JP2011043998A - Method and apparatus for analyzing image - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image analysis method and an image analysis apparatus to effectively identify an object. <P>SOLUTION: An identification object is identified from a picked-up image according to a suitably weighted pattern by: finding a feature amount of each unit area A3 of a pattern area A2; finding similarly a feature amount of each unit area A5 of an extracted area A4; finding a degree of matching composed of them by weighting for each method; determining whether each unit area of extracted area is an identification object or a background from a value of the composed degree of matching; finding a border line of the identification object and the background according to the determination concerned; finding an error between a position on an actual border line in a picked-up image and the border line obtained by the determination; and optimizing the pattern of weighting by changing serially so that the error concerned may become sufficiently small. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to an image analysis method and an image analysis apparatus for identifying an object.

  As a method for identifying an identification target object from captured image data in which a luminance value is determined for each pixel of the captured image, a method for obtaining an outline of the identification target object from luminance gradient information based on the captured image data, an image A method is known in which edge information of an object is extracted from a grayscale image based on data, and an object is extracted and a position is detected using an edge point sequence (for example, Patent Documents 1 and 2).

Japanese Patent Laid-Open No. 9-171553 Japanese Patent No. 2981382

However, in the method of acquiring the contour line based on the luminance gradient shown in Patent Document 1, a clear contour line is obtained unless there is a luminance difference that provides a sufficient luminance gradient between the object and the background. There is a problem that it cannot be obtained and cannot be identified unless there is a sufficient brightness difference between the object and the background.
In the case of the prior art shown in Patent Document 2, it is assumed that edge information can be obtained from an object, and sufficient edge information cannot be obtained unless there is a sufficient brightness difference between the object and the background. There was a problem that identification was difficult.

  An object of the present invention is to enable stable identification of an object.

  The invention according to claim 1 is an image analysis method performed by an image analysis apparatus that identifies an identification target based on captured image data of a captured image including an identification target having a specific pattern on the surface and a surrounding background. A reference feature amount obtaining step of obtaining a feature amount of a texture pattern serving as a reference by a plurality of methods for a predetermined unit region in a pattern region including only the specific pattern specified in advance in the captured image; The extracted region including at least the pattern region and the surrounding background in the captured image is divided into unit regions having the same size as the unit region of the pattern region, and the feature amount is obtained for each unit region by the respective methods. For each unit area of the comparison feature quantity acquisition step and the extraction area, a degree of coincidence with the pattern area is obtained for each feature quantity of each technique, and for each feature quantity of each technique, A degree of coincidence obtaining step for obtaining a degree of coincidence obtained by weighting the degree of match individually, and a step of determining whether each unit area of the extraction region is an identification target or a background from the value of the degree of coincidence, A boundary line obtaining step for obtaining a boundary line between the identification target and the background in the extraction region from the plurality of unit regions determined as the identification target in the determination step, a position on the actual boundary line in the captured image specified in advance, and the An error detection step for detecting an error from the boundary line acquired in the boundary line acquisition step, and the matching degree acquisition step and determination while sequentially changing the weighting pattern for the matching degree for each feature amount of each method By repeating the steps, the boundary line acquisition step and the error detection step, a weighting pattern in which the error is less than or equal to a threshold is specified, and a captured image is obtained according to the acquired weighting pattern. Characterized by identifying the identification target from.

  The invention described in claim 2 has the same configuration as that of the invention described in claim 1, and the plurality of methods for obtaining the feature values use a part or all of the 14 feature value calculation methods of the co-occurrence matrix. It is characterized by that.

  The invention according to claim 3 is an image analysis apparatus for identifying the identification target based on captured image data of a captured image including an identification target having a specific pattern on the surface and a surrounding background. A reference feature amount acquisition means for obtaining a feature amount of a texture pattern serving as a reference by a plurality of methods for a predetermined unit region in a pattern region including only the specific pattern specified in advance in the captured image; and at least in the captured image A comparison feature amount acquisition unit that divides the extraction region including the pattern region and the surrounding background into unit regions having the same size as the unit region of the pattern region, and obtains the feature amount for each unit region by the respective methods. For each unit area of the extraction area, the degree of coincidence with the pattern area is obtained for each feature amount of each method, and the degree of coincidence for each feature amount of each method is individually weighted A degree of coincidence obtaining means for obtaining a degree of coincidence obtained by combining the image, a determination means for determining whether each unit area of the extraction area is an identification target or a background from the value of the degree of coincidence synthesized, and identification by the determination means A boundary line acquisition means for obtaining a boundary line between the identification target and the background in the extraction area from a plurality of unit areas determined to be targets, and a position on the actual boundary line in the captured image specified in advance and the boundary line acquisition means An error detecting means for detecting an error with the acquired boundary line, and obtaining the degree of coincidence by the coincidence degree obtaining means while sequentially changing the weighting pattern for the degree of coincidence for each feature amount of each method, A weighting pattern in which the error is equal to or less than a threshold value by repeating determination by the determination unit, acquisition of the boundary line by the boundary line acquisition unit, and error detection by the error detection unit. Identify, identifying the identification object from the captured image in accordance with the acquired weighting pattern.

  The invention described in claim 4 has the same configuration as that of the invention described in claim 3, and the plurality of methods for obtaining the feature values use a part or all of the 14 feature value calculation methods of the co-occurrence matrix. It is characterized by that.

According to the first and third aspects of the present invention, a feature amount is obtained and synthesized from a pattern region of an identification target in captured image data by a plurality of methods, and a boundary between the unit region having the matching feature amount and the background of the identification target A line is obtained, and a retry pattern is repeated so that an error from an actual boundary position becomes small, thereby specifying a weighting pattern related to the degree of matching of feature amounts. For this reason, the target object is identified from the feature amount based on the unique pattern of the identification target, and the identification target can be effectively identified regardless of the brightness difference from the background.
In addition, since a suitable weighting pattern for a plurality of methods related to a texture pattern is specified, it is possible to eliminate the influence of bias in the identification accuracy of the method and realize more accurate identification.

  The inventions of claims 2 and 4 use the 14 feature quantity calculation method of the co-occurrence matrix established as a technique for obtaining the feature quantity of the texture pattern, so that the identification target of various patterns is more effective. Can be identified.

It is a top view which shows the whole electronic component mounting apparatus which concerns on this Embodiment. It is a block diagram which shows the control system of an electronic component mounting apparatus. It is a functional block diagram of an image analysis device. It is a flowchart of learning data creation processing. It is explanatory drawing which shows the various area | regions set in the captured image of captured image data. FIG. 6A is an explanatory diagram illustrating the relationship between the pattern region and the unit region, and FIG. 6B is an explanatory diagram illustrating the relationship between the extraction region and the unit region. FIG. 7A is an explanatory diagram showing an example of the luminance value in each pixel of the unit area, FIG. 7B is an explanatory diagram showing a concept of displacement in the co-occurrence matrix, and FIG. 7C is a co-occurrence matrix. It is explanatory drawing shown. It is a diagram which shows the normal distribution of the feature-value of a pattern area | region. It is a diagram which shows the relationship between the difference | error of a feature-value, and a matching degree. It is explanatory drawing which showed whether each unit area | region is a solder. It is explanatory drawing which overlapped and illustrated the actual outline and the production | generation outline. It is explanatory drawing which shows the example of the captured image at the time of actual mounting operation.

(Overall configuration of the embodiment)
An embodiment of the present invention will be described with reference to FIGS. In the present embodiment, an image analysis apparatus 10 for identifying an identification object having a specific texture in a captured image as a background is applied to the electronic component mounting apparatus 100, and a solder pattern that is the mounting position of the electronic component is identified. An example in the case of being used for the purpose of identifying as an object is shown.

(Electronic component mounting equipment)
The electronic component mounting apparatus 100 mounts various electronic components on a substrate. As shown in FIG. 1, a plurality of electronic component feeders 101 that supply electronic components to be mounted as electronic component mounting means. A feeder bank 102 as an electronic component supply unit that holds a plurality of electronic component feeders 101 side by side, a substrate transport unit 103 that transports a substrate in a certain direction, and a substrate transport path by the substrate transport unit 103. A mounting operation unit 104 for performing an electronic component mounting operation on a substrate, a head 106 as a component holding means for holding an electronic component by holding a suction nozzle 105 that sucks the electronic component, and a head 106 within a predetermined range. An XY gantry 107 serving as a head moving means for driving and transporting to an arbitrary position and a head 106 are mounted to image a substrate. There is provided with a CCD camera 108 as an imaging means for identifying the solder pattern that performs mounting of an electronic component, and a control unit 120 for controlling the respective configuration of the electronic component mounting apparatus 100.
The image analysis device 10 is incorporated in the control device 120 to identify the solder pattern from the captured image of the substrate of the CCD camera 108 for mounting electronic components, and to reflect accurate position information in the positioning operation of the head 106. Used.
In the following description, one direction orthogonal to each other along the horizontal plane is referred to as an X-axis direction, the other direction is referred to as a Y-axis direction, and a vertical vertical direction is referred to as a Z-axis direction.

The substrate transport unit 103 includes a transport belt (not shown), and transports the substrate along the X-axis direction by the transport belt.
As described above, the mounting operation unit 104 for mounting electronic components on the substrate is provided in the middle of the substrate transfer path by the substrate transfer means 103. The substrate transport unit 103 transports the substrate to the mounting work unit 104 and stops, and holds the substrate by a holding mechanism (not shown). That is, a stable electronic component mounting operation is performed while the substrate is held by the holding mechanism.

The head 106 includes a suction nozzle 105 that holds electronic components by air suction at the tip, a Z-axis motor 111 (see FIG. 2) that is a drive source that drives the suction nozzle 105 in the Z-axis direction, and a suction nozzle 105. And a rotation motor 112 (see FIG. 2), which is a rotation drive source for rotating the electronic component held via the rotation centering around the Z-axis direction.
Further, each suction nozzle 105 is connected to a negative pressure generation source, and suction and suction of electronic components are performed by performing suction suction at the tip of the suction nozzle 105.
That is, with these structures, at the time of mounting, the electronic component is sucked from the predetermined electronic component feeder 101 at the tip of the suction nozzle 105, and the suction nozzle 105 is lowered toward the substrate and rotated at the predetermined position. The mounting work is performed while adjusting the orientation of the electronic components.
The above-described CCD camera 108 is mounted on the head 106, and the head 106 is driven by the XY gantry 107 so as to be positioned at the imaging position of the substrate.

The XY gantry 107 includes an X-axis guide rail 107a that guides the movement of the head 106 in the X-axis direction, and two Y-axis guide rails 107b that guide the head 106 in the Y-axis direction together with the X-axis guide rail 107a. , An X-axis motor 109 (see FIG. 2) that is a drive source that moves the head 106 along the X-axis direction, and a Y-axis that is a drive source that moves the head 106 in the Y-axis direction via the X-axis guide rail 107a And a motor 110 (see FIG. 2). Then, by driving each of the motors 109 and 110, the head 106 can be conveyed to almost the entire region between the two Y-axis guide rails 107b.
Each motor recognizes the rotation amount of each motor by the control device 120 and is controlled so as to obtain a desired rotation amount, thereby positioning the suction nozzle 105 and the CCD camera 108 via the head 106. Is going.
Further, due to the necessity of electronic components, both the feeder bank 102 and the mounting work unit 104 are arranged in a region where the head 106 can be transported by the XY gantry 107.

The feeder bank 102 includes a plurality of flat portions along a plurality of XY planes, and a plurality of electronic component feeders 101 are arranged and mounted on the flat portions along the X-axis direction.
The CCD camera 108 is held in a state of being directed downward by the head 106, and the solder pattern is identified by imaging the substrate from above.

(Control device)
FIG. 2 is a block diagram showing a control system of the electronic component mounting apparatus 100. As shown in FIG. 2, the control device 120 mainly includes an X-axis motor 109, a Y-axis motor 110 of the XY gantry 107, a Z-axis motor 111 that moves the suction nozzle 105 up and down in the head 106, and the suction nozzle 105. A CPU 121 that performs operation control of the rotary motor 112 that performs rotation and the CCD camera 108, executes various processes and controls according to a predetermined control program, and a system ROM 122 that stores programs for executing various processes and controls , RAM 123 serving as a work area for various processes by storing various data, I / F (interface) (not shown) for connecting the CPU 121 and various devices, and input of data required for various settings and operations An operation panel 124 for performing various operations and setting data for executing various processes and controls are stored. And EEPROM126 that, and a display monitor 125 for displaying the results of tests that contents and later in various settings. Moreover, each motor 109-112 mentioned above is connected to the control apparatus 120 via the motor driver which is not shown in figure.

(Image analysis device)
FIG. 3 is a functional block diagram of the image analysis apparatus 10. Each function to be described later of the image analysis device 10 is actually realized by the CPU 121 of the control device 120 executing a predetermined program.
The image analysis apparatus 10 includes a main control unit 11 that performs overall control, an image input unit 20 that acquires captured image data from the output of the CCD camera 108, and an image dividing unit that divides a rectangular image region based on the captured image data. 30, an error measurement point input unit 40 for setting a plurality of points designated on the boundary line between the solder pattern in the captured image and its background by the operation panel 124 as an error measurement point, and a texture (pattern) of the solder pattern The feature quantity extracting unit 50 for extracting the feature quantity, the feature quantity synthesizing section 60 for synthesizing the feature quantity by determining parameters to be described later for synthesizing the extracted feature quantities, and the error between the error measurement point and the output shape And an error detection unit 70 for making a determination.
With these configurations, the image analysis apparatus 10 performs in advance a learning data creation process for acquiring appropriate parameters for synthesizing the feature amount of the solder pattern before performing the mounting operation of the electronic component. Sometimes, a process for identifying a solder pattern from an image captured by the CCD camera 108 is executed based on parameters obtained by the learning data creation process.

(Learning data creation process)
Next, the learning data creation process of the solder pattern performed by the image analysis apparatus 10 will be described in the order of the processes shown in the flowchart of FIG.
First, the image input unit 20 captures an image of a board including a solder pattern with the CCD camera 108, and the image is stored as captured image data (step S1).

Next, the error measurement point G is input by the operator, and the error measurement point input unit 40 performs a process of recording the position in the image designated by the input operation (step S2).
FIG. 5 shows a captured image captured by the CCD camera 108. An outermost rectangular area A1 indicates the range of the captured image, and H indicates a captured solder pattern.
Since the surface of the solder is uniformly covered with a granular pattern peculiar to the solder, at the time of mounting an electronic component, the characteristic amount of the pattern is obtained to identify the background, and the boundary line between the solder pattern H and the background The contour line R is obtained and an electronic component is mounted. In this learning data creation process, the operator sets the contour R based on the captured image displayed on the display device 125 in order to acquire appropriate parameters (described later) necessary for identifying the contour R from the captured image. By visually recognizing and inputting a plurality of arbitrary locations (six in this example) on the contour line as error measurement points G from the operation panel 124 by, for example, specifying a position by a cursor operation or the like, the image analysis apparatus 10 The position coordinates in the captured image area of the input error measurement point G are stored in the RAM 123.

Next, the pattern area A2 for acquiring the feature quantity of the solder pattern is input by the operator, and the feature quantity extraction unit 50 performs a process of recording the position and range of the area designated by the input operation. (Step S3).
The size of the pattern area A2 is not limited, but it is more preferable that the pattern area A2 has at least about 30 unit areas A3 of 8 × 8 pixels to be described later. Is not included.
That is, similarly to the input of the error measurement point G described above, based on the captured image displayed on the display device 125, the operator specifies and inputs the four corners of the specified range from the operation panel 124 by, for example, a cursor operation or the like. Thus, the pattern area A2 is designated and input, and the image analysis apparatus 10 stores the coordinates indicating the position and range of the input pattern area A2 in the captured image area in the RAM 123.

Next, the feature amount extraction unit 50 performs a process of acquiring the image feature amount of the texture pattern from the input pattern region A2 (step S4: reference feature amount acquisition step). In this process, the feature quantity extraction unit 50 functions as a “reference feature quantity acquisition unit”.
A co-occurrence matrix is used to acquire the image feature quantity of the texture pattern. The co-occurrence matrix is one of the statistical texture feature calculation methods among the texture features. Contrast and local uniformity that is one of the spatial patterns of the image from the co-occurrence matrix defined below. The feature amount can be obtained for 14 types of elements.

First, in order to generate a co-occurrence matrix, as shown in FIG. 6A, the pattern area A2 described above is divided by a horizontal direction dividing line LH and a vertical direction dividing line LV, and a square unit of 8 × 8 pixels is obtained. A process of cutting out the area A3 is performed. As described above, the number of unit areas A3 obtained by the division is about 30 (48 in the figure). As will be described later, since the feature quantities based on each unit area A3 are averaged and the standard deviation is obtained, 30 (48 in the figure) unit areas A3 are prepared as statistically sufficient numbers.
Then, a co-occurrence matrix is generated individually for each unit region A3.
The 8 × 8 unit image A3 shown in FIG. 7A will be described as an example. As shown in FIG. 7B, the luminance value i of the image (32 gradations: i = 0 to 31). A matrix shown in FIG. 7C having a probability Pδ (i, j) of a pixel whose luminance value is j, which is separated by a displacement δ = (r, θ) of length r in the direction of angle θ from the pixel of FIG. Is the co-occurrence matrix. In the present embodiment, the number of individual pixels whose luminance value is j with respect to the pixel having the luminance value i with the above-described displacement is input to the element of i rows and j columns. In the present embodiment, one co-occurrence matrix is generated for both directions of displacement r = 1 and angles θ = 0 ° and θ = 180 °. Similarly, co-occurrence matrices of θ = 45 (225), 90 (270), and 135 (315) are also generated.

  Then, for the co-occurrence matrix of each unit region A3, 14 types of feature quantities fi (i = 1 to 14) are calculated by the following formulas (1) to (14). The fi value is a real number. For example, f1 = angular second moment, f2 = contrast, f3 = entropy, etc. It is well known that the fi value has the ability to distinguish texture patterns. The significance of these texture features is explained in detail in “New Edition, Image Analysis Handbook, Supervised by Mikio Takagi / Yoshihisa Shimoda”, so the description is omitted here.

なお、各式において、ｎ＝31（最大輝度値）であり、また、次式(15)〜(18)の通り定義されるものとする。

In each equation, n = 31 (maximum luminance value) and is defined as the following equations (15) to (18).

  When the feature value fi value (i = 1 to 14) is obtained for all the unit regions A3 by the above processing, the average value μi and the standard deviation of all the unit regions A3 for each type of feature value (for each value of i). σ i is calculated and stored in the RAM 123 or the EEPROM 126 as a learning fi value as a feature amount of the pattern area A2. The feature amount fi of the pattern area A2 is generally considered to follow a normal distribution as shown in FIG.

Next, the feature quantity synthesis unit 60 initializes a synthesis parameter for the feature quantity (step S5).
The various feature quantities fi for each of the formulas (1) to (14) described above are obtained by the degree of coincidence E with the feature quantities calculated in the same way for the unit areas of the image other than the pattern area in the subsequent processing. In this case, the degree of coincidence E obtained for each of various feature amounts (for each i) is individually weighted and synthesized. The synthesis parameter indicates a combination pattern of coefficients wi (i = 1 to 14) for multiplying the matching degree E for each feature amount as a weight.
The synthesis parameter is defined to be w1 + w2 + w3 +... + W14 = 1, and when the synthesis parameter is initialized, w1 = 1. That is, all the other w2, w3, w4,. That is, only the degree of coincidence E regarding the feature amount f1 is obtained.

Next, a new extraction area A4 that wraps the entire solder in the input image 5 is set.
The feature amount synthesizing unit 60 identifies whether or not the pattern is a solder pattern based on the above synthesis parameter, and in order to form the outline of the solder pattern, the operator inputs a new extraction area A4 that wraps the entire solder. The feature amount synthesis unit 60 performs processing for recording the position and range of the region designated by the input operation (step S6).
The extraction area A4 is not limited in size as long as it includes a solder pattern, but as shown in FIG. 5, it is required to have a range including all the error measurement points G described above. A range that can surround the solder pattern is desirable. The input method is the same as that in the case of the pattern area A2.

Furthermore, as shown in FIG. 6B, the feature amount synthesis unit 60 divides the extraction region A4 into the same 8 × 8 pixel unit region A5 as in the case of the pattern region A2 (step S7: comparison feature amount acquisition step). ).
And the feature-value fi calculated | required from 14 type | formula mentioned above for every unit area | region A5 of extraction area | region A4 is calculated. At this time, the feature amount is calculated from the combination parameters in which wi (i = 1 to 14) ≠ 0. At this time, the feature quantity synthesis unit 60 functions as a “comparison feature quantity acquisition unit”.

Next, the feature amount synthesizing unit 60 calculates the degree of coincidence E between each unit area A5 of the extraction area A4 and the learning fi value of the area A3 of the pattern area A2 (step S8: coincidence degree acquisition step).
That is, for each unit area A5, an error δi with respect to the average value μi of the corresponding learning fi values for each feature amount fi is calculated. Then, the degree of coincidence ei is calculated for each feature amount fi of each unit region A5.
For the calculation of the degree of coincidence ei, the function eval shown in FIG. 9 is used. In this function eval, the vertical axis represents the degree of coincidence ei with the maximum value 1, and the horizontal axis represents the error δi. The function eval only needs to be monotonously decreased, and the change in the slope is arbitrary. In other words, a straight line that decreases monotonously may be used, but here, a curve is adopted in which the slope approaches 0 and the slope becomes maximum in the middle as the degree of coincidence approaches 0 and 1. That is, the value of ei is 0 ≦ ei ≦ 1, ei = 1 (δi = 0), ei = 0 (δi ≧ βσi), and the middle is monotonously decreasing. .sigma.i is a standard deviation of the learning fi value, .beta. is a coefficient for evaluating an error .delta.i between the feature value fi value and the average value .mu.i of the learning fi value, and the larger .beta. .
In this embodiment, for example, β = 1.5.
Thereby, the degree of coincidence ei is calculated for each feature amount fi of each unit region A5. Then, according to each coefficient wi of the synthesis parameter, each ei is multiplied and summed, and the degree of coincidence E is calculated for each unit region A5.
At this time, the feature amount synthesis unit 60 functions as a “matching degree acquisition unit”.

Next, the feature amount synthesis unit 60 determines whether or not each unit region A5 is solder based on the matching degree E of each unit region A5 of the extraction region A4 (step S9: determination step).
Such a determination is made based on whether or not the degree of coincidence E of each unit region is equal to or greater than the threshold by preparing a threshold C of the degree of coincidence in advance. That is, if E is greater than or equal to the threshold value C, solder is used.
Note that the value of the threshold C may be appropriate (determined by a test or the like), but here, for example, C = 0.8.
At this time, the feature amount combining unit 60 functions as a “determination unit”.

As shown in FIG. 10, according to the determination result of the degree of coincidence E of each unit area A5 of the extraction area A4, as shown in FIG. 10, the “binary image” (1 = solder) identified as the solder and the background having the unit area A5 as one unit. , 0 = background) can be obtained (unit area A5 which is solder is shown by shading).
On the other hand, the error detection unit 70 generates a solder pattern outline RA by connecting the centers of the unit areas A5 that are the background and adjacent to the unit area A5 that is the solder (step S10). : Boundary line acquisition step).
At this time, the error detection unit 70 functions as “boundary line acquisition means”.

Then, the error detection unit 70 detects the error D of the generated contour RA with respect to the actual contour R (step S11: error detection step).
The actual contour line R and the generated contour line RA are superimposed and shown in FIG. The error D is calculated based on the error measurement point G registered in step S2. That is, the shortest distance d (pixel distance) from each error measurement point G is calculated for the generated contour line RA. Assuming that the shortest distance obtained by this is d1, d2,..., The error D = max {d1, d2,.
At this time, the error detection unit 70 functions as “error detection means”.

  Next, the error detection unit 70 compares the error D with a preset reference value α (step S12). If the error detection unit 70 exceeds the reference value α, the current synthesis parameter setting identifies the solder as the background. Therefore, it is assumed that an appropriate degree of coincidence can be derived. As described above, the reference value α is an important value for determining the suitability of the synthesis parameter. If this value is set to a large value, the appropriate synthesis parameter can be easily obtained. Accuracy is reduced. If the reference value α is set to a small value, an appropriate synthesis parameter cannot be easily obtained, but the solder pattern identification accuracy is improved. Therefore, it is necessary to set an appropriate value according to the purpose by a test or the like.

On the other hand, if it is determined in step S12 that the error D is equal to or less than the reference value α, it is determined that the current synthesis parameter setting is not appropriate and the synthesis parameter setting is changed (step S13).
That is, the coefficient of coincidence wi with respect to the feature value fi is changed independently. For example, the coefficient of coincidence wi for each feature value fi is changed in 6 steps in increments of 0.2, 0, 0.2, 0.4, 0.6, 0.8, and 1.0 (may be divided more finely). As described above, the entire coefficient wi is normalized. In the present embodiment, the parameters are changed throughout the product space w1 × w2 ×.
When a new synthesis parameter is determined, the process returns to step S8, the degree of coincidence E of each unit area A5 of the extraction area A4 is newly obtained, the unit area A5 to be soldered is obtained, and the contour line is obtained therefrom. A process of generating RA and comparing the error D with the reference value α is retried.
As a result, the process is repeated until a synthesis parameter in which the error D is equal to or less than the reference value α is found.
Further, when the error D becomes equal to or smaller than the reference value α in step S12, combinations of numerical values of the coefficients wi of the synthesis parameter at that time (for example, w6 = 0.5, w8 = 0.5, other w = 0, etc.) (A numerical value group) is recorded as learning data (learning synthesis parameter).
Thereby, the learning data creation process ends.

(Processing when mounting electronic components)
Next, a process performed when the electronic component is mounted will be described.
In a state where the electronic component is sucked by the suction nozzle 106 described above and the head 105 is moved to the vicinity of the substrate mounting position, the CCD camera 108 performs imaging of the substrate mounting position, and acquires captured image data.
Then, the feature amount synthesizing unit 60 cuts out a part of the captured image (for example, the central part) as an extraction region, and executes the same processing as in steps S7 to S10 described above.
That is, the extracted region of the captured image is divided into unit regions, the feature amount for i that is not the coefficient wi = 0 of the matching degree ei is calculated, and the solder of each unit region is determined according to the synthesis parameter acquired in the learning data creation process. The degree of coincidence of is calculated. Then, a contour line is calculated based on the unit region determined to be solder. Since the area surrounded by the contour line is the solder pattern at the mounting position, the head 105 is moved so as to position the suction nozzle 106 at the position surrounded by the contour line, and mounting of the electronic component is executed. .

(Effect of embodiment)
In the image analysis apparatus 10, the feature amount is acquired and synthesized from the pattern region A2 of the solder H in the captured image data by the equation 14 of the co-occurrence matrix, and the background of the solder H identification target from the unit region A5 having the matching feature amount. And a retry pattern is repeated so as to reduce the error from the actual boundary position, thereby specifying a weighting pattern related to the matching degree of the feature amount. For this reason, the solder H is identified from the feature amount based on the unique pattern of the solder H, and the identification target can be effectively identified regardless of the presence or absence of the brightness difference from the background.
For this reason, for example, as shown in FIG. 12, even if the solder H and the substrate K, the wiring pattern N, and the pad U having a different pattern from the solder are present in the imaging region A1, the outline RA indicating the solder shape is accurately obtained. Can be identified.

In addition, since a suitable synthesis parameter related to the 14 equations of the co-occurrence matrix related to the texture pattern is specified, it is possible to eliminate the influence of the bias in the identification accuracy of each method and realize more accurate identification.
Since the 14 feature quantity calculation method of the co-occurrence matrix established as a technique for obtaining the texture pattern feature quantity is used, it becomes possible to more effectively identify the identification target of various patterns.

(Other)
In the image analysis apparatus 10 described above, the method derived from the co-occurrence matrix is used for calculating the feature amount of the solder pattern. However, another method capable of characterizing the pattern may be used, or the method of the co-occurrence matrix. And other methods may be used in combination. For example, as a pattern analysis technique, for example, a frequency, density histogram, run length matrix, and fractal dimension obtained by Fourier transform of an area image may be used.
Even in that case, it is possible to obtain the same effect as the above-described embodiment.

  In the image analysis apparatus 10 described above, only the pattern area A2 is used in the learning data creation process, and the image information of the solder background area is not used, but it is also possible to use it. In that case, for example, the unit area A6 (see FIG. 5) is cut out from the background for each error measurement point, and the fi value is obtained from the unit area A6. Accordingly, by comparing with the fi value of the pattern area A2, for example, the fi values can be divided into those having a large opening and those not having so. As a result, it is not necessary to rotate the entire product space of the coefficient wi serving as a weight, and it is possible to reduce the number of repetitions and quickly acquire an appropriate synthesis parameter.

  In the image analysis device 10 described above, in the learning data creation process, in the retry determination in step 12, the end condition is determined by comparing the error D with the reference value α, but the error D is displayed on the display device 125. And the operator may determine whether to retry. Even in that case, it is possible to obtain the same effect as the above-described embodiment.

  In the image analysis apparatus 10 described above, in the learning data creation process, the retry condition in step 12 is set as the process termination condition when the error D falls below the reference value α. The number of times may be set, and the synthesis parameter that minimizes the error within the specified number may be used as the learning data. Even in this case, it is possible to obtain substantially the same effect as in the above-described embodiment, and it is possible to further speed up the processing.

Further, the error measurement points G are not limited to the six points described above. It may be at least one or more. Even in that case, it is possible to obtain the same effect as the above-described embodiment.
Further, although the shortest distance d between the contour lines RA and R is used as the target of the error D described above, other objects may be used. For example, the number of error measurement points G may be set to be larger, and the areas and centroids may be obtained and compared between the solder regions. Even in that case, it is possible to obtain the same effect as the above-described embodiment.

DESCRIPTION OF SYMBOLS 10 Image analysis apparatus 11 Main control part 20 Image input part 30 Image division part 40 Error measurement point input part 50 Feature quantity extraction part 60 Feature quantity synthetic | combination part 70 Error detection part 100 Electronic component mounting apparatus 120 Control apparatus A1 Captured image A2 Pattern area | region A3 Unit area A4 Extraction area A5 Unit area G Error measurement point (boundary point)
H Solder pattern (identification target)

Claims (4)

An image analysis method performed by an image analysis apparatus that identifies an identification target based on captured image data of a captured image including an identification target having a specific pattern on a surface and a background around the identification target,
A reference feature amount acquisition step for obtaining a feature amount of a texture pattern as a reference by a plurality of methods for a predetermined unit region in a pattern region including only the specific pattern specified in advance in the captured image;
The extracted area including at least the pattern area and the surrounding background in the captured image is divided into unit areas having the same size as the unit area of the pattern area, and the feature amount is determined for each unit area by the respective methods. A comparison feature amount obtaining step to be obtained;
For each unit area of the extraction area, the degree of coincidence with the pattern area is obtained for each feature amount of each method, and the degree of coincidence obtained by weighting the degree of coincidence for each feature amount of each method is obtained. A matching degree acquisition process;
A determination step of determining whether each unit area of the extraction area is an identification target or a background from the synthesized matching value;
A boundary line acquisition step for obtaining a boundary line between the identification target and the background in the extraction region from the plurality of unit regions determined as the identification target in the determination step;
An error detection step of detecting an error between a position on an actual boundary line in the captured image designated in advance and the boundary line acquired in the boundary line acquisition step;
With
Weighting that makes the error equal to or less than a threshold by repeating the matching degree acquisition step, determination step, boundary line acquisition step, and error detection step while sequentially changing the weighting pattern for the matching degree for each feature amount of each method. Identify patterns of
An image analysis method, wherein the identification target is identified from a captured image in accordance with the acquired weighting pattern.   The image analysis method according to claim 1, wherein the plurality of methods for obtaining the feature amount use a part or all of the 14 feature amount calculation methods of the co-occurrence matrix. An image analysis apparatus for identifying the identification target based on captured image data of a captured image including an identification target having a specific pattern on a surface and a background around the identification target,
A reference feature amount obtaining unit for obtaining a feature amount of a texture pattern serving as a reference by a plurality of methods for a predetermined unit region in a pattern region including only the specific pattern specified in advance in the captured image;
The extracted area including at least the pattern area and the surrounding background in the captured image is divided into unit areas having the same size as the unit area of the pattern area, and the feature amount is determined for each unit area by the respective methods. A comparison feature amount obtaining means to be obtained;
For each unit area of the extraction area, the degree of coincidence with the pattern area is obtained for each feature amount of each method, and the degree of coincidence obtained by weighting the degree of coincidence for each feature amount of each method is obtained. A matching degree acquisition means;
A determination means for determining whether each unit area of the extraction area is an identification target or a background from the synthesized matching value;
Boundary line acquisition means for obtaining a boundary line between the identification object and the background in the extraction area from the plurality of unit areas determined as the identification object by the determination means;
Error detection means for detecting an error between a position on an actual boundary line in the captured image designated in advance and the boundary line acquired by the boundary line acquisition means;
With
The matching degree is acquired by the matching degree acquisition unit, the determination is performed by the determination unit, the boundary line is acquired by the boundary line acquisition unit, and the error is detected while sequentially changing the weighting pattern for the matching degree for each feature amount of each method. By repeating the error detection by the means, the weighting pattern in which the error is less than or equal to the threshold value is minimized,
An image analysis apparatus for identifying the identification target from a captured image according to the acquired weighting pattern.   The image analysis apparatus according to claim 3, wherein the plurality of methods for obtaining the feature amount use a part or all of the 14 feature amount calculation methods of the co-occurrence matrix.

Images