JP2834153B2 - Adaptive learning type general purpose image measurement device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION Object of the Invention; (Industrial Application Field) The present invention is particularly applicable to various fields such as sensor and identification of parts in a production line, image measurement (counting) in materials and medical fields, and the like. A device for various image measurement (including counting and identification) relating to the shape and number of objects captured as a binary image. In particular, by learning, two-dimensional patterns can be identified or two-dimensional patterns at high speed according to the target application. More particularly, the present invention relates to a general-purpose image measurement device for counting and weighing geometric feature amounts (number, perimeter, etc.) of a pattern at high speed.

(Prior art) Conventionally, as this type of image measuring device, image processing methods considered to be necessary to obtain a desired result for a predetermined purpose are realized by sequentially serially combining as shown in FIG. (Sequential and procedural). For example, in a binary image including many particles of two different sizes with different radii, when counting the number of each particle, first find all the particles (black circles) on the screen (white background), Label each particle with a label (number) to distinguish, measure the diameter and area of each particle, determine which type of particle,
The number of determinations is integrated to finally determine the number of the two types of particles.

In such a conventional apparatus, since a number of complicated image processing methods are realized in combination one after another, the measurement time for the processing is enormous, the apparatus is inevitably large, and is not suitable for high-speed processing. . Moreover, a decisive problem of the measuring device is that as the complexity of the corresponding image (the number of particles in the above example) increases, the processing time generally increases in proportion thereto. Another problem is that since such a series of processes is a special process considered for a predetermined purpose, such a device becomes a single-purpose device only for the predetermined purpose, That is, it is necessary to change the processing method for each application. Furthermore, there is a problem that can be applied only to a problem in which what and what procedure should be measured is clearly understood.

(Problem to be Solved by the Invention) In contrast to the above-described serial measurement method, a parallel / adaptive method such as a perceptron or a neural network model in pattern recognition has been considered. 25th
The figure shows the structure of the neural network, which is based on the information processing of the brain, and is a layer (input layer, intermediate layer, output layer) in which many elements that model the characteristics of neurons are arranged in parallel. ) Are connected in multiple stages, their coupling coefficients are set as variable weights (parameters), and a large number of data for learning are sequentially input so that a desired output (measurement result) is obtained with respect to the input. It is based on a learning method in which the coupling coefficient is successively corrected based on the error of the output each time.

However, in this measurement method, although the processing time is fast due to the parallel processing, the convergence speed of the learning is very slow, it takes time, and sometimes a good result cannot be obtained due to the local optimal solution. This measurement method is versatile because it is an adaptive method, but it is clear how many elements and how to connect to the problem and what to take at the input. However, there is a drawback that must be done by trial and error. Further, in the element used for the measurement method using the neural network, the values of input and output signals are limited to two values of “0” and “1” or a value between them as a model of a nerve cell. Therefore, a large number of input signals are multiplied by weights, added, and subjected to non-linear conversion to become one output.

FIG. 26 shows such a measurement method, which is composed of weighted sum and nonlinear conversion neuron elements 10 and 11. However, the expression form of information in such a measurement method is meaningful as a model of neural information processing, but is not so important for practical application, and is rather limited and inefficient. Therefore, the actual situation is that the above-mentioned measuring methods are hardly applied as practical image measuring devices.

On the other hand, there is a system disclosed in Japanese Patent Publication No. 58-47064 as a system for identifying a two-dimensional pattern at a high speed or measuring a geometric feature of the pattern at a high speed.
In this method, the operation required for character identification is performed in advance by another machine, and the result must be obtained. In addition, the main purpose is to limit the application to character reading and to have the feature that character extraction is not necessary, and it is difficult to handle the size and shape of the target pattern. Both adaptability and versatility were poor for use in applications tailored to field targets. Furthermore, it is difficult to predict in advance the environment (disturbance, for example, fading, image quality, etc.) at each use site, so that an optimal and adaptive method suitable for each situation cannot be taken.

The present invention has been made in view of the above circumstances, and an object of the present invention is to perform a two-stage feature extraction having a relatively simple structure based on a theoretical consideration of basic conditions of image measurement. Another object of the present invention is to provide an adaptive learning-type general-purpose image measurement device capable of adaptively performing image measurement for use that is adapted to an arbitrary object or environment by faster learning.

(Means for Solving the Problems) The present invention relates to an adaptive learning-type general-purpose image measuring apparatus, and an object of the present invention is to specify an image pickup means for two-dimensionally picking up an object to be measured and a measuring range. Window signal generating means for generating a window signal in synchronization with the output of the video signal of the imaging means,
A plurality of sets of autocorrelation mask patterns having different sizes for taking the Nth order autocorrelation, switching a cutout image size according to the size of the selected autocorrelation mask pattern, and outputting an image output by the imaging means An image extracting means for extracting a plurality of sets of local images in a time-series manner within the range of the window signal from among the signals; and a local image from the image extracting means and the selected autocorrelation mask pattern. A correlation calculating means for calculating an Nth order autocorrelation coefficient value based on the Nth order autocorrelation coefficient value calculated by the correlation calculating means together with the teaching data set by the teaching means for teaching a class in a learning mode. Storage means for storing each time the window signal is generated; and a plurality of N-order autocorrelation coefficient values and teaching data stored in the storage means, which are analyzed by multiple regression analysis to obtain the measurement object. A statistical feature calculating means for calculating a statistical feature value for recognizing the class, and calculating a coefficient for obtaining the statistical feature value by linearly integrating the Nth order autocorrelation coefficient value; In this case, the Nth-order autocorrelation coefficient value calculated for the unknown measurement target by the correlation calculation means is linearly integrated using the coefficient and the statistical characteristic value, and the unknown measurement target is And recognition means for automatically recognizing the class.

(Operation) Many problems in practical image measurement include measuring the area and perimeter of a single object captured in an image frame (on the screen), counting the number of holes, and measuring the number of holes on the screen. The problem of counting the number of the same objects can be considered. Also, the problem of recognition for determining what an object is by measurement is considered to be a problem of measurement (shape measurement) in a broad sense here. Measurement results for such various image measurement problems are generally given as measurement values (numerical values), and therefore, image measurement can be considered as a problem of feature extraction from an image. The following three items are important as the basic conditions.

1) Invariance related to translation 2) Additivity regarding screen 3) Adaptive learning possibility Condition 1) is that the measurement value (for example, area) does not change at the same value regardless of where the object is in the image frame. Means. Condition 2) is, for example, that, when the entire screen is divided into a plurality of parts, the measured value on the entire screen is equal to the measured value on the divided screen, as is clear from the case of counting particles on the screen. Means sum. Furthermore, condition 3) is
This is important as a condition for versatility that is the object of the present invention.

To satisfy the above basic conditions 1) to 3), the following F1
And F2 feature extraction in two stages.

F1: A large number of general basic features satisfying the conditions 1) and 2) are extracted as initial features for all the screens (geometric feature extraction).

F2: The optimal features for various applications are adaptively extracted by learning by linearly combining the initial features (statistical feature extraction).

For example, as the geometric feature extraction of F1, feature extraction using a local autocorrelation mask based on the concept of N-order autocorrelation (see Japanese Patent Publication No. 58-47064) can be used.
Further, the statistical feature extraction of F2 can be realized using a method such as linear regression analysis in multivariate analysis. The linear combination in the statistical feature extraction F2 is necessary to maintain the condition 2).

The measurement concept of the present invention as shown in FIG. 2 based on the two-stage feature extraction of F1 and F2 satisfies the above-mentioned basic conditions 1) to 3) of image measurement, and has a simple structure and is easy to implement as a parallel device. It becomes a dynamic and adaptive measurement method, and image measurement can be adaptively performed by high-speed learning for various applications.

The measuring apparatus of the present invention performs image measurement of a two-dimensional pattern in a general-purpose and high-speed real-time manner based on the above-described operation.

Embodiment First, the measurement principle of the present invention will be described with reference to the drawings.

As shown in FIG. 3, the measurement object is a TV camera (for example,
The image is captured as a grayscale (multi-valued) image in the image frame through the CCD 1. This measurement method can be applied to gray-scale images (signal SG1) as it is. However, in actual image measurement, binary images are mainly handled mainly for the shape of the measurement target. Consider a binary image binarized by the binarization circuit 2. In this case, the grayscale image is binarized by a binarization circuit 2 into a target area (value 1: black background) and a background area (value 0: white background) at a predetermined threshold, and an image memory (M × M pixels)
3 is input.

An image f ( r ) (here, r = (i, j)) captured in an image frame (in the image memory 3, hereinafter simply referred to as a screen)
Is extracted by the geometric feature extraction unit 4 for the two-dimensional vector representing the position on the screen. Specifically, the extracted feature is an N-order autocorrelation function (see the Electronic Technology Research Institute, Research Report No. 818, issued in July 1981). In addition,
Hereinafter, the underline indicates a vector.

The extraction of the initial feature is based on the idea of the above equation (1), and shows a position invariant, that is, a feature that is invariant with respect to any translation f ( r ) → f ( r + b ) within the target image frame. I have. However, the displacement direction (a _1, ... _{a N)} how to take are countless idea, a practical point of view, for example, from N to up to the second order, and the reference point in the displacement direction in order to satisfy the condition 2)
What is necessary is just to set it to the local 3x3 direction around r . FIG. 4 shows an example of the local displacement direction. In this case, the Nth-order autocorrelation function of the equation (1) is calculated by scanning the entire screen once with a 3 × 3 local frame (center r 1 ). For binary images,
This apparently is _{f (r) f (r +} a 1) ... f (r +
a _N ) = 1, that is, counting the number of local patterns in the entire screen where all (N + 1) pixels in a certain function on a 3 × 3 local frame are “1”. Become. Such a local pattern of (N + 1) pixel points exists in the number of combinations for selecting (N + 1) points from 3 × 3 = 9 ways (N = 0, 1, 2 for up to second order) when simply considered. However, in consideration of scanning, patterns that move by parallel movement within the local frame have the same value, and after all, there are 25 independent patterns up to the second order shown in FIG. In the pattern of FIG. 5, "1" indicates a pixel to be examined, and "*" indicates a pixel to be ignored. These patterns are called local autocorrelation masks. Once scanning the entire screen by using these mask M _{j (j} = 1~25), by obtaining the respective product-sum, local autocorrelation coefficients x _j of the 25 types corresponding to M _j as the initial feature The procedure of the calculation is shown in steps S1 to S3 in FIG. The details of the calculation circuit system are described in JP-B-58-47064.

The initial feature x _j from the geometric feature extraction unit 4 is input to the statistical feature extraction unit 5, and next, the statistical feature extraction (F2) as the adaptive feature extraction will be described.

The initial features x _j obtained as described above are general and basic features that do not depend on the measurement task. However, information on the target image necessary for the measurement task is extracted as a whole. Therefore, these linear combinations are constructed in a configuration as shown in FIG. Thus, y _i as a new feature (measured value itself or predicted approximate value) effective for the measurement task is obtained. here,
M represents the number of measurement values measured simultaneously. The initial feature x _j and the new feature y _i are respectively represented by a vertical vector x = (x ₁ ,.
₂₅ ) ′ and y = (y ₁ ,..., Y _n ) ′.
= [A _ij ] ′, the above equation (2) can be simply expressed as y = A ′ x (2 ′) (where “′” indicates transposition). Here, the coefficient A may be found trivially depending on the problem. For example, the area of the measurement object captured in the binary image is clearly obtained as y ₁ = x ₁ because the zero-order autocorrelation coefficient (mask M ₁ ) counts the number of pixels of “1”. . In addition, other measurements may be represented by linear combinations due to logical considerations. However, such artificial methods are difficult and uncommon. Therefore, in the present invention, it is considered that the optimum coefficient A is automatically obtained by learning as a general-purpose method, and for this purpose, a multivariate analysis method as shown in FIG. 8 is used.

This is an effective and direct method for multiple regression analysis, one of the multivariate analysis methods, especially for measurement tasks. Now, a set of measurement target images used for learning is represented by G = f _i (r), where i = 1 to
As L), a correct measured vector for the image f ₁ z _i
And the initial feature vector calculated in step S10 is x _i (the output is y _i = A ′ x ₁ ).
At this time, the optimum coefficient A is obtained by the following average 2 by multiple regression analysis.
By the squared error Is minimized in step S12, and the optimal solution is explicitly given by the following equation (4).

A = R _xx ^-1 R _xz (4) where Are the autocorrelation matrix of the initial feature x _i and the input data, respectively.
a cross-correlation matrix of x _i and z _i. The -1 power indicates an inverse matrix.

Using the coefficient A obtained in steps S11 and S12 by the learning as described above, the system shown in FIG. 3 performs a good estimation of the measurement vector with respect to the input image f ( r ) according to the equation (2 '). Can be. If the correct measurement vector z
_{If i} can be written in the form of the above equation (2), the correct coefficient A is automatically obtained by learning.

The recognition of the measurement object is also possible. When the multiple regression analysis, consider distinguishing code of the recognition target class measurement vector (z _i and y), may be performed similar learning. For example, in the case of recognizing an object composed of two classes, in the learning mode, the coefficient A is learned with z = (1, 0) 'for class 1 and z = (0, 1)' for class 2. Then, in the recognition mode, the recognition unit 6 compares the value of the feature y ₁ and y ₂ for obtained for the input target image _{y = (y 1, y 2} ) '( step S22), and is characterized y _i Recognition is performed by discriminating class 1 if it is large, and class 2 if it is not. The same applies to the case of the many (K) class. As described above, in the case of recognition, identification processing such as detection of the maximum value of the feature y _i is required.

Also, discriminant analysis can be used as learning of the coefficient A for recognition. In this case, the coefficient A is obtained as a solution (eigenvector) of the next eigenvalue problem so that the vector y optimally separates the K class as a space.

To _{X B A = X W AΛ .........} (6) (A'X W A = I) where, lambda is the eigenvalue diagonal matrix, I is the identity matrix. X _W and X _B are in the class of the initial feature vector x , respectively.
This is an inter-class covariance matrix and is defined by the following equation.

 Where ω_j, _j, X_jAre the occurrence probabilities of class j, respectively.
Mean vector, covariance matrix, _TIs the overall mean vector
It is. As an identification method in the case of this discriminant analysis, for example,
Input (y) To the average vector of each class ( _j= A '
_j) To calculate the distance to the class that gives the minimum distance
A method for identifying the input is conceivable.

The above measurement principle shows that Euler's formula in topological mathematics can be almost approximated by the model.
Importantly, Euler's formula (as a program)
Rather than teaching, this system automatically learned from the learning examples.

An image measuring apparatus according to the present invention is based on the above-described measuring principle, and one embodiment thereof will be described with reference to FIG.

The industrial television camera 101 captures an image of the measurement target 100 and sequentially outputs a two-dimensional image IS in a time-series manner.
102 and the output image synthesis switching circuit 154. The sampling circuit 102 sequentially converts the video image IS into digital values (here, binary values) according to the clock CLK from the timing generation circuit 103.
Convert to The clock CLK is about 6 MHz when the entire screen is made up of 320 horizontal pixels and 240 vertical pixels. The binary image BP from the sampling circuit 102 is input to an output image synthesizing switching circuit 154 and a partial image extracting circuit 110 as shown in detail in FIG. 9, and the partial image extracting circuit 110 outputs one horizontal line of the image. The time-sequential binary image BP is converted into parallel image data a _{11 to} a ₅₅ of 5 × 5 = 25 pixels by being input to the shift registers 111 to 114 that perform the delay by each time. Nine pixels are selected from the parallel image data a _{11 to} a _55. This selection is made by setting the output nine pixels to P _{1 to} P _{9 in} FIG. 3 × 3 as shown in FIG.
9 or 9 pixels of a 5 × 5 square as shown in FIG. In this way, the relationship between the mask pattern and the size of the target image can be changed while maintaining a similar shape. Further, it can be easily increased to 7 × 7, 9 × 9 by adding a circuit. Further, by preparing a plurality of sets of mask patterns, simultaneous processing can be performed in parallel. With these functions, it is possible to extract features using a mask pattern of an appropriate size without changing the magnification of the imaging system in accordance with the size of the target pattern.

Selection of a pixel is performed by a selection signal SL from a control unit including the CPU 150 and the like.
, For example, by the multiplexer 115 shown in FIG. For example one of the pixels a ₁₁ or a ₂₂ is selected and output.

The image DP cut out as described above is input to the matching circuit 120 in the correlation circuit 160, and takes a correlation with an autocorrelation mask in a translation invariant sense. Matching circuit 120
Is matched in the range of the window signal WS generated by the window signal generation circuit 140, so that the window signal generation circuit 140 will be described first. The configuration of the window signal generation circuit 140 is as shown in FIG.
An example in which a circuit for processing only the hatched portion (window) of FIG. 14 is described. The horizontal synchronization signal HD, the vertical synchronization signal VD and the clock CLK are input from the timing generation circuit 103,
The counters 141 and 145 generate an X coordinate XS and a Y coordinate YS. That is, the counter 141 counts the clock CLK using the horizontal synchronization signal HD as a clear signal, and the counter 145 counts the horizontal synchronization signal HD using the vertical synchronization signal VD as a clear signal. The output XS of the counter 141 and the output YS of the counter 145 are coordinate values X ₁ , X ₂ and Y ₁ , Y ₂ shown in FIG. 14 which are set in the control unit by digital comparators 142, 143 and 146, 147, respectively.
After being compared, if an AND circuit 144, 148 and 149 of the X ₁ ≦ X coordinate _{XS ≦ X 2, Y 1 ≦} Y coordinate YS ≦ Y ₂ only window signal WS is outputted from the AND circuit 149. That is, the AND circuit 144 when the output XS of the counter 141 exists between the set values X ₁ and X ₂
Output AN1 becomes “1” and the output YS of the counter 145 becomes the set value.
Output AN2 becomes "1" of the AND circuit 148 when it is between Y ₁ and Y _2, the window signal WS output from the AND circuit 149 becomes "1" when the output AN1 and AN2 are "1" Output. Therefore, the window signal WS becomes "1" when it is in the shaded portion in FIG. The window signal WS is input to the matching circuit 120, and designates a range for matching. The correlation histogram is counted only when the window signal WS is 1 (or when 0), and the end of the window signal WS is a signal to end the counting. The window signal WS is effective because it can be used to process only the target, cut out according to the size of the target, or scan the screen one after another at arbitrary intervals.

In the matching circuit 120, for example, matching with the autocorrelation mask as shown in FIG. 5 is performed by taking the logical product of the "1" portion in FIG. This may be the sum of “1” parts when the image is multi-valued. The mask pattern is not limited to a 3 × 3 square as shown in FIG.
15 A hexagonal pattern as shown in FIG. The effect of the correlation by the hexagonal pattern is
Smaller variations in the pixel-to-pixel distance compared to the square pattern with the center pixel in the center, and the number of independent patterns can be reduced to 15 in the sense of invariant translation, which increases the efficiency compared to the case of Fig. 5. And so on. As an actual configuration, a method of shifting the clock phase by 180 degrees for each scanning line can be considered. FIG. 16 shows examples of independent autocorrelation masks (No. 1 to No. 1) in the sense of invariant translation of a hexagonal pattern.
5) is shown. “1” and “*” in the pattern are the fifth
This is the same as in the case of FIG.

The correlation data CL that has been matched by the matching circuit 120 is input to the counting circuit 130, and the outputs corresponding to the number of the autocorrelation masks are sequentially counted by a plurality of counters. The counted data is stored in the RAM 152. The data can be read from the RAM 152 at any time. After the data is stored, the data of the counting circuit 130 is initialized by the horizontal / vertical synchronization signal from the timing signal generation circuit 103.

The output image synthesis switching circuit 154 outputs the video image IS and the binary image BP from the television camera 101 to the switching signal PS from the CPU 150.
And the window signal WS can be displayed as needed. The CPU 150 performs necessary settings of the sampling circuit 102, the partial image cutout circuit 110, and the counting circuit 130, performs learning and calculation, and further performs measurement and output by a program stored in the ROM 151. An input device such as a keyboard is connected to the CPU 150 via an interface 153, and input of teaching data,
Instructions for learning sampling, output of results, and instructions for various settings are input. For example, the setting of the cut-out image size (5 × 5, 3 × 3) corresponds to the mask pattern No.
Since 1 represents the area (size) of the target pattern, if this value is larger than a certain value, 5 × 5 is used, and if it is smaller than a certain value, 3 × 3 is used.

In such a configuration, the object 100 captured by the television camera 101 is 25-dimensional vector as a matching histogram by the mask pattern (count value), is the feature extraction that is, as an initial characteristic x _j. Only the inside of the window is to be measured, and the initial feature _xj is extracted each time the window signal WS ends. In the learning process, a plurality of initial features x _j are stored in the RAM 152 together with information taught according to a plurality of instructions. In the operation analysis process, a plurality of data in the memory are processed by a multivariate analysis technique, and parameters for subsequent measurement are determined.
By the parameter is measured course, been determined for the initial features x _j extracted from an unknown object, the measurement result is output. FIG. 17 shows a flow of the learning process, and FIG. 18 shows a flow of the calculation. The steps of measurement and output are as shown in FIG.

Next, an actual application example of the above-described image measuring device will be described. For example, a “face wearing glasses” as shown in FIG. 20 and a “face wearing no glasses” as shown in FIG.
Can be applied to a system that identifies First, while presenting the "face with glasses shown in Fig. 20", set the windows so that each one fits into the window, and sample while teaching the class (whether or not you wear glasses). go. The sampling data is composed of a 25-dimensional vector and a class (“1” or “2”). Next, each of these faces is input so that a slight disturbance enters (for example, 10 times while giving a slight angular displacement). The same operation is performed for the "face without glasses" in FIG. Thus, the 90 initial features x are stored in the RAM 152, and the learning process ends. Subsequently, the CPU 150 performs a discriminant analysis in a calculation process, and a statistical feature optimal for discriminating between a face wearing glasses and a face not wearing is extracted by this analysis. In FIG. 22, the data of each face is one lump, and the class 1 and the class 2 are scattered separately. However, in FIG. 23 (after the judgment analysis), the class 1 and the class 2 are clearly separated. I understand. As a result, a statistical feature capable of discriminating between a face wearing glasses and a face not wearing glasses is extracted, and the system can be applied to an unknown face. FIG. 22 shows the results of principal component analysis with the first principal component on the horizontal axis and the second principal component on the vertical axis.
FIG. 23 shows the results of discriminant analysis, with the horizontal axis representing the first projection axis and the vertical axis representing the second projection axis. Measurement (identification) in the process, for the unknown face data x, after Y = A X, the determination of the class the average value and the identification rules for each class in the discriminant space (the face that does not face / over wearing glasses) Do. The result is, for example, output in the form of a text “face wearing glasses” in the case of class 1.

Effect of the Invention As described above, according to the measuring device of the present invention, the user of the device learns according to the actual object, and the device itself automatically selects the most appropriate measuring method in that situation. Measurement can be performed at high speed, so it can be applied to difficult applications where it is difficult to judge what kind of identification should be used, and because it can learn including the environment of the site, it is resistant to disturbance factors at the site Measurement becomes possible. In addition, a window function, a mask size switching function, and the like can provide an effective and adaptive feature extraction capability to a target, so that the device is not only adaptive but also easy to use. Furthermore, the object may exist anywhere in the window, high-speed real-time processing is possible due to parallel computation, and adaptive learning means that the same device can be applied to various uses (versatility). It is not necessary to give an algorithm of whether to perform measurement, and an image as an example and an answer to the image (measured value to be obtained) may be shown to the apparatus.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a conceptual diagram showing a basic measuring method of the present invention, FIG. 3 is a structural diagram showing a measuring principle of the present invention, and FIG. FIG. 5 is a diagram showing a displacement direction, FIG. 5 is a diagram showing a local autocorrelation mask up to the second order, FIG. 6 is a flowchart showing an operation example of initial feature extraction, and FIG. 7 is a diagram showing a linear combination of statistical feature extraction. ,
FIG. 8 is a flowchart of the operation, FIG. 9 is a detailed view of the partial image extracting circuit, FIGS. 10 and 11 (A) and (B) are views showing examples of the output, and FIG. FIG. 13 is a block diagram showing an example of a window signal generating circuit. FIG. 14 is a diagram for explaining the coordinates of the window. FIGS. 15 (A), (B) and FIG.
FIG. 17 is a diagram showing a local autocorrelation mask in the case of a square, FIG. 17 is a flowchart showing a learning process, FIG. 18 is a flowchart showing computation, FIG. 19 is a flowchart showing measurement and output, FIG. 20 to FIG. FIG. 24 is a diagram for explaining an application example of the apparatus of the present invention, FIG. 24 is a conceptual diagram showing a conventional measurement method, and FIG.
FIG. 26 and FIG. 26 are diagrams showing a parallel / adaptive method. 1,101 TV camera, 2 binarization circuit, 3 image memory, 4 geometric feature extraction unit, 5 statistical feature extraction unit, 6 identification unit, 100 measurement object , 102 ... sampling circuit, 103 ... timing generation circuit, 110 ...
Counting circuit, 140 ... Window signal generation circuit, 150 ... CPU, 151 ...
... ROM, 152 ... RAM.

──────────────────────────────────────────────────続 き Continued on the front page Examiner Shigekazu Ishii (56) References JP-A-59-34108 (JP, A) JP-B-58-47064 (JP, B2) (58) Fields investigated (Int. Cl. ⁶ , DB name) G06T 7/60 JICST Science and Technology Literature Database

Claims (4)

(57) [Claims] 1. An image pickup means for two-dimensionally picking up an image of a measurement object, a window signal generation means for generating a window signal indicating a measurement range in synchronization with an output of a video signal of the image pickup means, and an N-order autocorrelation. And has a plurality of sets of autocorrelation mask patterns of different sizes for extracting the local image from the video signal output by the imaging means in time series within the range of the window signal. Image extracting means, correlation calculating means for calculating an Nth-order autocorrelation coefficient value based on the local image from the image extracting means and the selected autocorrelation mask pattern, and a class for learning mode. Storage means for storing the Nth autocorrelation coefficient value calculated by the correlation calculation means together with the teaching data set by the teaching means to be taught each time the window signal is generated; and a plurality of Nth autocorrelation values stored in the storage means. The self-correlation coefficient value and the teaching data are analyzed by multiple regression analysis to calculate a statistical feature value for recognizing the class to be measured, and the Nth autocorrelation coefficient value is linearly integrated to obtain the statistical feature value. A statistical feature calculating means for calculating a coefficient for obtaining a value, and in the case of a measurement mode, an N-order autocorrelation coefficient value calculated by the correlation calculating means for an unknown measurement object is linearly calculated using the coefficient. An adaptive learning type general-purpose image measurement apparatus, comprising: a recognition unit for automatically recognizing the unknown measurement target class from the integrated characteristic value and the statistical characteristic value. 2. The adaptive learning type general-purpose image measuring apparatus according to claim 1, wherein said statistical feature calculating means uses discriminant analysis instead of said multiple regression analysis. 3. The apparatus according to claim 1, further comprising: means for sampling and binarizing the video signal; and means for displaying the captured image, the binarized image and / or the window signal. Adaptive learning type general-purpose image measurement device. 4. The adaptive learning type general-purpose image measuring apparatus according to claim 1, wherein said autocorrelation mask pattern has a hexagonal shape.