JP2010140082A - Image processing method and image processor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce a circuit scale, and to quickly and satisfactorily perform pattern matching of an image without changing a threshold. <P>SOLUTION: An image processor is provided with: an edge extraction part 2 of an original image, for generating an edge image from an original image input by an original image input part 1; a binarization part 3 for an original image, for binarizing the edge image generated by the edge extraction part 2 of the original image based on mutually different thresholds, and for generating a plurality of binary images; a correlation part 8 for calculating a correlation value between each of the plurality of binary images generated by the binarization part 3 for an original image and a binary image for reference stored in a memory part 7; and a correlation value selection part 9 for selecting the correlation value indicating that correlation is the highest from among a plurality of correlation values calculated by the correlation part 8. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an image processing method and an image processing apparatus, and more particularly, to an image processing method and an image processing apparatus that perform pattern matching satisfactorily for grayscale images that are difficult to binarize and enable image alignment and extraction of specific patterns. .

  In an automated production line or the like, an image processing apparatus is used in which a moving object is photographed with a camera or the like, and the position and orientation of an object in the image are automatically measured and the type of the object is determined.

  FIG. 7 is a block diagram showing the configuration of the image processing apparatus described in Patent Document 1. In the image processing apparatus shown in FIG. 7, a DOG (Difference of Gaussian) filtering unit 42 inputs an image captured by an original image input unit (camera) 41 and performs DOG filtering processing. The threshold processing means 43 ternarizes (-1, 0, +1) the output of the DOG filtering means 42. The ternarized image is stored in the second memory 48. The DOG filtering unit 45 performs DOG filtering processing on the reference pattern 44 stored in a predetermined memory. The threshold processing unit 46 ternizes the output of the DOG filtering unit 45. The ternarized image is stored in the first memory 47. The similarity calculation means 49 calculates the similarity between the image stored in the first memory 47 and the image stored in the second memory 48 and stores it in the score map 50 that is a storage unit. The control means 51 searches the score map 50 and finds the one with the highest similarity.

JP-A-6-76062 (paragraphs 0056-0066, FIG. 1)

  In the image processing apparatus shown in FIG. 7, since a DOG filter is used as a spatial bandpass filter, there is a problem that calculation is complicated and processing time is required. In addition, when the DOG filter is realized by hardware, there is a problem that the circuit becomes complicated and large-scale.

  In the image processing apparatus shown in FIG. 7, the reference pattern is stored in the apparatus in advance. Since the reference pattern is multi-bit data like the input original image, there is a problem that when a large number of reference patterns are stored, the memory capacity required for the storage means becomes large and the circuit becomes large. Another problem is that it takes a long time to search and select a desired reference pattern from a large number of reference patterns.

  Furthermore, in order for the threshold processing means 43 and 46 to ternarize the output images of the DOG filtering means 42 and 45, a threshold must be set. For example, when the image processing apparatus is used for an application such as recognition of natural terrain of an autonomous robot, it is preferable to input a scene in which sunshine conditions and lighting conditions change, and to change the threshold appropriately. However, since there is no human input in the autonomous robot that moves offline, there is a problem that the threshold setting cannot be easily changed.

  Accordingly, an object of the present invention is to provide an image processing method and an image processing apparatus capable of reducing the circuit scale and performing good pattern matching at high speed without changing the threshold value. .

  An image processing method according to the present invention includes a reference image input step of inputting a reference image to be a reference pattern for pattern matching, and an edge of the reference image for generating a reference edge image in which density change is emphasized from the input reference image A reference image binarization step that binarizes the reference edge image generated in the extraction step, the reference image edge extraction step, and generates a reference binary image; and an image to be subjected to pattern matching The original image input step for inputting the edge image, the edge extraction step for generating the edge image in which the density change is emphasized from the input original image, and the edge image generated in the edge extraction step for the original image are different from each other. Binarization based on the original image and generating a plurality of binary images, and the binarization step generated by the original image binarization step. A correlation step for calculating a correlation value between each of the binary images and the reference binary image generated in the reference image binarization step, and a plurality of correlation values calculated in the correlation step A correlation value selection step for selecting a correlation value indicating the highest correlation from the correlation value selection step, and an image identification step for outputting data for specifying an original image used as a basis for calculation of the correlation value selected in the correlation value selection step. It is characterized by including.

  An image processing apparatus according to the present invention generates a reference image that emphasizes a change in density from a reference image input unit that inputs a reference image used as a reference pattern for pattern matching and a reference image that is input by the reference image input unit. A reference image edge extraction unit, a reference image binarization unit that binarizes the reference edge image generated by the reference image edge extraction unit, and generates a reference binary image; From a memory unit that temporarily stores a reference binary image generated by the binarization unit, an original image input unit that inputs an image to be subjected to pattern matching, and an original image input by the original image input unit, The edge extraction unit of the original image that generates the edge image that emphasizes the density change and the edge image generated by the edge extraction unit of the original image are binarized based on different threshold values to generate a plurality of binary images. Correlation for calculating a correlation value between an original image binarization unit, each of a plurality of binary images generated by the original image binarization unit, and a reference binary image stored in the memory unit A correlation value selection unit that selects a correlation value that indicates the highest correlation among a plurality of correlation values calculated by the correlation unit, and an original that is a basis for calculating the correlation value selected by the correlation value selection unit. And an image identification unit that outputs data for specifying an image.

  According to the present invention, the circuit scale can be reduced and image pattern matching can be performed at high speed and satisfactorily without changing the threshold value.

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

  FIG. 1 is a block diagram showing an example of the configuration of an image processing apparatus according to the present invention. In the image processing apparatus shown in FIG. 1, an original image input unit 11 captures a scene to be subjected to pattern matching with a camera, A / D converts the imaged signal into digital data, and outputs the digital data to the edge extraction unit 12. The edge extraction unit 12 converts the input digital data of the original image into a multi-valued edge image by emphasizing a portion where the density change of the original image is large by two-dimensional differentiation, and outputs the edge image to the binarization conversion unit 130. . The binarization conversion means 130 is composed of a plurality of binarization means 131, 132,. The multi-valued edge image from the edge extraction means 12 is input to all the binarization means 131, 132,. The plurality of binarizing means 131, 132,..., 13n binarize the input edge images with different predetermined threshold values ("0" or "1"), and binarize the binary image. , 13n are output to binary correlation means 141, 142,..., 14n provided corresponding to each of the conversion means 131, 132,. That is, the binary image output from the binarizing unit 131 is input to the binary correlating unit 141, the binary image output from the binarizing unit 132 is input to the binary correlating unit 142, and the binarizing unit 13n The binary image to be output is input to the binary correlation means 14n.

  Further, before the scene is imaged, processing using each block in the broken line portion 21 shown in FIG. 1 is performed in advance. That is, the reference image input means 17, the edge extraction means 18 and the binarization means 19 that perform the same processing as the original image input means 11, the edge extraction means 12 and the binarization means 131, 132,. Thus, a template image for pattern matching is created from the reference image. Unlike the original image, only one binarizing means 19 is provided.

  The size of the template image is smaller than the size of the image input by the original image input unit 11. That is, in this embodiment, it is assumed that an image similar to the reference image is found from the scene captured by the original image input unit 11. However, the present invention can be applied even if the size of the template image is the same as the size of the image input by the original image input means 11. Further, the reference image is not stored in advance in the storage means, but is captured by the reference image input means 17 such as a camera when pattern matching is performed.

  Specifically, the reference image input means 17 that is an image pickup device such as a camera picks up a scene as a reference image in advance, A-D converts the image pickup signal into digital data, and outputs it to the edge extraction means 18. . The edge extracting unit 18 converts the input digital data of the reference image into a multi-valued edge image by the same two-dimensional differentiation process as that of the edge extracting unit 12 and outputs the multi-valued edge image to the binarizing unit 19. The binarizing means 19 binarizes the edge image input from the edge extracting means 18 with a preset threshold (“0” or “1”), and stores the binary image in the memory 20. The memory 20 stores a plurality of binary correlation means 141, 142,... Configured in accordance with the output timing of the binary images from the binarization means 131, 132,.・ Output to all 14n.

  The binary correlation means 141, 142,..., 14n are binary images of the original image input from the corresponding binarization means 131, 132,. And the two-dimensional coordinate value on the image of the point having the highest correlation and the correlation value thereof are output to the correlation value selection means 15. The correlation value selection means 15 selects the value having the highest correlation from the correlation values input from the binary correlation means 141, 142,..., 14n, and the two-dimensional coordinate value on the image is sent to the position identification means 16. Output. The position identification unit 16 outputs the coordinate value input from the correlation value selection unit 15 to an interface of a host device such as a control CPU for navigation.

  The components shown in FIG. 1 other than the original image input unit 11 and the reference image input unit 17 can be realized by a computer that executes processing according to a program.

  Next, the operation of each unit of the image processing apparatus shown in FIG. 1 will be described with reference to the flowchart of FIG. 2 and the explanatory diagram of FIG. FIG. 2 is a flowchart showing the operation of the image processing apparatus. FIG. 3 is an explanatory diagram illustrating an example of edge filter processing.

[Operation Example of Edge Extraction Units 12 and 18]
The edge extraction unit 12 performs a two-dimensional differentiation process on the digital data of the image input from the original image input unit 11 to generate a multi-valued edge image (step S21). Further, the edge extraction unit 18 performs a two-dimensional differentiation process on the digital data of the image input from the reference image input unit 17 to generate a multi-valued edge image (step S21).

  The edge extraction units 12 and 18 perform edge filter processing such as Sobel filter processing and Laplacian filter processing on a small region such as 3 × 3 pixels or 5 × 5 pixels centered on the target pixel in the input image, for example. An edge image is obtained by applying edge filtering to all pixels of the input image.

  Further, by using the edge filter described below, the edge extraction means 12 and 18 can perform powerful edge extraction as compared with the general edge filter processing described above. That is, a 3 × 3 pixel minute region is extracted centering on the target pixel in the input image, and each filter coefficient shown in each of FIGS. 3A to 3D and the gradation value of the extracted minute region image are displayed. And the multiplication results for 9 pixels are summed to obtain the absolute value of the sum. Then, the largest value among the four absolute values calculated using the respective filter coefficients shown in FIGS. 3A to 3D is set as the output value of the target pixel. By performing the above processing for all the pixels of the input image, an edge image having higher sensitivity than an image obtained by general edge filter processing can be obtained.

  When the filter coefficients shown in FIG. 3A are used, the edge extraction means 12 and 18 are arranged so that the minute region centered on the pixel of interest (near region in the range of several pixels centered on the pixel of interest: in this example) The absolute value of the value obtained by subtracting the sum of the shade values of the lower side from the sum of the shade values of the upper side in the range of 3 pixels in the vertical and horizontal directions is calculated. When the filter coefficient shown in FIG. 3B is used, the edge extraction means 12 and 18 calculate the absolute value of a value obtained by subtracting the sum of the gray values on the left side of the neighboring region from the sum of the gray values on the right side. is doing. When the filter coefficient shown in FIG. 3C is used, the edge extraction means 12 and 18 calculates the absolute value of the difference obtained by subtracting the sum of the shade values of the upper left area from the sum of the shade values of the lower right area. The value is being calculated. When the filter coefficients shown in FIG. 3D are used, the edge extraction means 12 and 18 calculate the absolute value of the value obtained by subtracting the sum of the shade values of the lower left region from the sum of the shade values of the upper right region of the neighboring region. Is calculated. In this example, the range of 3 × 3 pixels in the vertical and horizontal directions is set as the neighborhood region, but the size of the neighborhood region is not limited to 3 × 3 pixels.

  The edge extraction unit 12 outputs the multi-valued edge image by the edge filter process to the binarization units 131, 132,..., 13n (step S22). The edge extracting unit 18 outputs a multi-valued edge image obtained by the edge filter process to the binarizing unit 19 (step S22).

[Operation example of binarization means 131, 132,..., 13n and binarization means 19]
Each of the binarizing units 131, 132,..., 13n binarizes the edge image input from the edge extracting unit 12 with a predetermined threshold value, and each pixel value is “0” or “1”. A binary image is generated, and the binary image is output to the binary correlation means 141, 142,..., 14n (step S23). The binarizing unit 19 binarizes the edge image input from the edge extracting unit 18 with a predetermined threshold value, and generates a binary image having each pixel value “0” or “1”, thereby generating a binary image. Is output to the memory 20 (step S23). In the process of step S23, the binarizing means 131, 132,..., 13n and the binarizing means 19 are equal to or greater than a threshold value at which the gray value of the image is set for each of all pixels of the input edge image. If it is, “1” is output, and if it is less than the threshold, “0” is output.

  In addition, a threshold value adjusted in advance is set in the binarizing means 19, and different threshold values are set in advance in each of the binarizing means 131, 132,. It is assumed that the threshold value set in the binarizing means 19 is not the optimum threshold value in the binarizing means 131, 132,. This is because that. That is, different threshold values are set in advance for each of the binarization means 131, 132,. That is, in the present embodiment, it is not required to change the threshold value according to the difference in environmental conditions. Therefore, the present embodiment can be suitably applied to an application in which the threshold setting cannot be easily changed.

  Since the binarization means 131, 132,..., 13n operating at the time of inputting the original image are configured in plural, it is possible to execute binarization simultaneous parallel processing with different threshold values.

  Further, the threshold values set in the binarizing means 131, 132,..., 13n are determined based on image characteristics by, for example, creating a grayscale histogram from the input edge image, and a plurality of threshold values based on the respective image characteristics. The threshold values may be automatically set.

[First operation example of binary correlation means 141, 142,..., 14n]
The binary correlation means 141, 142,..., 14n are correlation values between the binary image input from the corresponding binarization means 131, 132,..., 13n and the binary image input from the memory 20, respectively. Is calculated (step S24). That is, the binary image input from the memory 20 is used as a template image, and pattern matching is performed for each of the binary images input from the binarizing means 131, 132,. Since the binary image input from the memory 20 is a template image, the image size of the binary image input from the memory 20 is larger than the image size of the binary image input from the binarization means 131, 132,. Is also small. An example of the calculation of the correlation value r is shown in equation (1).

      r = Σ | f−g | (1)

  In Expression (1), g is a binary image input from the memory 20, that is, a bit value (“0” or “1”) of the template image. f is a bit value of a partial image obtained by cutting out a region having the same size as the template image from the binary image input from the binarizing means 131, 132,. The range of Σ is the number of pixels of the template image. Therefore, according to equation (1), for all the pixels of the template image and the matching target image, the absolute difference value of f and g is calculated at the corresponding pixels of the template image and the matching target image, and the total value is correlated. It becomes the value r.

  In this example using equation (1), the smaller the correlation value r, the higher the correlation between the template image and the matching target image. The binary correlation means 141, 142,..., 14n shift the region of the partial image cut out from the binary image input from the binarization means 131, 132,. Then, the correlation value r is calculated. Further, it is shifted by one pixel in the vertical direction, and the entire binary image inputted from the binarizing means 131, 132,..., 13n is scanned, and the binarizing means 131, 132,. Correlation values r are calculated for all partial images that can be cut out from the input binary image.

[First Operation Example of Correlation Value Selection Unit 15]
The correlation value selection means 15 selects the value indicating the highest correlation among the correlation values r calculated by the binary correlation means 141, 142,..., 14n, that is, the minimum value of the correlation value r in this example. Then, the selected correlation value r is output to the position identifying means 16 together with the partial image cut-out coordinate value at that time (step S25). The partial image cutout coordinate value is, for example, a value indicating the upper right coordinate of the partial image.

[Second operation example of binary correlation means 141, 142,..., 14n]
Another example of the calculation of the correlation value r is shown in equation (2).

      r = Σ (f xnor g) (2)

  In the equation (2), Σ, f, g represents the same data as Σ, f, g in the equation (1). Xnor represents a value obtained by inverting the exclusive OR of f and g. Hereinafter, inverting exclusive OR is referred to as exclusive OR inversion.

  FIG. 4 is a circuit diagram showing a configuration example for realizing the calculation of the equation (2). In the configuration shown in FIG. 4, f and g are input serially for each corresponding pixel in synchronization with the synchronization signal CLK. The flip-flop 31 aligns the input timings of f and g and outputs the same to the exclusive OR inversion circuit 32. The exclusive OR inversion circuit 32 outputs “0” to the adder 33 if the bits of f and g are different, and outputs “1” to the adder 33 if they are the same. Therefore, the adder 33 adds up the number of bits in which f and g are the same. The adder 33 outputs the added value to the multi-bit flip-flop 34. The multi-bit flip-flop 34 holds a cumulative value of the same number of bits of f and g that are output values of the adder 33.

  By configuring the cumulative value held by the flip-flop 34 to be fed back to the adder 33, the adder 33 adds the fed back value and the value newly input from the exclusive OR inversion circuit 32. To update the total number of bits in which f and g are the same. At the stage where f and g of all the pixels are inputted, the final correlation value r is outputted to the output of the flip-flop 34 shown in FIG.

  Instead of the adder 33 and the multi-bit flip-flop 34 shown in FIG. 4, an up counter that uses the output signal of the exclusive OR inversion circuit 32 as an enable input may be used.

[Second Example of Operation of Correlation Value Selection Unit 15]
In the example using Expression (2), the correlation between the template image and the matching target image increases as the correlation value r increases. When using equation (2), as in the case of using equation (1), the binary correlation means 141, 142,..., 14n are input from the binarization means 131, 132,. The region of the partial image cut out from the binary image is shifted by one pixel, the entire image is scanned, and all correlation values r are calculated. Correlation value selection means 15 selects a value indicating the highest correlation among the correlation values r, in the example using equation (2), the maximum value of correlation values r, and selects the selected correlation value r at that time. Are output to the position identifying means 16 together with the partial image cutout coordinate values.

  The position identification unit 16 outputs the coordinate value indicating the partial image cut out when the correlation value r selected by the correlation value selection unit 15 is calculated to another device that uses the output of the image processing device. The other apparatus may be provided separately from the image processing apparatus having the configuration illustrated in FIG. 1, or may be another part of the image processing apparatus that partially includes the configuration illustrated in FIG. May be.

  As an example, another apparatus (or another part of the image processing apparatus) cuts out an image specified by the partial image cut-out coordinate value output from the position identification unit 16 from the original image, and in an automated production line, An image for observing a specific part is displayed on a display, or a specific product is identified in a product sorting line to control a sorting mechanism.

  As described above, in this embodiment, instead of performing complex and time-consuming image pattern matching such as normalized correlation of multi-bit data, correlation calculation of binary data is performed. Image pattern matching can be performed with a high-speed and small-scale circuit.

  Also, instead of performing a complex and time-consuming image filter such as a DOG filter or a LOG (Laplace Of Gaussian) filter for binarization pre-processing, it is a small size such as 3 × 3 pixels or 5 × 5 pixels. Since the edge extraction filter is used, image pattern matching can be performed with a high-speed and small-scale circuit.

  Further, in the binarization of the image, the threshold value is not changed in accordance with a change in environmental conditions or the like, but includes a plurality of binarization means based on different threshold values and a plurality of binary correlation means corresponding to each of them. Since the one that outputs the value having the highest correlation among them is automatically selected, it is possible to perform image pattern matching without setting a threshold with human intervention.

  Further, instead of storing an image of multi-bit data as a reference pattern, a reference image is input offline, that is, before a pattern matching target scene is input, and an edge of the reference image is extracted and a binary image is stored in memory. Since the reference pattern stored in the memory 20 is binary data, the memory capacity of the memory 20 can be reduced, and image pattern matching can be achieved with a high-speed and small-scale circuit. Can be realized.

  From the above, the image processing method and the image processing apparatus according to the present embodiment are used in, for example, an automated production line or the like for external recognition in autonomous guidance in transportation means such as a vehicle, a ship, and an aircraft, or When the position and orientation of an object in an image is automatically measured or the type of an object is automatically determined in an external environment recognition application for an autonomous robot, the surface state of the object is uneven or the lighting conditions are unstable. Even in such a case, the object can be recognized satisfactorily and stably. Further, when obtaining three-dimensional information of an object by the principle of stereo vision, it is possible to stably extract the three-dimensional information even in a situation where a plurality of objects may exist in the field of view. Further, when measuring the movement of the recognition target object, the object can be measured at high speed and with high reliability.

  FIG. 5 is a block diagram showing the main part of the image processing apparatus according to the present invention. As shown in FIG. 5, the image processing apparatus includes a reference image input unit 4 (corresponding to the reference image input unit 17 in FIG. 1) that inputs a reference image used as a reference pattern for pattern matching, and a reference image input unit 4. The reference image edge extraction unit 5 (corresponding to the edge extraction unit 18 in FIG. 1) that generates a reference edge image in which the density change is emphasized from the input reference image, and the reference image generated by the reference image edge extraction unit 5 A reference image binarization unit 6 (corresponding to the binarization means 19 in FIG. 1) for binarizing the edge image for reference and generating a reference binary image, and a reference image binarization unit 6 A memory unit 7 (corresponding to the memory 20 in FIG. 1) that temporarily stores the generated binary image for reference, and an original image input unit 1 (an original image input in FIG. 1) that inputs an image to be subjected to pattern matching Mean 1 ), An original image edge extracting unit 2 (corresponding to the edge extracting means 12 in FIG. 1) for generating an edge image in which density change is emphasized from the original image input by the original image input unit 1, and The edge image generated by the edge extraction unit 2 is binarized based on different threshold values, and a binarization unit 3 for original image that generates a plurality of binary images (corresponding to the binarization means 131 to 13n in FIG. 1). And a correlation unit 8 that calculates a correlation value between each of the plurality of binary images generated by the original image binarization unit 3 and the reference binary image stored in the memory unit 7 (FIG. 1). And a correlation value selection unit 9 for selecting a correlation value indicating the highest correlation among a plurality of correlation values calculated by the correlation unit 9 (correlation value selection in FIG. 1). The correlation value selection unit 9 selects And an image identification unit 10 for outputting data identifying the original image becomes the basis for calculation of the function value (corresponding to the position identification means 16 in FIG. 1).

  FIG. 6 is a flowchart showing the main steps in the image processing method according to the present invention. As shown in FIG. 6, the image processing method includes a reference image input step (step S1) for inputting a reference image to be a pattern matching reference pattern, and a reference edge in which density change is emphasized from the input reference image. Reference image 2 for generating a reference binary image by binarizing the reference edge image generated in the reference image edge extracting step (step S2) for generating an image and the reference image edge extracting step. A valuation step (step S3), an original image input step (step S4) for inputting an image to be subjected to pattern matching, and an edge of the original image for generating an edge image in which density change is emphasized from the input original image The edge image generated in the extraction step (step S5) and the edge extraction step of the original image is binarized based on different threshold values, A binarization step for original images for generating a number of binary images (step S6), a plurality of binary images generated in the binarization step for original images, and a binarization step for reference images A correlation step (step S7) for calculating a correlation value with the reference binary image thus selected, and a correlation value indicating the highest correlation among the plurality of correlation values calculated in the correlation step is selected. A correlation value selection step (step S8) and an image identification step (step S9) for outputting data for specifying the original image that is the basis of calculation of the correlation value selected in the correlation value selection step are included.

  In the image processing apparatus, the correlation unit has a partial image having the same size as the size of the reference image in each of the plurality of binary images (the size is larger than the size of the reference image) generated by the original image binarization unit. And the image identification unit is configured to output coordinate data that identifies the partial image of the original image that is the basis of the calculation of the correlation value selected by the correlation value selection unit. Also good. In such a configuration, a desired image portion can be specified from the original image.

  In addition, the image processing apparatus calculates correlation values while shifting the partial image cutout region, so that the correlation value is obtained for all partial images that can be cut out from the binary image generated by the binarization unit for the original image. It may be configured to calculate.

  In addition, the image processing apparatus has an absolute value of a value obtained by subtracting the sum of the shade values of the lower side from the sum of the shade values of the lower side from the sum of the shade values of the upper side in the vicinity region of the range of several pixels around each pixel of the input image. And a value obtained by subtracting the sum of the gray values of the right side from the sum of the gray values of the left side of the neighboring region and the horizontal differentiation unit (included in the edge extracting units 12 and 18 in FIG. 1; see FIG. 3A). The vertical differentiation means for calculating the absolute value (included in the edge extraction means 12 and 18 in FIG. 1; see FIG. 3B) and the sum of the gray values in the upper left area of the neighboring area to the total of the gray values in the lower right area From the upper right lower left oblique differentiation means (which is included in the edge extraction means 12 and 18 in FIG. 1; see FIG. 3C) and the sum of the gray values in the upper right area of the neighboring area. The absolute value of the difference of the total gray value in the lower left area The upper left lower right oblique calculating means (included in the edge extracting means 12 and 18 in FIG. 1; see FIG. 3D), the absolute value obtained by the upper left lower right oblique differentiating means, and the upper right lower left oblique The absolute value obtained by the differentiating means, the absolute value obtained by the vertical differentiating means, and the absolute value obtained by the horizontal differentiating means are compared, and a selecting means (selecting the edge extracting means 12, 18 in FIG. 1) is selected. May be included).

  In the image processing apparatus, the correlation unit inverts the exclusive OR of the binary image value generated by the reference image binarization unit and the binary image value generated by the original image binarization unit. A coincidence detection circuit (corresponding to the exclusive OR inversion circuit 32 in FIG. 1), and a holding circuit such as a register that holds the cumulative value of the number of pixels whose value obtained by the coincidence detection circuit is “1” Flip-flop 34) and an adder (corresponding to adder 33 in FIG. 1) that adds the cumulative value held by the holding circuit and the inversion of the exclusive OR obtained by the coincidence detection circuit May be.

  The present invention can be applied to an automated production line, and can be applied to uses such as external recognition in transportation means and autonomous robots.

It is a block diagram which shows an example of a structure of the image processing apparatus by this invention. It is a flowchart which shows operation | movement of an image processing apparatus. It is explanatory drawing which shows an example of an edge filter process. It is a circuit diagram which shows the structural example which implement | achieves the calculation of (2) Formula. It is a block diagram which shows the principal part in the image processing apparatus by this invention. It is a flowchart which shows the main steps in the image processing method by this invention. 10 is a block diagram illustrating a configuration of an image processing apparatus described in Patent Document 1. FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Original image input part 2 Original image edge extraction part 3 Original image binarization part 4 Original image binarization part 5 Reference image edge extraction part 6 Reference image binarization part 7 Memory part 8 Correlation part 9 Correlation value selection unit 10 Image identification unit 11 Original image input unit 12 Edge extraction unit 15 Correlation value selection unit 16 Position identification unit 17 Reference image input unit 18 Edge extraction unit 19 Binarization unit 20 Memory 31 Flip-flop 32 Exclusive OR Inverting circuit 33 Adder 34 Flip-flop 130 Binary conversion means 131 to 13n Binary means 141 to 14n Binary correlation means

Claims (10)

A reference image input step for inputting a reference image to be a pattern matching reference pattern;
An edge extraction step of a reference image for generating a reference edge image in which density change is emphasized from the input reference image;
A reference image binarization step that binarizes the reference edge image generated in the reference image edge extraction step to generate a reference binary image;
An original image input step for inputting an image to be subjected to pattern matching;
An edge extraction step of the original image for generating an edge image in which the density change is emphasized from the input original image;
Binarization step for original image that binarizes the edge image generated in the edge extraction step of the original image based on different threshold values, and generates a plurality of binary images;
Correlation step of calculating a correlation value between each of the plurality of binary images generated in the original image binarization step and the reference binary image generated in the reference image binarization step. When,
A correlation value selection step for selecting a correlation value indicating that the correlation is highest among the plurality of correlation values calculated in the correlation step;
And an image identification step of outputting data for specifying an original image that is a basis for calculating the correlation value selected in the correlation value selection step. An image processing method using a reference image having a size smaller than that of an original image,
In the correlation step, a correlation value is calculated between the size of the reference image and the partial image of the same size in each of the plurality of binary images generated in the binarization step for the original image,
The image processing method according to claim 1, wherein, in the image identification step, coordinate data for specifying a partial image of the original image that is a basis for calculation of the correlation value selected in the correlation value selection step is output. The correlation value is calculated for all partial images that can be cut out from the binary image generated in the binarization step for the original image by calculating the correlation value while shifting the partial image cut-out region in the correlation step. Image processing method. In the edge extraction step, an image processing method for generating an edge image by performing calculation of a neighborhood region of a range of several pixels around each pixel of an input image,
The edge extraction step includes
A lateral differentiation step of calculating an absolute value of a value obtained by subtracting the sum of the shade values of the lower side from the sum of the shade values of the upper side in the vicinity region;
A longitudinal differentiation step of calculating an absolute value of a value obtained by subtracting the sum of the shade values of the right side from the sum of the shade values of the left side of the neighborhood region;
An upper right lower left diagonal differentiation step of calculating an absolute value of a value obtained by subtracting the sum of the shade values of the lower right region from the sum of the shade values of the upper left region of the vicinity region;
An upper left lower right diagonal differentiation step for calculating an absolute value of a value obtained by subtracting the sum of the shade values of the lower left region from the sum of the shade values of the upper right region of the vicinity region;
The absolute value obtained in the upper left lower right oblique differentiation step, the absolute value obtained in the upper right lower left oblique differentiation step, the absolute value obtained in the vertical differentiation step, and the lateral differentiation step. The image processing method according to claim 1, further comprising: a selection step of comparing the absolute values with each other and selecting a maximum value. The correlation step is
A match detection step for inverting the exclusive OR of the binary image value generated in the reference image binarization step and the binary image value generated in the original image binarization step;
A cumulative holding step of holding a cumulative value of the number of pixels whose value obtained in the match detection step is “1”;
5. The method according to claim 1, further comprising: an adding step of adding the cumulative value obtained in the cumulative holding step and the inversion of the exclusive OR obtained in the coincidence detecting step. Image processing method. A reference image input unit for inputting a reference image used as a reference pattern for pattern matching;
An edge extraction unit for a reference image that generates a reference edge image in which density change is emphasized from the reference image input by the reference image input unit;
A reference image binarization unit that binarizes the reference edge image generated by the edge extraction unit of the reference image and generates a reference binary image;
A memory unit for temporarily storing a reference binary image generated by the reference image binarization unit;
An original image input unit for inputting an image to be subjected to pattern matching;
From the original image input by the original image input unit, an edge extraction unit of the original image that generates an edge image that emphasizes density change;
An original image binarization unit that binarizes the edge image generated by the edge extraction unit of the original image based on different threshold values, and generates a plurality of binary images;
A correlation unit that calculates a correlation value between each of a plurality of binary images generated by the original image binarization unit and a reference binary image stored in the memory unit;
A correlation value selection unit that selects a correlation value indicating that the correlation is highest among a plurality of correlation values calculated by the correlation unit;
An image processing apparatus comprising: an image identification unit that outputs data for specifying an original image that is a basis for calculating a correlation value selected by the correlation value selection unit. An image processing apparatus that uses a reference image having a size smaller than that of an original image,
The correlation unit calculates a correlation value between the size of the reference image and the partial image of the same size in each of the plurality of binary images generated by the original image binarization unit,
The image processing apparatus according to claim 6, wherein the image identification unit outputs coordinate data that specifies a partial image of the original image that is a basis for calculating the correlation value selected by the correlation value selection unit. The correlation unit calculates correlation values for all partial images that can be cut out from the binary image generated by the original image binarization unit by calculating a correlation value while shifting the partial image cutout region. Image processing device. An edge extraction unit is an image processing device that generates an edge image by performing calculation of a neighborhood region of a range of several pixels around each pixel of an input image,
The edge extraction unit
Lateral differentiation means for calculating an absolute value of a value obtained by subtracting the sum of the shade values of the lower side from the sum of the shade values of the upper side in the vicinity region;
Vertical differentiation means for calculating the absolute value of the value obtained by subtracting the sum of the shade values of the right side from the sum of the shade values of the left side of the neighborhood region;
Upper right lower left oblique differentiation means for calculating an absolute value of a value obtained by subtracting the sum of the gray value of the lower right region from the sum of the gray value of the upper left region of the vicinity region;
Upper left lower right oblique differentiation means for calculating an absolute value of a value obtained by subtracting the sum of the gray value of the lower left region from the sum of the gray value of the upper right region of the vicinity region;
The absolute value obtained by the upper left lower right oblique differentiation means, the absolute value obtained by the upper right lower left oblique differentiation means, the absolute value obtained by the vertical differentiation means, and the absolute value obtained by the lateral differentiation means The image processing apparatus according to claim 6, further comprising: a selecting unit that compares the two and selects a maximum value. The correlation part is
A coincidence detection circuit that inverts the exclusive OR of the binary image value generated by the reference image binarization unit and the binary image value generated by the original image binarization unit;
A holding circuit for holding a cumulative value of the number of pixels whose value obtained by the coincidence detection circuit is “1”;
10. The image processing according to claim 6, further comprising: an adder that adds a cumulative value held by the holding circuit and an inversion of an exclusive OR obtained by the coincidence detection circuit. apparatus.

Images