JP2009163383A - Method and apparatus for evaluating feature of outline of curved object - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To extract the outline of a curved object from an image and precisely evaluate its features. <P>SOLUTION: When the outline of a curved object is extracted from an image to evaluate its features, a multiplicity is determined that is the number of repeated detections when the features are detected again as set intervals including the features are narrowed down toward the features in sequence. An evaluation value matching the multiplicity is used. Using the multiplicity, the features of the peaks of the outline can be evaluated. Further, evaluation values matching indeterminate points generated randomly on the outline can be added. Also, the maximum value of evaluation values for the indeterminate points can be made smaller than the minimum value of evaluation values for the peaks. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a characteristic evaluation method and apparatus for a curved object contour, and in particular, extracts a contour of a curved object from an image suitable for use in positioning of a component in a surface mounting apparatus for electronic components, The present invention relates to a characteristic evaluation method and apparatus for a curved object outline suitable for use in evaluating the characteristics, particularly depressions and protrusions.

  In image recognition such as analysis, positioning, defect inspection, discrimination, and semantic understanding using an image, the contour line to be recognized is important. Therefore, it has been widely practiced to extract abstract data such as straight lines and arcs from contour lines and perform recognition using them.

  For example, in Patent Document 1, a contour line is described by a direction code (0 to 7), a change in a curve is calculated, and an averaged result is generated, and depressions and protrusions are detected based on this. .

  In Patent Document 2, the rate of change (curvature) is monitored in the tangential direction of the contour line, and a convex curve segment, a concave curve segment, and a straight line segment are extracted based on the sign and size.

  In Patent Document 3, straight lines and arcs are simply repeated by starting from an initial state in which the contour line is decomposed to an arbitrary length, shortening if the error exceeds the reference value, and increasing if the error is less than the reference value. Is extracted. At this time, the increase / decrease value at the time of correction is characterized by being set to ½ of the previous increase / decrease value, and fast convergence similar to the binary search is effective.

  Furthermore, Patent Document 4 describes that a contour line of a component is extracted using a conical shell that spreads in a spiral shape to obtain a component feature on the contour line.

Japanese Examined Patent Publication No. 1-30183 Japanese Patent Laid-Open No. 3-250380 Japanese Patent No. 3883993 JP 2001-251097 A

  However, each of the prior arts only outputs one point of coordinates, which is the tip of the depression or protrusion, and the tangential direction thereof, and has a well-shaped and large depression or protrusion, and conversely a bad or small depression or protrusion. The protrusion could not be distinguished.

  That is, in FIG. 1, the depression 3 and the protrusions 2, 4, 5, 6, 7, and 8 of the object 1 can be extracted even by the conventional technique. However, the features most desired to be used when recognizing the object 1 are the well 3 having a good shape and a large depression or protrusion, and the protrusions 2, 4, 5, and 8. Protrusions 6 and 7 that are poorly shaped or small indentations or protrusions. However, the prior art does not output information for the distinction. In order to distinguish what should be used from what should be discarded, it is necessary to output an evaluation value for that purpose.

  The protrusions 6 and 7 are not always unnecessary. As illustrated in FIG. 2, when there is an object 1 ′ similar to a circle and the only changes appearing on the outer periphery thereof are protrusions 6 and 7, these are important positioning information. That is, the necessity is unnecessary, and the evaluation value in question needs to be the highest in the evaluation values of the protrusions 6 and 7 in the case of a circular shape. However, conventionally, such an evaluation value cannot be appropriately determined.

  The present invention has been made to solve the above-described conventional problems, and an object of the present invention is to make it possible to appropriately evaluate the characteristics of the contour line.

  In the present invention, when a contour of a curved object is extracted from an image and the feature is evaluated, when a set section including the feature is sequentially fit toward the feature, the feature is detected again. The multiplicity which is the number of repeated detections is obtained, and the evaluation value corresponding to the multiplicity is used to solve the above problem.

  Here, using the multiplicity, it is possible to evaluate the characteristics of the depressions and protrusions (collectively referred to as peaks) in the contour line.

  Furthermore, an evaluation value corresponding to an indefinite point randomly generated on the contour line can be added.

  Further, the maximum value of the indefinite point evaluation value can be made smaller than the minimum value of the mountain evaluation value.

  The present invention also extracts a contour line of a curved object from an image and evaluates the feature of the curved object contour line when the set section including the feature is sequentially fit toward the feature. A feature evaluation apparatus for a curved object contour line, comprising: means for obtaining multiplicity which is the number of repeated detections when a feature is detected again; and means for obtaining an evaluation value corresponding to the multiplicity Is to provide.

  According to the present invention, it is possible to obtain a correct evaluation value of the feature of the contour line of a curved object. In particular, the multiplicity can appropriately represent the depth (height) and the shape of the mountain.

  Further, by adding an evaluation value corresponding to an indefinite point, which is an indefinite feature, to an evaluation value corresponding to the multiplicity, which is a firm feature, it is possible to recognize even a curved object having almost no feature on the outer periphery.

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

  FIG. 3 is a diagram showing a block configuration of the main part of the graphic feature data extracting apparatus for carrying out the present invention.

  In the present embodiment, for example, an image input unit 104 that captures and inputs the object 1 illuminated by the illumination 100 with the television camera 102, an image contour line extraction unit 106 that extracts the contour line of the object from the image, The figure feature data extraction unit 110 and object feature data database 112 for extracting object feature data from the line and all the Remarks (dents and protrusions) from the contour lines extracted by the image contour line extraction unit 106 are found. R generating unit 120 for outputting, R evaluating unit 122 for embedding unnecessary evaluation value ν in each R output from R generating unit 120, and T generating unit for outputting an indefinite point (Transient) from the contour line 124, a T evaluation unit 126 that embeds an unnecessary evaluation value ν in each T output from the T generation unit 124, and the whole is controlled to image the next object, and object feature data Extracted, and a control unit 130 for output to the database 112 of the object feature data are mainly provided.

  Hereinafter, examples will be described in detail.

  The present invention relates to a method for outputting good R and good T from a contour line.

  In FIG. 1, R is a Remark, that is, a depression or a protrusion, and is a depression 3, a protrusion 2, 4, 5, 6, 7, 8.

  T is a Transient, that is, an indefinite point (random point), and is an indefinite point 9 that is randomly generated on the circumference when the number of R is insufficient.

  FIG. 4A shows the outline structure. The contour line is a broken line, and is a row of vertex coordinates 31 that proceed in the direction of travel 30. The tangential direction 33 can be obtained by leveling the neighboring vertex row 32 at each vertex.

  First, length evaluation values ζ are attached to all contour lines. ζ has a value of 0 to 1, and the larger the length λ of the contour line, the larger ζ.

  In FIG. 5, the contour 1 of the object to be recognized is the longest (λ = MAX), the contour 13 of the external foreign object is the next longest, and the contour 12 of the scratch or pattern inside the recognition object is the shortest. (Λ = MIN).

  FIG. 6A shows an example of a graph (function) of λ and ζ. This function is ζ = 1 when λ = MAX and ζ = C1 when λ = MIN. C1 is a positive constant and may be an arbitrary value. For example, C1 = 0.1.

  The function may be appropriate if it is an increasing function of λ. In the embodiment, a quadratic function having a base at coordinates (MIN, C1) and passing through the center point 70 is set. Note that the graph from the center point 70 in the latter half of the graph to the coordinates (MAX, 1) is the rotation of 180 ° in the first half.

  According to this evaluation function, ζ of the contour line 1 in FIG. 5 is 1.0, ζ of the contour line 12 is 0.1, and ζ of the contour line 13 is 0.2 (for example). Yes.

  Next, R is generated.

  FIG. 7A shows the structure of R. R has a tangent direction of coordinates and contour lines. Moreover, it has a comprehensive evaluation value (= necessity degree) ν (0.5 to 1.0). The other data is a material for generating ν.

  FIG. 4 shows R target coordinates output in this embodiment. FIG. 4A shows a shape sandwiched between straight lines. In this case, the vicinity of the vertex 34 closest to the intersection is the R target coordinate. FIG. 4B shows a curved shape, and the case where the circumference is symmetrical both microscopically and macroscopically. In this case, the vicinity of the vertex 40 closest to the intersection with a symmetrical center line (not shown) is the R target coordinate. FIG. 4C shows a shape sandwiched between a straight line and a curve, which is a typical example in the case of microscopic left-right symmetry and macroscopic left-right asymmetry. In this case, an intermediate coordinate between the coordinate 43 and the coordinate 45, for example, the coordinate 44 is a target coordinate of R. Here, the coordinate 43 is the R coordinate obtained in the micro range, and the coordinate 45 is the R coordinate obtained in the macro range.

  FIG. 8 is a flowchart showing a processing procedure of the R generation unit 120, which will be described in order according to this flow.

  In step S1, a sequence of peripheral lengths Γ (Γ1, Γ2,...) Is prepared in descending order. Γ may be determined appropriately from the object size.

  In step S2, it is positioned at the first Γ1. The number of R generations k is set to 0.

  In step S3, the following processing is continued as long as the next circumference Γ exists. The process ends when all Γ are processed.

  In step S4, it is positioned at the first contour line.

  In step S5, the start point S of the section (S0 in FIG. 9) is placed at the start point of the contour line.

  In step S6, the following processing is continued as long as the next contour line exists. When all the contour lines have been processed, the process returns to step S3.

  In step S7, the state transition flag state is set to OFF (not having the best mountain). Mountain is a general term for depressions / projections.

  In step S8, the following processing is continued until S reaches the end point of the contour line. However, this is a process necessary when the contour line is closed (= 1 round). When the contour line is opened, the process may be interrupted when reaching the point where Γ has returned from the end point.

  In step S9, the mountain detection flag mnt is turned OFF (not mountain).

  In step S10, the vertex E where the circumference from S is equal to or greater than Γ for the first time is searched. That is, in FIG. 9, the current Γ = Γ1, the current S = S0, and if the distance between the vertices is continuously accumulated starting from S, the vertex E0 equal to or exceeding Γ1 is reached. This S0 to E0 is called the current section.

  In step S11, a mountain-shaped preliminary inspection is performed. This has the following inspections.

  (1) The number of vertices (excluding S0 and E0) in the current section is sufficiently large. This can be, for example, a constant = 3 or more. However, the constant is arbitrary.

  (2) The distance S0E0 of the opening should be large to some extent in order to remove beards. This can be, for example, 0.2 × Γ1 or more. However, the constant is arbitrary.

  (3) All the vertices (except S0 and E0) of the current section are unevenly distributed on either side of the straight line S0E0.

  (4) The depth M0G0 of the current section is sufficiently large. This can be, for example, 0.2 × Γ1 or more. However, the constant is arbitrary.

  M0 is obtained as follows. That is, when the circumference of the current section is γ (γ ≧ Γ1), the circumference from S0 is measured, and the point of γ / 2 is obtained and set as M0. Therefore, since M0 generally falls between vertices, it is important to obtain accurate coordinates by linear interpolation. G0 is the center of S0E0.

  If all the preliminary inspections are passed, the process goes to the next step S13. If either fails, go to step S17.

  In step S13, an intersection angle inspection is performed. This requires that S0E0 and M0G0 are close to orthogonal (= 90 °). This is determined by | cos θ | ≦ D1. For example, D1 = cos 80 °. However, the constant is arbitrary.

  In step S14, left-right symmetry inspection is performed. This requires that βs and βe, which are tangential directions of S0 and E0, are opened at the same angle with respect to the center line M0G0.

  The calculation is performed as follows. Let the direction cosine of the vector M0G0 be (cos α, sin α). By appropriately leveling the vicinity of S0, the tangential direction βs in the direction of the drawing can be obtained. The same applies to βe. Both are obtained in the form of direction cosine.

  φs = βs-α and φe = βe-α are calculated.

  Next, φ = φs + φe is calculated. If perfect left-right symmetry, φ = 0 ° = (1, 0). Therefore, in this inspection, this is determined by | sinφ | ≦ D2. For example, D2 = sin 10 °. However, the constant is arbitrary.

  In step S15, an evaluation value e = | sin θ || cos φ | e has a value of 0 to 1, and approaches 1 as it is closer to right angle and closer to left-right symmetry. The closer to 1, the better. e is used to find the best mountain when there is a sequence of mountains.

  In step S16, mnt is turned ON. This indicates that one mountain has been found.

  In step S17, the state transition flag state is viewed.

  If it is determined in step S18 that mnt is ON and the best mountain is not present, if this is a mountain, in step S19, the current mountain is copied to the best mountain and set as the best mountain candidate. That is, if it is found, it is not output immediately. Set ON to state.

  In step S20, if mnt is OFF and has the best mountain, if it is not a mountain, it can be output for the first time, so the best mountain is outputted as R in step S21. Set “state” to OFF, and again there is no mountain.

  Hereinafter, processing when outputting the best mountain as R will be described.

  In FIG. 9, the best mountain is assumed to be in the section S0E0. Here, the same name as S0E0 used for the description of the current section is used first, so be careful not to confuse it.

  (1) Find the vertex P0 having the longest distance from the straight line S0E0. When there are a plurality of points, they may be the vertices of the same maximum distance near the center of the start and end.

  (2) A perpendicular is dropped from P0 to S0E0, and the foot is set to H0.

  (3) On H0P0 (length h), near the center (for example, within h / 4 to (3/4) h), with an appropriate spread, from H0 to P0, for example, divided into four equal parts and five Generate a sequence of points. However, a constant for determining the position may be appropriate.

  (4) A straight line orthogonal to H0P0 (and thus parallel to S0E0) is drawn from each occurrence point (for example, straight line 50).

  (5) The intersections 51 and 52 of the straight line 50 and the S0 side and the E0 side of the current section are obtained.

  (6) The midpoint 53 between the intersections is output. Repeat the above for 5 bottles.

  (7) Draw the least square approximation line 54 from the five midpoint sequences.

  (8) A perpendicular is dropped from P0 to the straight line 54, and the leg of the perpendicular is z0. z0 is the output coordinate Q1 of this mountain.

  (9) The tangential direction τ0 of the mountain at z0 is made from the direction obtained by rotating the vector H0P0 by ± 90 °. In FIG. 9, since the current section is counterclockwise, it rotates by + 90 ° (if it is clockwise, it rotates by −90 °).

  Thereby, the output Q of the mountain can be obtained. That is, Q = (coordinate z0, tangential direction τ0). Note that in FIG. 10, Q is output for each Γ. That is, Γ1 to Q1, Γ2 to Q2,... And Γ5 to Q5 continue.

  All the outputs Q including Q1 are inspected to determine whether they are in the vicinity of ε of individual Rs already registered. This is because the number of R is not increased unnecessarily. Here, ε may be determined appropriately, for example, 0.1 times the object length.

  Here, the first Q1 is registered. In fact, there are no mountains around. The registered Q is called R.

  Copy ζ of the current contour line to ζ of R.

  The interval length γ (actual length from S0 to E0) of Q1 is copied to γ of R. This γ must not be changed in subsequent processing.

  Also, m called R multiplicity is set to 1.

  The number k of R is increased by one.

  Since Q2 of the second time is in the vicinity of ε of R, Q2 is not registered. Instead, it has the role of correcting R.

  That is, in FIG. 11, the coordinates of Q1 (= R) and Q2 are q1 and q2, respectively, and the tangential direction (direction cosine) is ω1 and ω2, respectively.

  The new coordinate q ′ is obtained by dividing q1 and q2 into Γ2: Γ1.

  The new tangential direction ω ′ is obtained by internally dividing ω1 and ω2 into Γ2: Γ1. As a method at this time, an opening angle δ is obtained, and δ is close to 0 ° (actually almost less than 2 °), and sin δ (signed) may be internally divided into Γ2: Γ1. That is, sin β = (Γ2 / (Γ1 + Γ2)) sinδ of the correction angle β (not shown), and therefore cosβ = √ (square of 1−sinβ). ω ′ is obtained by rotating ω1 by β.

  The above coordinate correction and angle correction are very important. By this modification, the R coordinate and the tangential direction can be stabilized.

  Finally, the multiplicity m of R is increased by 1 to 2. This completes the correction by Q2.

  Repeat the same thing after Q3. When the process is repeated up to the final Q5, the multiplicity m of R is 5.

  The role of multiplicity will be described later.

  If it is determined in step S22 that the mountain is the best, if it is a mountain, the mountain is continuous. Accordingly, in step S22, the best mountain and the current mountain are selected. That is, if the evaluation value of the current mountain is higher in step S23, the current mountain is copied to the best mountain. If it is the same or lower, do nothing.

  In step S24, whether or not there is a current mountain, the circumference of S0 is moved by δ1, and the next step is S1. However, δ1 is the vertex position displacement, for example, δ1 = 1 (moves one by one). However, this constant is arbitrary.

  The following repeats the same from step 8 for the new S1.

  In step S21, it should be noted that a mountain is not output from a circle or in-phase, that is, a feature-free area.

  That is, in FIG. 12, it is assumed that the peak M0 indicated by the middle point has already been output and becomes R (the immediately preceding R).

  After that, an attempt is made to output the mountain M1, but it can be seen that the shape is almost the same as the previous R.

  That is, for example, when the distance of the opening SE is the same, the distance from M to SE is the same, and the relative directions of the tangential directions βs and βe of M0 and M1 are the same, You can conclude that there is.

  In such a case, the mountain M1 is not output, and the immediately preceding R itself is deleted from the output list.

  When the processing is completed for all Γ in step S3, R generation ends.

  After the generation of R is completed, the R evaluation unit 122 evaluates R.

  FIG. 6B shows an example of the evaluation function ω with multiplicity m.

  First, a maximum value (MAX) and a minimum value (MIN) are obtained for the m of the generated R set. FIG. 13 shows the multiplicity (indicated by M) of the generated R set. MAX = 5 and MIN = 1.

  As a feature, as shown in FIG. 14, a more preferable shape is good and a large multiplicity is output to a large mountain, and conversely, a small multiplicity is output to a bad or small mountain. Note that

  Returning to FIG. 6B, since the multiplicity m is entered in each of R, the function value ω in the figure is obtained with that m. The function used here may be made according to the same rules as those in the case of ζ evaluation in FIG. The constant C2 may be appropriate, for example, C2 = 0.2.

  Next, FIG. 6C shows an evaluation function ψ having a length γ. In the generated set of R, the maximum value (MAX) and the minimum value (MIN) are obtained for the γ.

  Since the length γ is entered in each R, the function value ψ in the figure is obtained with the γ. The function used here may be made according to the same rules as those used for ζ evaluation in FIG. The constant C3 may be appropriate, for example, C3 = 0.2.

Next, an evaluation value η of shape and size is set for each of R. Η of R is
η = multiplicity evaluation value × length evaluation value = ω × ψ
It is.

Next, a comprehensive evaluation value ν is set for each of R. Ν of R is
ν = 0.5 + 0.5ζη
It is. According to this equation, ν has a value of 0.5 to 1.0, the contour evaluation value ζ is larger, and the shape and size evaluation value η is larger.

  In FIG. 5, the black circle indicates that it is R and indicates its ν value. As a feature, note that a larger value of ν is output as R at a preferred point (note that the minimum value of R is 0.5).

  In particular, in FIG. 13, in the multiplicity, both R63 and R64 are m = 4 and cannot be distinguished, but it should be noted that in FIG. 2, there is a large difference between ν = 0.9 and ν = 0.6, respectively. This is due to the action of the outline evaluation value ζ.

  FIG. 15 shows the output R. When R is used for object recognition, for example, the shape of the triangle R1R2R9 is learned and recorded, and it is checked whether the same shape appears during recognition.

  In the figure, 12 Rs are output, but this number may not be sufficient. Depending on the application, information about the outer periphery may be desired.

  For example, the points T randomly arranged on the outer peripheral portion can be used for recognition information if the number is large.

  FIG. 7B describes the structure of T.

  T has a tangential direction of coordinates and contour lines. Moreover, it has a comprehensive evaluation value (= necessary degree) ν (0 to 0.5).

  For example, assume that the triangle R2R8T3 is learned. In recognition, since it is a different image, it generally occurs at a different place, for example, Tx.

  However, the triangle R1R8Tx is very similar to the learning triangle R2R8T3, so it can be output as a learning object possibility.

  Whether T is necessary or not should be according to the number k of R. FIG. 16 shows a generation function of the necessary number N of T. This can also be created based on the same principle as the ζ generation function shown in FIG. 6A (upside down). A sufficient number RE of R may be appropriate, for example, RE = 25. The function outputs k = RE and Nt = 0, that is, T is unnecessary. On the other hand, when k = 0, TS pieces are output. TS is a necessary number that can be recognized only by T, and may be determined from the size of the object and the like. For example, a constant TS = 80 may be assigned.

  The generation of T is quite simple. That is, in FIG. 15, it first occurs at the starting point of each contour line, proceeds around the circumference for each distance calculated in advance, and occurs there. However, it does not occur when R is in the vicinity of ε.

  The distance traveled by T may be determined from the quotient obtained by dividing the distance of the entire outline set by Nt.

  Ζ of T is a copy of the evaluation value of the contour line.

  Ν of T is set to ν = 0.5ζ. Thus, even when the contour is the maximum evaluation value ζ = 1, ν = 0.5, which is less than the minimum value of R, which is convenient for discriminating the two.

  FIG. 5 shows R and T finally output. The black circle is R and the white circle is T. Although not shown in the figure, each has a tangential direction of the outline. The ν value is one digit after the decimal point for easy understanding.

  According to this, obviously, ν is very high for features that you want to use, and low for features that you do not want to use. Note in particular that the internal noise 12 has a low ν and the external noise 13 has a low ν.

  In this way, a correct evaluation value ν of R (recess / projection) can be obtained. ν is an excellent guideline. It is a recognition target, and a stronger feature appears as a larger ν.

That is,
ν = 0.5 + 0.5ζη
= 0.5 + 0.5ζ (λ) ω (m) ψ (γ)
Then, ζ (λ) increases as the length λ of the parent contour increases. ω (m) increases as the multiplicity m increases. ψ (γ) increases as the circumference length γ increases. As described above, the ν value increases only when all three terms are large, and this is a correct evaluation value of R.

  In particular, as shown in FIGS. 6A, 6B, and 6C, ζ (λ), ω (m), and ψ (γ) are ranges of values of λ, m, and γ (MAX and MIN), respectively. Note that the function follows. That is, whether the evaluation value is high or low depends on the MAX value and MIN value of the set.

  For example, in FIG. 6B, the evaluation value ω (m) of the multiplicity m depends on the range of m. That is, as described above, as illustrated in FIG. 2, when there is a circular object 1 ′ and there are slight depressions / projections 6 and 7 on the circumference, their multiplicity is, for example, m = 1 and m There are times when only = 2. In this case, MAX = 2, MIN = 1, and the one obtained m = 2 is the maximum evaluation value ω = 1. This is a correct evaluation of the multiplicity of the depression / projection.

  Here, the multiplicity m is an index representing the depth (height) of the depression / projection and the shape.

  That is, in FIG. 14, (a), (b), and (c) are all depressions (projections) having the same opening length e. However, the multiplicity is different as shown in the figure. The reason why the multiplicity of (b) is smaller than that of (a) is that the depth (height) is small, and the reason why the multiplicity of (c) is smaller than that of (a) is that the shape of the tip is bad. Because.

  Thus, the value of multiplicity m is effective as one of the evaluations of R because the deeper the depression (projection) (the higher the height) and the better the shape.

  Further, recognition can be performed by R alone or a combination of R and T. At that time, each evaluation value ν is an excellent guideline, and only the good R or good T can be taken out to create the best set. Thereby, it is possible to recognize even a curved object having almost no feature on the outer periphery.

  In the embodiment, a quadratic function is used as the evaluation value generation function. However, any function may be used as long as it is a monotonically increasing or decreasing function. A straight line may be used as an extreme example. Even in that case, the effect is almost the same.

  In the embodiment, R (dent / protrusion) is used as a feature of the contour that can be evaluated with multiplicity m. However, it exists in addition to R. For example, L (straight line), C (circle), etc. are examples.

  FIG. 17A shows a case where the straight line portion (L) of the contour line is extracted by Γ (not shown) having different lengths. At this time, if the length of Γ used is five, m = 5 may be set. As shown in (b), when there is a section with severe irregularities and can be detected as a straight line when viewed at a long distance, but not at a short distance, only three lengths as shown in the figure. Therefore, the multiplicity m = 3.

  FIG. 18 is extracted with Γ (not shown) having different arc lengths (C) of the contour line. At this time, if the length of Γ used is five, m = 5.

  Thus, the multiplicity exists in every feature shape extracted from the contour line, and can be a scale representing the size of the feature shape and the goodness of the shape.

Diagram showing an example of the shape of a curved object Diagram showing an example of a circular object The block diagram which shows the principal part structure of the graphic feature data extraction apparatus for implementing this invention The figure which shows the target coordinate of R The figure which shows the example of provision of the evaluation value by this invention The figure which shows the example of the evaluation function used by this invention The figure which similarly shows the structure of R and T A flow chart showing the procedure for generating R The figure for demonstrating the processing method of an Example similarly The figure for explaining the processing method similarly The figure for explaining the processing method similarly The figure for explaining the processing method similarly The figure for explaining the processing method similarly The figure which shows the multiplicity of this invention The figure for demonstrating the processing method of an Example The figure which similarly shows the example of the production | generation function of the required number Nt of T The figure which shows the example of extraction of the linear part of an outline The figure which shows the example of extraction of the circular arc part of an outline

Explanation of symbols

DESCRIPTION OF SYMBOLS 1, 1 ', 12, 13 ... Contour line 104 ... Image input part 106 ... Image outline extraction part 110 ... Graphic feature data extraction part 112 ... Graphic feature data database 120 ... R generation part 122 ... R evaluation part 124 ... T generation Unit 126 ... T evaluation unit 130 ... Control unit

Claims (5)

When extracting the contour line of a curved object from an image and evaluating its features,
When the setting section including the feature is sequentially fit toward the feature, the multiplicity which is the number of repeated detections when the feature is detected again is obtained,
A characteristic evaluation method for a curved object outline, wherein an evaluation value corresponding to the multiplicity is used.   The feature evaluation method for a curved object outline according to claim 1, wherein the feature of the peak of the outline is evaluated using the multiplicity.   3. The characteristic evaluation method for a curved object outline according to claim 1, further comprising adding an evaluation value corresponding to an indefinite point randomly generated on the outline.   4. The characteristic evaluation method for a curved object outline according to claim 3, wherein a maximum value of the indefinite point evaluation value is smaller than a minimum value of the mountain evaluation value. In a feature evaluation apparatus for a curved object contour that extracts a contour of a curved object from an image and evaluates its characteristics,
Means for obtaining a multiplicity which is the number of repeated detections when the feature is detected again when the setting section including the feature is sequentially fit toward the feature;
Means for obtaining an evaluation value according to the multiplicity;
A characteristic evaluation apparatus for a curved object outline characterized by comprising: