicated structure. On the other hand, since the leaves of the plant have a flat shape, the input is easy. It also has a variety of shapes required for identification.

[0003] The outline of a leaf is considered to be one characteristic for distinguishing species. However, the contour of the leaf varies even in the same species. In addition, the leaves can be classified into similar structural groups, making it difficult to qualitatively identify them. Therefore, the recognition needs to be based on the statistical property of the feature amount indicating the shape.

[0004] In order to obtain a feature value, it is necessary to represent a leaf based on a certain expression method. There are various methods for expressing the contour of a flat object. For example, a hierarchical linear approximation of a curve (AK
Mackworth.On reading sketch maps.In Proc.5th I
nt. Joint Conf. ArtificialIntell., pages 598-606,
Cambridge, MA, 1977. and LS Davis. Understandin
g shape: Angles and sides. IEEE Trans. Comput., C-
26 (3), 1977.) recursively improves the polygonal approximation of the curve, but in this method there can be several different descriptions for a closed curve. Several curve segmentation methods have been proposed based on the critical point of the contour curvature (DD Hoffman and
WA Richards. Representing smooth plane curves f
or recognition: Implications for figure-ground rev
ersal.In Proc. Nat. Conf, Artificial Intell., page
s 5-8, Pittsburgh, PA, 1982. and M. Leyton. A pro
cess-grammar for shape.Artificial Intelligence, 3
4: 213-247, 1988.). The critical point of curvature is invariant to the similarity transformation and is uniquely determined. However, since the value of the curvature depends on the local shape of the curve, the recognition method based on the value of the curvature is not suitable for the shape of the leaf due to local shape variation.

[0005]

SUMMARY OF THE INVENTION In the present invention, a method for hierarchically expressing the shape of a leaf contour by a polygon having a vertex at the critical point of the curvature of the contour is presented, and a hierarchical recognition method based on the method is presented. Present. First, a statistical value of a certain shape feature is calculated from the hierarchical polygonal approximation of the leaf learning data. The similarity of the species is defined based on this statistic. Recognition is performed hierarchically using this similarity. Recognition by the presented method has been successful.

[0006]

According to a first aspect of the present invention, a polygon having a vertex at a critical point of curvature is constructed from a contour of an object image input from an image input device such as a scanner or a digital camera. The feature is that the contour shape features are described hierarchically by polygon features.

According to a second aspect of the present invention, a polygon having a vertex at a critical point of curvature is constructed from a contour of an object image input from an image input device such as a scanner or a digital camera, and features of the polygon are used. Thus, the contour shapes are hierarchically classified.

According to a third aspect of the present invention, there is provided means for constructing a polygon having a vertex at a critical point of curvature from the contour of an object image input from an image input device such as a scanner or a digital camera, and the feature of the polygon. The method further comprises means for classifying objects by using the method, and a part for controlling the detailed classification of the shape by the number of sides of the polygon.

According to a fourth aspect of the present invention, there is provided means for constructing a polygon having a vertex at a critical point of curvature from a leaf contour of a plant image input from an image input device such as a scanner or a digital camera, and the polygon. Is characterized in that the characteristics of the leaves of the plant are described hierarchically.

According to a fifth aspect of the present invention, a polygon having a vertex at a critical point of curvature is constructed from a contour of a leaf of a plant image input from an image input device such as a scanner or a digital camera. Is used to classify plants hierarchically.

According to a sixth aspect of the present invention, there is provided means for constructing a polygon having a vertex at a critical point of curvature from a contour of a leaf image of a plant input from an image input device such as a scanner or a digital camera. And means for classifying plants by using the feature of (1) and a part for controlling the detailed classification of the shape by the number of sides of the polygon.

[0012]

DESCRIPTION OF THE PREFERRED EMBODIMENTS (Leaf Shape of Maple Family) The recognition method according to the present invention is based on a hierarchical expression method of shape features.
First, the structure of the leaf shape is analyzed, and then the shape of its components is analyzed in more detail. The structure of the shape of the leaf and the shape of the detail are approximated by a polygon having the vertex of the critical point of the curvature of the contour of the leaf as a vertex, and recognition is performed based on the obtained feature amount. This recognition method also takes into account variations in leaf shape of the same species. Experiments with several plants of the maple family have shown that the structure of the leaves and the shapes of their constituents can be indicators for recognizing plant species.

In the experiment of the present invention, as shown in FIG.
A species of maple leaf 1 is used. In FIG. 1, they are displayed using lowercase letters of alphabets (a) to (i). These can be roughly classified by the number of vertices A corresponding to the sharp corners of the outline of the leaf 1. The seeds (g) to (i) in FIG. 1 have three vertices A, and the other seeds (a) to (f) have five vertices A. The basic structure of the leaf 1, triangular pieces 1 _1, as shown in FIG. 2
It is thought that ~ ₁₅ protrudes in all directions.
The leaves can be identified by the structure composed of the fragments and the shape of the fragments.

(Leaf Recognition Method) The recognition method proposed in the present invention employs a hierarchical procedure. First, the global structure of the outline of the leaf and the rough shape of the fragments constituting the leaf are captured, and classification is performed based on the obtained feature amounts. Species that cannot be classified at this stage capture the more detailed shape of the fragments and classify them based on the obtained feature values. FIG. 3 is a flowchart showing an example of the hierarchical procedure.

(Leaf Representation Method) In order to obtain a characteristic amount required for recognition, it is necessary to represent a leaf shape based on a certain expression method. The representation method presented in the present invention is based on a polygon approximation having a vertex at a critical point of the curvature of the contour.

(Expression of Global Shape of Leaf) In order to express the general structure of a leaf and the structure of a fragment, a polygonal approximation connecting critical points of the curvature of the contour is used. Therefore, the expression is invariant to the similarity transformation. The polygonal approximation can represent the shape of the bottom part with the leaf handle, the positional relationship between the fragments, and the angle formed.

First, the curvature of the contour is calculated. The curvature of the contour is calculated using the following equation (1) using the contour smoothed using a Gaussian filter (F. Mokhtarian and A. Mac)
kworth. Scale-based description and recognition of
planar curves and two-dimensional shapes.IEEE Tr
ans. on PAMI, PAMI-8 (1): 34-43, 1986.).

[0018]

(Equation 1)

Here, (X (t), Y (t)) are parameter displays of the coordinates of points on the contour smoothed by the Gaussian filter. The point above each coordinate variable indicates the derivative by t.

Next, a critical point is determined from the curvature, that is, a point at which the derivative of the curvature by t becomes zero. The critical point where the absolute value of the critical value is large corresponds to the sharp corner of the contour.
Negative values result in concave corners. The A ₁ to A ₅ points having large positive threshold as shown in FIG. 4 is referred to as the apex of the leaf. From the critical points of this curvature, those having a positive high curvature value exceeding a certain threshold corresponding to the vertex are selected. Next, a minimum point B having a negative curvature between two adjacent vertices is obtained. This point corresponds to the sharp corner of the concave. By connecting these points, a polygon approximation of the upper part of the contour of the leaf 1 can be obtained. This polygon approximation represents the rough shapes of the fragments and their positional relationships. If the leaf has three vertices, for example, A ₂ , A ₃ , A ₄ , the leaf is composed of four line segments C ₃ , C ₄ , C ₅ , C ₆ and has _five vertices A _{1 to} A ₅ Is eight line segments C _{1 to} C ₈
It consists of. A polygon approximation consisting of n line segments is
Relative arc length l _i (i = 1,...,
n−1) and the angle θ _i (i = 1,...,
n-1), (l ₁ ,..., l _n , θ ₁ ,..., θ _n-1 ).

(Expression of the shape of a fragment in the middle of a leaf) A more detailed shape of a fragment can be described by increasing the number of polygonal segments that approximate the contour. Since the shape of the pieces 1 ₃ middle as shown in FIG. 5 are relatively stable compared to others, it is performed to improve the polygonal approximation of this debris. Although one debris eg 1 ₃ was expressed in two segments C _4, C ₅ in the previous stage, four segments C ₄₁ In this step, C _42, C _51, C
Approximated by ₅₂ .

Since the detailed shape of the contour cannot be obtained stably at some critical points, another method is used. Between the vertex of the fragment and the concave corners on both sides of the fragment, a contour division point for a new line segment is obtained.This point is defined as the line connecting the concave corner and the division point, the division point and the vertex. The difference between the sum of the lengths of the connecting line segments and the arc length of the contour between the concave corner and the vertex is determined to be minimized.

(Similarity of Leaf) Recognition is performed using the feature amount obtained by the above expression method. However, since the leaf contour has the same kind of variation, a statistical method is used for recognition. Required. First, the statistics (mean and variance) of each feature l ₁ and θ _j are calculated from the leaf learning data of each species.
The average is represented by equation (2), and the variance is represented by equation (3).

[0024]

(Equation 2)

Based on this statistic, the similarity to each species is defined as the probability that the observed sample belongs to a species based on the assumption that the distribution of each feature is normal and independent of each other. Define. In other words, if the i-th arc length of the polygon approximation of the leaf to be recognized is A (Equation (4)) and the i-th angle is B (Equation (5)), Equations (6) and (7) Are defined respectively, and the similarity is expressed as a product as shown in Expression (8). The range of similarity is from 0 to 1. This similarity is defined similarly for both stages.

[0026]

(Equation 3)

(Recognition method) Before recognizing leaf species, parameters necessary for calculating similarity of each species are calculated from learning data. Recognition is performed as follows. First, leaves are classified into two groups according to the number of vertices. Next, a polygon approximation representing the global shape of the leaf is obtained, and the similarity is calculated based on the approximation. The similarity is determined for each species, and those whose values exceed a certain threshold are determined to be the seeds of the leaf. At this stage, there may be some similarities of species above the threshold. If the leaf cannot be identified in the first stage, the shape of the fragment in the middle of the leaf is approximated by a polygon and the similarity is calculated in the same manner. Final recognition is performed based on this similarity value.

(Recognition Means) FIG. 6 is a block diagram for explaining an example of recognition means suitable for use in the embodiment of the present invention.
As shown, CPU 11, memory 12, display device 1
3, hard disk 14, keyboard 15, CD-RO
A computer system comprising a recording medium 16 such as an M, a CD-ROM drive 17, an image input means 18 such as a scanner and a digital camera is prepared, and a computer-readable recording medium such as a CD-ROM is provided with the contour judgment of the present invention. And a program for determining the vertices of the polygon, the curvature of the contour, and the like. Further, image data input from the image input means 18 such as a scanner is temporarily stored in the hard disk 14 or the like. Then, when the program is started, the temporarily stored image data is read, and a recognition process is performed to determine whether or not an image of a specific pattern exists in the input image. Output to etc.

[0029]

As is apparent from the above description, according to the present invention, a polygon having a vertex at a critical point of curvature is obtained from the contour of an object image input from an image input device such as a scanner or a digital camera. The feature of the shape of the object is extracted from the feature of the polygon, or the object is specified, and more specifically, the plant can be classified using the feature of the leaf of the plant.

[Brief description of the drawings]

FIG. 1 is a diagram showing an example of the shape of a leaf of the maple family.

FIG. 2 is a diagram showing an example of a basic structure of a leaf.

FIG. 3 is a flowchart illustrating an example of a hierarchical classification.

FIG. 4 is a diagram illustrating an example of polygonal approximation.

FIG. 5 is a diagram showing an example of a polygonal approximation of a broken shape in the middle of a leaf.

FIG. 6 is a configuration diagram for explaining an example of a recognition unit suitable for use in the embodiment of the present invention.

[Explanation of symbols]

1 ... leaves, ₁ 1 to 1 ₅ ... debris, A, A ₁ to A ₅ ... point having a positive threshold value (vertex), B ... minimum point of negative curvature, C ₁ -C ₈ ... line, 11 ... CPU, 12 ... memory, 13 ... display device, 1
4: Hard disk, 15: Keyboard, 16: CD-
ROM, 17 CD-ROM drive, 18 image input means.

Claims (6)

[Claims] 1. A polygon having a vertex at a critical point of curvature is constructed from a contour of an object image input from an image input device such as a scanner or a digital camera, and a contour shape feature is hierarchically represented by the polygon feature. A method for extracting the shape features of an object image that is described in a hierarchical manner. 2. A polygon having a vertex at a critical point of curvature is constructed from a contour of an object image input from an image input device such as a scanner or a digital camera, and a contour shape is formed using characteristics of the polygon. An image object recognition method that classifies hierarchically. 3. A means for constructing a polygon having a vertex at a critical point of curvature from an outline of an object image input from an image input device such as a scanner or a digital camera, and using the characteristics of the polygon to form an image object. An image object recognition device comprising: means for classifying; and means for controlling detailed classification of a shape by the number of sides of a polygon. 4. A means for constructing a polygon having a vertex at a critical point of curvature from a contour of a leaf of a plant image input from an image input device such as a scanner or a digital camera. An object image shape feature extraction device that describes leaf features hierarchically. 5. A polygon having a vertex at a critical point of curvature is constructed from a contour of a leaf of a plant image input from an image input device such as a scanner or a digital camera, and an image is formed using characteristics of the polygon. An image object recognition method that classifies plants hierarchically. 6. A means for constructing a polygon having a vertex at a critical point of curvature from a contour of a leaf image of a plant image input from an image input device such as a scanner or a digital camera.
An image object recognition apparatus comprising means for classifying plants using the characteristics of the polygon and means for controlling detailed classification of the shape by the number of sides of the polygon.