JP2005316758A - Method and apparatus for detecting position - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method and an apparatus for detecting a position in which the reduction of a position detection rate due to changes in signal noise and background noise of an image can be prevented and the accuracy of sub-pixel can be obtained. <P>SOLUTION: After performing positional detection at rough accuracy obtained from normalized correlation operation by using a compressed image, a cross spectrum between the frequency spectrum of image data on the position and the frequency spectrum of a reference pattern is calculated and threshold processing is applied to the cross spectrum data to remove signal noise and background noise and detect the position. The number of effective data points is secured by using the linearity of a phase of a frequency transfer function to obtain position detection accuracy in each sub-pixel. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a position detection method for an IC chip or the like in a semiconductor assembly apparatus such as a wire bonding apparatus or a tape bonding apparatus, and an apparatus therefor.

  Conventionally, a reference pattern (teach image) is registered before bonding, and a pattern for searching for the position of the reference pattern in an image (search image) captured by a CCD camera or the like at the time of alignment after the start of bonding Matching is used. A normalized correlation technique is used for pattern matching using grayscale images.

  A great deal of processing time is required to perform a normalized correlation directly between the reference pattern and the search image. For this reason, both the reference pattern and the search image are compressed in the previous stage, and raster scanning is performed with the compressed image with a reduced data amount of the image, and a wide range search is performed by normalized correlation processing (coarse detection). In addition, the position of the compressed search image having the maximum correlation value obtained after the search with the compressed image is detected, and the relative position between the detected search image position and the uncompressed actual size reference pattern is changed. Normalized correlation processing is performed to detect the position of the search image having the maximum correlation value. In the normalized correlation process, a search using a compressed image is performed to shorten the processing time for position detection.

  FIG. 7 is a flowchart showing the flow of conventional position detection processing. As shown in FIG. 7, the reference pattern is obtained by moving the camera onto the IC chip by a manipulator during teaching so that the fixed point of the IC chip is positioned at the intersection of the cross lines on the monitor, The image is converted and stored in the image frame memory of the image processing apparatus. A predetermined area of the image stored in the image frame memory is cut out, and the cut out image data is stored and registered in the memory as a teach image f1 as a reference pattern (step S50). Further, the image data of the teach image f1 is compressed to generate a compressed teach image fc1, which is stored and registered in the memory as the compressed teach image fc1 (step S51). In the compression processing, for example, sampling is performed every n pixels in the vertical and horizontal directions of the image data, and the compressed teach image fc1 is generated from the sampled data. It should be noted that the registration of the teach image at the time of teaching is performed for each location where position detection is performed.

  Next, the position detection process will be described. In the position detection, the video signal of the camera is digitally converted and stored as a search image f2 in the image frame memory of the image processing apparatus (step S52), and sampling is performed every n pixels in the vertical and horizontal directions of the search image f2. The compressed image data fc2 is generated (step S53). Normalized correlation processing is performed between the entire generated compressed image fc2 and the compressed teach image fc1 (step S54), and the position (xc, yc) of the maximum correlation value fc2max (xc, yc) on the compressed image fc2 is detected. (Step S55). Next, normalized correlation processing is performed while changing the relative positions of the search image f2 centered on the position (xc, yc) detected in the compressed image fc2 and the reference pattern f1 (step S56), and the maximum correlation value is obtained. fmax (xd, yd) is obtained and its position (xd, yd) is detected (step S57).

  Note that an embodiment using normalized correlation processing as a position detection method in a wire bonding apparatus is disclosed in Patent Document 1.

JP2002-190021

  In position detection in a wire bonding apparatus or the like, it is necessary to process an image on which background noise or the like is superimposed. However, in the normalized correlation processing, when signal noise other than the position detection target object, background noise, etc. are superimposed on the search image, the accuracy of position detection varies due to the noise, and the wire or lead on the pad of the IC chip In some cases, the bonding position may vary.

  In order to raise the position detection accuracy to a certain level and reduce variations in position detection accuracy, it is necessary to eliminate image noise in the calculation process of the normalized correlation process. As noise processing of an image on which noise, background noise, and the like are superimposed, there are filtering of an input image and bandpass processing using a frequency domain of image data.

  However, the filtering of the input image requires raster scanning using a matrix in order to perform the filtering process directly on the real signal, and requires a great amount of processing time. On the other hand, in order to perform threshold processing such as bandpass on the image data converted into the frequency domain, orthogonal transform such as Fourier transform is applied to the entire search image f2 in step S52 shown in FIG. 7 or the image cut out in step S56. Need to be performed, leading to an increase in processing time. As a result, the balance between processing speed and position detection accuracy is poor, and it has been difficult to construct an efficient algorithm. For this reason, it is a problem to improve detection accuracy by removing noise contained in an image at an early stage of the processing process to improve processing speed and removing noise.

  Therefore, the present invention has been made in view of the above problems, and prevents a decrease in position detection rate due to a change in image signal noise and background noise to obtain subpixel accuracy, as well as an IC chip or the like. An object of the present invention is to provide a position detection method and apparatus capable of performing position detection processing in a short time.

  The position detection method according to the present invention includes a matching step for detecting a position of image data that most closely matches the reference pattern by matching the reference pattern and image data from the imaging means, and a reference centered on the position detected in the matching step. A cutout step of cutting out an area having the same size as a pattern from the image data, a crossspectrum calculation step of calculating a crossspectrum between the image data cut out in the cutout step and a reference pattern, and a calculation in the crossspectrum calculation step The cross spectrum amplitude data is compared with a preset threshold value, a frequency removal step for removing the frequency of the amplitude data below the threshold value, and the cross spectrum excluding the frequency removed in the frequency removal step is subjected to inverse Fourier transform, Inverse Fourier transformed data A peak value detecting step for detecting a position up to the peak value of the data, and a phase shift amount calculating step for calculating a phase shift amount between the image data at the position detected in the previous peak value detecting step and the reference pattern. Features.

  The matching step of the position detection method according to the present invention is a normalized correlation process between an image compressed reference pattern and image compressed image data from an imaging means.

  The phase shift amount calculating step of the position detection method according to the present invention includes a transfer function between a reference pattern frequency spectrum obtained by Fourier transforming a reference pattern and an image data frequency spectrum obtained by performing Fourier transform on the image data cut out at the cutout step. The phase shift amount is calculated from the phase characteristics of

  In addition, the position detection device according to the present invention has a matching unit that detects the position of image data that most closely matches the reference pattern by matching the reference pattern and image data from the imaging unit, and the position detected by the matching unit. Cutting means for cutting out an area having the same size as the reference pattern from the image data, cross spectrum calculating means for calculating a cross spectrum between the image data cut out by the cutting means and the reference pattern, and the cross spectrum calculating means The amplitude data of the cross spectrum calculated in step 1 is compared with a preset threshold value, a frequency removing unit that removes the frequency of the amplitude data below the threshold value, and an inverse Fourier transform of the cross spectrum excluding the frequency removed by the frequency removing unit. The position to the peak value of the inverse Fourier transformed data A peak value detecting means for output, characterized in that a phase shift amount calculation means for calculating a phase shift amount between the image data and the reference pattern position detected by year peak value detecting means.

  The matching means of the position detection apparatus according to the present invention is a normalized correlation process between an image compressed reference pattern and image data from an image compressed image pickup means.

  Further, the phase shift amount calculation means of the position detection apparatus according to the present invention includes a transfer function between a reference pattern frequency spectrum obtained by Fourier transforming a reference pattern and an image data frequency spectrum obtained by Fourier transforming the image data cut out by the cutout means. The phase shift amount is calculated from the phase characteristics of

  According to the position detection method and the apparatus according to the present invention, after performing position detection with coarse accuracy obtained from the normalized correlation calculation using the compressed image, the cross spectrum data between the image data and the reference pattern is detected. Threshold processing is performed, and signal noise and background noise are removed to perform position detection. Thereby, it is possible to prevent a decrease in position detection rate and position detection accuracy due to the influence of noise.

  Further, according to the position detection method and apparatus according to the present invention, by utilizing the linearity of the phase of the frequency transfer function, an effective number of data points can be secured, so that the position accuracy accuracy in units of subpixels can be obtained.

  Further, according to the position detection method and apparatus according to the present invention, the normalized correlation process is not required while changing the relative positions of the search image and the teach image after the coarse detection by the normalized correlation process. The detection process can be performed in a short time.

  Embodiments of a position detecting method and apparatus in a semiconductor assembly apparatus according to the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a wire bonding apparatus using a position detection device according to the present invention, FIG. 2 is a block diagram showing a configuration of a position detection device using a position detection method according to the present invention, and FIG. FIG. 4 is a flowchart for explaining a position detection method according to the present invention in a wire bonding apparatus.

  As shown in FIG. 1, a wire bonding apparatus 1 includes a camera 3 as an imaging means, a bonding arm 4 having a capillary 4a attached to the tip, a bonding head 5 that drives the bonding arm 4 up and down, and a bonding arm 4 And an XY table 6 as a positioning means for mounting the bonding means and the imaging means comprising the bonding head 5 and positioning them by two-dimensionally moving them in the X and Y directions, and a bonding stage on which a bonding operation is performed by the bonding head 5 The bonding stage 8 includes a heater (not shown) that heats a substrate, a lead frame, and the like on which a plurality of chips are arranged and adhered in the longitudinal direction.

  Further, as shown in FIG. 1, a wire bonding apparatus 1 as a semiconductor assembly apparatus receives position signals as position detection means that receives an image pickup signal from a camera 3 and detects a fixed point deviation amount of a part to be bonded from image data. A signal for moving the XY table 6 manually is output to the control device 12, a monitor 11 that receives video output from the position detection device 10, a control device 12 having a microprocessor 12 a and a memory 12 b. A manipulator 15 and a driving device 13 that issues a driving signal to the bonding head 5 and the XY table 6 in response to a command signal from the control device 12 are provided. The memory 12b of the control device 12 stores a program for the microprocessor 12a for controlling the wire bonding apparatus 1, a position coordinate table that stores position coordinates of bonding points, and the like.

  Next, the position detection apparatus 10 used for the wire bonding apparatus 1 having the above configuration will be described with reference to FIG. As shown in FIG. 2, the position detection device 10 includes an FPGA (programmable array) 10a including a CPU, an arithmetic circuit, and an interface circuit, a data store memory 10b for storing a teach image as a reference pattern, and the like. A work memory 10d for temporarily storing data in calculation and control processing; a program memory 10c for storing a program for image processing executed by the CPU; an image frame memory 10e for storing image data from the camera 3; A camera driver 10f that outputs a control signal to the camera 3, an A / D converter 10g that converts the signal from the camera 3 into a digital signal, a monitor driver 10h that outputs a video signal to the liquid crystal monitor 11 and the like, and an external The communication driver 10j for communicating with the position detection device 10 and the position detection device 10 are connected to the PCI bus. It has a PCI bridge 10k for.

  The camera 3 captures an image using a signal such as a synchronization signal output from the camera driver 10f of the position detection device 10 or a trigger signal for starting imaging, and outputs a video signal to the A / D converter 10g of the position detection device 10. To do. The position detection device 10 samples the digital value digitally converted by the A / D converter 10g with a clock having a predetermined frequency, and stores the image data in the image frame memory 10e. The CPU and the arithmetic circuit built in the FPGA 10a perform processing such as normalized correlation processing, Fourier transform, and inverse Fourier transform described below. Further, the CPU and the arithmetic circuit built in the FPGA are configured to be able to process arithmetic operations and the like in parallel.

  Next, position detection processing in the wire bonding apparatus 1 and the position detection apparatus 10 having the above-described configuration will be described with reference to the flowcharts shown in FIGS. The position detection process is performed by the CPU built in the FPGA 10a of the position detection apparatus 10 executing a processing program stored in the program memory 10c.

  As shown in FIG. 3, first, a teach image is registered as a reference pattern used for position detection. To register the teach image, first, the camera 3 is moved onto the IC chip by a manipulator during teaching so that the fixed point of the IC chip is positioned at the intersection of the cross lines 11a on the monitor 11 shown in FIG. 3 is digitally converted by the A / D converter 10g and stored in the image frame memory 10e of the position detecting device 10 (step S1). Further, the control device 12 stores the position of the XY table 6 at this time as a reference coordinate in the position coordinate table of the memory 12b.

  Next, the CPU cuts out the image data in the window 11b displayed on the monitor 11 shown in FIG. 1 from the image frame memory 10e, and stores the cut-out image data as a teach image f1 (x, y) (step). S2). Note that the teach image f1 (x, y) is composed of a two-dimensional pixel array. Next, a Fourier transform process is performed on the teach image f1, and a frequency spectrum F1 (u, v) is created and stored in the data store memory 10b (step S3). The Fourier transform process uses a fast Fourier transform (FFT) obtained by speeding up discrete Fourier transform. The frequency spectrum F1 (u, v) generated from the teach image f1 (x, y) by the fast Fourier transform is composed of a two-dimensional matrix. Note that (u, v) represents each frequency variable in the (x, y) direction. Further, the image of the teach image f1 (x, y) is compressed, and the data of the compressed image is stored in the data store memory 10b as the compressed image fc1 (x, y) of the teach image (step S4). In the compression process, sampling is performed every n pixels in the vertical and horizontal directions of the image data, and a compressed image is generated from the sampled data.

  With the above processing, each image data of the teach image f1 (x, y), frequency spectrum F1 (u, v), and compressed image fc1 (x, y) of the reference pattern is stored in the data store memory 10b.

  Next, position detection processing will be described using the flowchart shown in FIG. First, the control device 12 reads the reference coordinates from the position coordinate table of the memory 12b, and moves the camera 3 mounted on the XY table 6 as the positioning means to the reference position. After the movement of the XY table 6 is stopped, the position detection device 10 captures image data in the vicinity of a fixed point on the IC chip from the camera 3 into the image frame memory 10e (hereinafter, this image data is referred to as a search image f2) ( In step S10), a search compressed image fc2 (x, y) is generated from the search image f2 (x, y) stored in the image frame memory 10e (step S11). Note that the search compressed image fc2 (x, y) is generated by performing the same compression process as in the teaching.

  Next, the compressed image fc1 (x, y) of the teach image stored in the data store memory 10b is read, and the compressed image fc1 (x, y) of the teach image and the compressed image fc2 (x, y) of the search image are read. ) Is performed (step S12). The entire compressed image fc2 (x, y) of the search image is subjected to raster scanning by the normalized correlation process, and the maximum correlation value fcmax (xc, yc) is detected (step S13). At this time, it is checked that the maximum correlation value fcmax (xc, yc) obtained by the normalized correlation processing of the compressed image is a value equal to or larger than the reference value M set in advance (step S14). The search image f2 (xc, yc) having the maximum correlation value is used in the subsequent processing. The reference value M is used to determine whether the search image in the normalized correlation process includes an image similar to the teach image. When the maximum correlation value is less than the reference value M, it is determined that the search image does not include an image similar to the teach image, an error warning is output, and the subsequent processing is not performed. When the maximum correlation value fcmax (xc, yc) is a value greater than or equal to the reference value M, the position (xc, yc) of the maximum correlation value fcmax (xc, yc) is set as the detection position (step S15). Note that the detected position (xc, yc) of the maximum correlation value fcmax (xc, yc) from the origin of the XY axis of the search image f2 is set as the deviation Lc of the normalized correlation process.

  Next, from the search image f2 (x, y) in the image frame memory 10e to the teach image f1 (x, y) centered on the position (xc, yc) of the maximum correlation value fcmax (xc, yc) obtained in step S15. An image f3 (x, y) having the same area is cut out (step S16). The extracted image f3 (x, y) is subjected to Fourier transform to generate a frequency spectrum F3 (u, v) (step S17).

Next, two image data of the frequency spectrum F1 (u, v) of the teach image f1 (x, y) and the frequency spectrum F3 (u, v) of the image f3 (x, y), which have been transformed into the frequency domain in advance, are obtained. On the other hand, the cross spectrum F4 (u, v) is obtained (step S18). The cross spectrum F4 (u, v) is obtained by the equation (1).
F4 (u, v) = F1 * (u, v) .F3 (u, v) (1)
Here, F1 * (u, v) represents a conjugate complex number of F1 (u, v). Further, (u, v) represents each frequency variable in the (x, y) direction.
The cross spectrum F4 (u, v) shown in the equation (1) forms a complex matrix, and each element of the matrix includes a real part and an imaginary part.

  Next, amplitude data | F4 (u, v) | of the amplitude component of the cross spectrum F4 (u, v) is obtained (step S19). When the amplitude data | F4 (u, v) | is FP, threshold processing is performed on the amplitude data FP of each frequency, and both the real part and imaginary part of the frequency having the amplitude data FP less than the threshold TL are 0. The cross spectrum F4TL (u, v) is calculated (step S20). That is, when FP <TL, F4TL (u, v) = 0, and when FP ≧ TL, data of F4TL (u, v) is held.

The cross spectrum F4 _TL (u, v) subjected to the threshold processing is subjected to inverse Fourier transform, and the cross-correlation function f4 (x, y) of the real signal regions x and y is calculated again (step S21). The position data (xs, ys) of the maximum value f4max (xs, ys) of the cross-correlation function f4 (x, y) in the region is detected (step S22). Note that the position (xs, ys) having the maximum correlation value obtained in step S22 has the accuracy of the sampling interval, that is, the accuracy in units of pixels.

FIG. 5B shows a cross-correlation function by performing threshold processing of the cross spectrum F4 (u, v) in step S20 and performing inverse Fourier transform on the cross spectrum F4 _TL (u, v) subjected to threshold processing. It is a graph which shows an example of the result of having calculated. In FIG. 5A, the cross-correlation function is calculated by performing inverse Fourier transform on the cross spectrum F4 (u, v) without performing the threshold processing of the cross spectrum F4 (u, v) in step S20. It is a graph which shows an example of the result obtained. FIG. 5 shows the cross-correlation function in the X-axis direction of the real signal region, the horizontal axis shows the image range composed of pixels, and the vertical axis shows the cross-correlation value.

  The detected image appears as an image in which the reflected light varies due to noise or changes in the surface state of the IC chip, and the luminance level changes due to the protrusion of the adhesive, and the cross spectrum F4 (u, v) is shown in FIG. As shown, unnecessary frequency components are included. For this reason, threshold processing is performed to remove unnecessary frequency components including noise and the like.

  In addition, an image obtained by operating the focus of the camera 3 and each illumination system during teaching is generally in a state where the user can easily determine the position detection target, that is, in a state where the outline is clear. When spectrum analysis is performed on this image, the frequency region including the contour information in the amplitude data has a large amplitude, and conversely, the frequency region that adversely affects the position detection accuracy such as background noise has a small amplitude. Therefore, by performing threshold processing, noise is removed as shown in FIG. 5B, and the peak detection value can be easily detected by performing inverse Fourier transform on the cross spectrum from which noise has been removed. . In addition, this finally affects the position accuracy, and it becomes possible to detect the position with higher accuracy and resistance to noise.

  In addition, although the parabolic approximation and the sinc function approximation are often used for the peak detection in the data after the cross spectrum and the inverse Fourier transform, the detection accuracy and processing time required for the wire bonding apparatus used in the production process cannot be satisfied. . Therefore, an algorithm that can secure an effective number of data points for obtaining sufficient detection accuracy and has a short processing time is required. For this reason, in the present invention, the movement amount obtained from the data after the inverse Fourier transform is used, and this is used as the phase shift amount in the frequency domain and multiplied by the data before the inverse Fourier transform to within one pixel. Detection is performed. This achieves good subpixel accuracy.

Hereinafter, detection of sub-pixel accuracy for position detection will be described. Teach image f1 (x,
The amount of displacement of the search image f2 (x, y) with respect to y) is defined as L (xl, yl). The accuracy of the position detected in step S22 shown in FIG. 4 is the sampling interval accuracy, that is, the pixel. When the positional deviation amount at this time is Ls (xs, ys) and the positional deviation amount within one sampling interval (less than one pixel) is Lr (xr, yr), L is L = Ls + Lr.

  Note that Ls is a distance to the maximum value f4max (xs, ys) of the cross-correlation function f4 in step S22. The present invention further achieves high accuracy by obtaining Lr.

  The frequency spectrum F1 (u, v) obtained by transforming the teach image f1 in step S3 shown in FIG. 3 into the frequency domain by FFT, and the frequency spectrum F3 (u) obtained by performing Fourier transform on the image f3 cut out in step S17 shown in FIG. , V) to calculate a transfer function G (u, v) (step S23). When the transfer function is G (u, v), the relationship between the input F1 (u, v) and the output F3 (u, v) is expressed by equation (2).

F3 (u, v) = G (u, v) · F1 (u, v) (2)
The expression (2) includes the shift amount (movement amount) Ls + Lr. By moving the expression (2) by −Ls, an expression including only the shift amount (movement amount) Lr below the decimal point of the pixel is obtained. It is done. Note that the phase shift amount in the frequency region of the movement amount (-Ls) is ej (u, v) Ls (step S24).
By correcting the equation (2) with the phase shift amount (ej (u, v) Ls), the equation (3) can be obtained.
ej (u, v) Ls.F3 (u, v) = ej (u, v) Ls.G (u, v) .F1 (u, v) (3)
If ej (u, v) Ls · G (u, v) on the right side of equation (3) is the transfer function Q (u, v), then Q (u, v) = ej (u, v) Ls · G ( u, v). Phase data for the frequencies u and v is obtained from the phase characteristics of the transfer function Q (u, v) (step S25).

  FIG. 6 shows the phase characteristics of Qxy (u, v). The vertical axis represents the phase, and the horizontal axis represents the frequency u or v. When the transfer function Q (u, v) includes a component of positional deviation of one pixel or more, a saw waveform is shown, but Q (u, v) represents G (u, v) as ej (u, v). ) Since it is corrected by Ls, linearity is shown as shown in FIG. The slope of the straight line is obtained from the data indicating the linearity of the phase with respect to the frequency u or v by the least square method, and the obtained slope value is set as a shift amount Lr after the decimal point of the pixel (step S26). By obtaining the slope using the least square method, it is possible to obtain a positional deviation amount L (xl, yl) with good subpixel accuracy (accuracy of less than one pixel).

  As described above, the detection position (deviation amount) of the fixed point on the IC chip is detected as the deviation amount Lc + Ls + Lr in the normalized correlation process.

  As described above, according to the position detection method and apparatus according to the present invention, after performing position detection with coarse accuracy obtained from the normalized correlation calculation using the compressed image, the image data and the reference pattern Since the threshold is set for the cross spectrum data and the position detection is performed by removing the signal noise and the background noise, it is possible to prevent the position detection rate and the position detection accuracy from being lowered due to the influence of the noise.

  Further, according to the position detection method and apparatus according to the present invention, by utilizing the linearity of the phase of the frequency transfer function, an effective number of data points can be secured, so that the position accuracy accuracy in units of subpixels can be obtained.

  In addition, the position detection device according to the present invention can perform the position detection process in a short time because the CPU and the arithmetic circuit built in the FPGA can process the arithmetic operation and the like in parallel.

It is a block diagram which shows the structure of the wire bonding apparatus using the position detection apparatus by this invention. It is a block diagram which shows the structure of the position detection apparatus using the position detection method by this invention. It is a flowchart which shows the registration procedure of the reference pattern used at the time of position detection. It is a flowchart for demonstrating the position detection method by this invention in a wire bonding apparatus. (A) is a graph showing an example of a result of calculating a cross-correlation function by performing inverse Fourier transform on the cross spectrum without performing threshold processing of the cross spectrum F4 (u, v), and (b) is a cross It is a graph which shows an example of the result of having performed the threshold process of spectrum F4 (u, v), performing the inverse Fourier transform to the cross spectrum which performed the threshold process, and calculating the cross correlation function. It is a figure which shows the phase characteristic of a transfer function Qxy (u, v). It is a flowchart which shows the flow of the calculation processing of the conventional positioning detection.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Wire bonding apparatus 3 Camera 4 Bonding arm 4a Capillary 5 Bonding head 6 XY table 7 Conveyance apparatus 8 Bonding stage 10 Position detection apparatus 10a FPGA
10b Data store memory 10c Program memory 10d Work memory 10e Image frame memory 10f Camera driver 10g A / D converter 10h Monitor driver 10j Communication driver 10k PCI bridge 11 Monitor 11a Cross line 11b Window 12 Controller 12a Microprocessor 12b Memory 13 Drive device 15 Manipulator

Claims (6)

  A matching step for detecting the position of the image data that most closely matches the reference pattern by matching the reference pattern with the image data from the imaging means, and an area of the same size as the reference pattern with the position detected in the matching step as a center. Cutting out from the image data, a cross spectrum calculating step for calculating a cross spectrum between the image data cut out in the cutting step and a reference pattern, and amplitude data of the cross spectrum calculated in the cross spectrum calculating step in advance Compared with the set threshold value, the frequency removal step for removing the frequency of the amplitude data below the threshold value, and the peak value of the data obtained by performing the inverse Fourier transform on the cross spectrum excluding the frequency removed in the frequency removal step and performing the inverse Fourier transform Up to Position detecting method, wherein a peak value detecting step, further comprising a phase shift amount calculation step of calculating a phase shift amount between the image data and the reference pattern position detected by year peak value detecting step of.   2. The position detection method according to claim 1, wherein the matching step is a normalized correlation process between the image-compressed reference pattern and the image data from the image-compressed image pickup means.   In the phase shift amount calculating step, the phase shift amount is calculated from the phase characteristic of the transfer function between the reference pattern frequency spectrum obtained by Fourier transforming the reference pattern and the image data frequency spectrum obtained by Fourier transforming the image data cut out in the extraction step. The position detection method according to claim 1, wherein:   Matching means for detecting the position of image data that most closely matches the reference pattern by matching the reference pattern with image data from the imaging means, and an area of the same size as the reference pattern with the position detected by the matching means as the center Cutting means for cutting out from the image data, cross spectrum calculating means for calculating a cross spectrum between the image data cut out by the cutting means and a reference pattern, and amplitude data of the cross spectrum calculated by the cross spectrum calculating means in advance Compared with the set threshold value, the frequency removal means for removing the frequency of the amplitude data below the threshold value, and the peak value of the data obtained by performing the inverse Fourier transform on the cross spectrum excluding the frequency removed by the frequency removal means and performing the inverse Fourier transform Peak value detection means to detect the position up to Position detecting device characterized by comprising a phase shift amount calculation means for calculating a phase shift amount between the image data and the reference pattern of position detected by the click value detecting means.   5. The position detecting apparatus according to claim 4, wherein the matching means is a normalized correlation process between the image compressed reference pattern and the image data from the image compressed image pickup means.   The phase shift amount calculation means calculates the phase shift amount from the phase characteristic of the transfer function between the reference pattern frequency spectrum obtained by Fourier transforming the reference pattern and the image data frequency spectrum obtained by Fourier transform of the image data cut out by the extraction means. The position detecting device according to claim 4, wherein

Images