JPH11144057A - Device and method for image recognition - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide the device and method for image recognition which can perform image recognition with good precision in a production line easily changing in environment. SOLUTION: In a template registering means 2, templates can be registered for one machine kind and one image to be recognized and when templates are registered for one machine kind and one image to be recognized, a product discriminating means 3 performs pattern matching against templates for the printed logo type on a product surface.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image recognition apparatus and an image recognition method for recognizing a product by image recognition.

[0002]

2. Description of the Related Art In recent years, for example, in the semiconductor industry, products
Technology that recognizes products by character recognition of characters printed on the surface of (electronic components) has attracted attention, and this type of recognition technology uses barcodes to recognize products. In addition, there are recognition techniques for recognizing a production number, and further, recognition techniques for dealing with missing or stained printed characters.

For example, Japanese Patent Application Laid-Open No. 7-175894 discloses that
There is disclosed a technique for recognizing characters printed on the surface of an electronic component by a recognition method using a neural network, preventing erroneous mounting, and simultaneously controlling quality and production.

Japanese Patent Application Laid-Open No. Hei 6-44375 discloses a technique for detecting a different material by comparing image data on the upstream side with data on the downstream material in the substance management of a material marked with characters or the like. It is shown.

[0005]

However, in the prior art described above, the recognition of the product is performed only by the recognition of the characters printed on the product. It is not possible to detect even the matching between the product and the printed characters for parts that are printed and supplied with characters after they have been manufactured. In addition, a logo or the like printed on the surface of a product must be considered as a designed pattern in terms of appealing the product more than the role as character information, and it is not easy to match with the standard of characters.

There are many methods for speeding up the detection of recognition results using a learning function such as a neural network. However, when a product of high-mix low-volume production is to be recognized, the recognition target is learned. Within the period until
In many cases, production is terminated (cut off).

Further, when a template is created upstream of a production line and is used for recognition in another process using the template, there is a possibility that dirt or the like may be attached in a difference in environmental lighting or in a process in the middle.

An object of the present invention is to provide an image recognition apparatus and an image recognition method capable of performing accurate image recognition on a production line in which the environment is easily changed.

In particular, the present invention provides a technique for easily registering a template and performing pattern matching with high accuracy, and when model information of a recognition target is known in advance, the matching becomes more effective. It is an object of the present invention to provide an image recognition device and an image recognition method that can perform the operation.

[0010]

In order to achieve the above-mentioned object, the invention according to claim 1 provides an image pickup means for picking up a printed logo on a surface of a product conveyed on a production line, and an image pickup means for picking up an image. Template registration means for registering the printed logo image as a template, and pattern matching between the printed logo on the surface of the product and the template registered in the template registration means for a product after being transported on the production line. Product identification means for performing product identification by performing, and presentation means for presenting whether to register a logo image of a product currently in the identification space as a new template if the product identification result by the product identification means does not exceed the recognition threshold. In the template registration means, a plurality of templates can be registered for one model and one recognition target image. It is, the product identification means,
When a plurality of templates are registered for one model and one recognition target image, pattern matching with a plurality of templates is performed on a printed logo on the product surface.

According to a second aspect of the present invention, there is provided an image capturing means for capturing a printed logo on the surface of a product conveyed on a production line, and a template registration for registering a printed logo image captured by the image capturing means as a template. Means,
Product identification means for product identification by performing pattern matching between the printed logo on the surface of the product and the template registered in the template registration means, for products after being transported on the production line, A plurality of templates can be registered for one model and one recognition target image in the template registration means. If the product identification result by the product identification means does not exceed the recognition threshold, the template registration means The logo image of the product in the identification space is registered as a new template. When a plurality of templates are registered for one model and one image to be recognized, the product identification unit can register the logo on the product surface. The feature is that pattern matching with a plurality of templates is performed.

According to a third aspect of the present invention, in the image recognition device according to the first or second aspect, the template registering means captures an image after the image for template registration is captured by the image capturing means. The recognition target is divided into components by performing labeling of the obtained image, the feature amount of each component is detected, and a data structure including a component group is used as a template for each recognition target in which the feature amount of each component is detected. Create and prepare multiple templates for each part in one model and one recognition target image.
In this case, the product identification means performs pattern matching with the plurality of templates for each component in the product captured image.

According to a fourth aspect of the present invention, in the image recognition apparatus according to the first or second aspect, the template registering means performs pattern matching for each component to be recognized with a registered template. When the comparison with the recognition threshold value of each component is performed, only a component having a low matching rate is registered as a template in a data structure including a component group.

According to a fifth aspect of the present invention, in the image recognition apparatus according to the first or second aspect, the template registration means determines that the matching rate of the result of pattern matching between the recognition target image and the template is different. If it is close to the recognition target candidate, the result of matching is prepared as a template, and the product identification unit performs matching of the recognition target itself and matching of the image after matching of the recognition target itself in the subsequent pattern matching, The method is characterized in that after performing recognition from a group of recognition targets having similar shapes by matching of the recognition target itself, matching of the image after matching of the recognition target itself is performed for further confirmation.

According to a sixth aspect of the present invention, in the image recognition apparatus according to the fifth aspect, the template registration means determines that the matching rate of matching of the image after matching of the recognition target itself has a continuously high value. It is characterized in that the obtained frequency is detected, and when the frequency exceeds a certain value, the input image to be recognized is registered as a new template, while the previous template is deleted.

According to a seventh aspect of the present invention, in the image recognition apparatus of the third or fifth aspect, when the product identification means obtains a matching image by performing pattern matching, the product identification means performs labeling of the matching image. The feature amount of each part in the matching image is detected, and by comparing the feature amount of each part in the matching image and the recognition target image,
A new data structure consisting of a group of parts is created by selecting from the parts in the image to be recognized as a new part to be recognized, pattern matching is performed only on the selected parts, and a group of recognition objects with similar shapes Is characterized in that the parts to be recognized in the image are selected.

According to an eighth aspect of the present invention, in the image recognition apparatus according to the fifth or seventh aspect, the template registering means performs the matching after matching the parts having a low matching rate of the pattern matching for each part. Is added to the data structure as a template, while recognition target parts in the images of the recognition target group having similar shapes are selected to reduce the number of template candidates.

According to a ninth aspect of the present invention, in the image recognition apparatus according to the first aspect, the template registration unit captures a print logo on a surface of a product conveyed on a production line by the image capture unit, Performs pattern matching between the captured logo image as the recognition target image and the template image, and if the recognition target image and the model of the template match, deletes the mismatched part between the printing logo image and the template image. While registering as
If the model of the recognition target image does not match the model of the template, the matching part between the printed logo image and the template image is registered as a deletion area, and the deletion areas of both are set as deletion areas specific to the model to be recognized. Is performed before the recognition target image and the template image are deleted.

The invention according to claim 10 is the first invention.
In the image recognition device described above, the template registration unit first creates a representative template and stores the representative template in a first template.
Is registered as a second template, and the result of pattern matching between the recognition target and the template by the product identification means is registered as a second template.

According to the eleventh aspect of the present invention, in the image recognition apparatus according to the first, second or seventh aspect, the template registering means obtains a high degree of matching as a result of a plurality of pattern matchings. If it is not possible, labeling is performed in each pattern in a plurality of patterns, matching is performed for each component, optimal matching is performed in each component, and an image obtained by combining them is registered as one template. I have.

The invention according to claim 12 is the first invention.
0, the template registration unit registers an image of a common part of the recognition target images of all models as a representative template, and registers a new recognition target image and a representative image for a new recognition target image. The image of the common part with the target template image is generated as a new representative template image, and a result image obtained by pattern matching between the recognition target image and the template image is used as a unique template image for the recognition target. The product identification means is registered, and the result of pattern matching with the representative template is subjected to pattern matching with the unique template.

Further, the invention according to claim 13 is based on claim 1
In the image recognition device described above, the template registration unit deletes the detected recognition target from the image when the image capturing unit acquires a recognition target image at a correct position and detects a recognition target in the image by pattern matching. An image obtained by subtracting a basic image from an image captured by an image capturing unit is obtained as a template, using an image from which a recognition target has been deleted as a basic image.

[0023]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing a configuration example of an image recognition device according to the present invention. Referring to FIG. 1, the image recognition apparatus includes an image capturing unit 1 for capturing a printed logo on the surface of a product conveyed on a production line, and a template for registering a printed logo image captured by the image capturing unit 1 as a template. Product identification for registering the product by performing pattern matching between the printed logo on the product surface and the template registered in the template registration means 2 for the registration means 2 and the products conveyed on the production line. And means for presenting whether a logo image of a product currently in the identification space is to be registered as a new template when the result of product identification by the product identification means does not exceed the recognition threshold value. The registration means 2 can register a plurality of templates for one model and one recognition target image. Another means 3
When a plurality of templates are registered for one model and one recognition target image, pattern matching with a plurality of templates is performed on a printed logo on the product surface.

FIG. 2 is a flowchart for explaining the processing operation of the image processing apparatus of FIG. In the description of FIG. 2, for simplicity, two types of models A and B are used as products.
It is assumed that different types of models are manufactured. Referring to FIG. 2, the operator inputs model information (for example, information on model A and model B) to the image recognition apparatus in order to recognize whether the product is of model A or B. (Step S1). It is also assumed that templates are registered in the template registration means 2 for each of the models A and B together with the input of model information on the models A and B. For example, as shown in FIG. 3, image1, image2,.
In addition, image1 ′, i
.. are registered.

After such initial setting, a print logo on the surface of the product conveyed on the production line is picked up by the image pick-up means 1, and the image is subjected to preprocessing.
(Step S2). Thereafter, pattern matching is performed between the preprocessed image and the template registered in the template registration means 2 (step S3).

In the present case, as shown in FIG.
are registered as image1, image2,..., and image1 ', image1
Since 2 ′,... Are registered, the preprocessed image and the template image1, image2,.
pattern matching is performed with ge1 ′, image2 ′,.

As a result, the image just captured and each template image1, image2,..., Image
When all the degrees of coincidence with 1 ′, image2 ′,... Are lower than the recognition threshold (when the degree of coincidence (similarity) is low) (step S4), the image that has just been picked up is selected from either model A or B. Whether to register as a template is presented to the operator (step S5).

Thus, when the operator registers this as a template of the model A, for example, it is newly registered as a template of the model A (step S6).

On the other hand, when it is determined in step S5 that the operator does not register the image as a template, the currently captured image is re-recognized (step S7).

As described above, in the image processing apparatus of FIG. 1, the template for the recognition target image of the model A is stored in the image
, Image2,..., The template for the recognition target image of model B is image1 ′, image2 ′,..., Each model has a plurality of templates, and when performing pattern matching in step S3 of the flowchart of FIG. Set a threshold value, perform pattern matching with all templates, and if all of the pattern matching results (matching degrees) with each template do not exceed the threshold value (when it is determined that all templates are not similar) ), This recognition target image can be added as a new template.

The image recognition apparatus shown in FIG. 1 prevents mistakes in assembling parts due to the common use of parts due to rationalization of production, etc., setting of a dark room where it is difficult for an operator to intervene, and fixing of an inflexible optical system. It can also be applied to production lines that focus on manual assembly without the need to use closed optical systems that are difficult for humans to intervene. In other words, the captured image (input image) that is affected by the unevenness of the surface shape roughness due to environmental lighting that is difficult to stabilize, the accuracy of the mold for product molding, the difference of the manufacturer, etc.
By preparing a plurality of templates for one model with respect to the variation of, it is possible to accurately recognize the product without performing erroneous recognition, erroneous determination, and excessive recognition.

Therefore, for example, recognition of a logo printed on the front of a product that is being irregularly produced in a large variety and in small quantities is performed by a production line mainly composed of hands (for example, ceiling, line side,
This is useful in a case where the lighting is performed on a production line where environmental lighting that is independently operated is installed on the next ceiling and the next line side.

In the image processing apparatus shown in FIG. 1, a plurality of types of products (for example, two types of products A and B) are conveyed on a production line. It was necessary to present to the operator whether or not to register as (registering means 4 was required). However, for example, only one type of product (for example, product A) is transported on the production line. In such cases, only one model is registered as a template, and as a result of pattern matching between the captured image and the template,
When the degree of coincidence is lower than the recognition threshold (when the degree of coincidence is low), the captured image can be automatically registered as a template of one model. In this case, the image recognition device can be configured as shown in FIG. 4 instead of FIG. That is, the presentation means 4
Need not be provided.

In the image recognition apparatus shown in FIG. 1 or FIG. 4, the template registration means 2 more specifically, after the image for template registration is captured by the image capturing means 1, performs labeling of the captured image. The recognition target is divided into components, the feature amount of each component is detected, and a data structure including a component group is created as a template for each recognition target in which the feature amount of each component is detected.
A plurality of templates can be prepared for each component in one recognition target image. In this case, the product identification unit 3
Pattern matching with a plurality of templates is performed on each component in a product captured image.

For example, assuming that the logo attached to the product of the model A is a logo composed of α, β, and γ as shown in FIG.
image1 is image1, image1 of each of the three parts α, β, γ, image1α, image
e1β and image1γ. If the logo attached to the product of the model A is a logo composed of α, β, and γ as shown in FIG.
(Logo as recognition target) is, as a whole, image2
Image2 has three parts α, β,
Each image image2α, image2β, im of γ
age2γ.

Thus, the logo of FIG.
image1α, image1β, image of ge1
As 1γ, a logo composed of α, β, and γ as shown in FIG. 6 is used as image2α, im of image2 of model A.
The data can be registered as age2β and image2γ in a data structure having the model A as a tag as shown in FIG. If the template is registered in a data structure as shown in FIG. 7, pattern matching is performed independently for each component, and the detection result is set as a tag, model A, based on the overall matching rate.

When a data structure as shown in FIG. 7 is used, when the image to be recognized is composed of a plurality of parts, the parts in the plurality of template images are decomposed and recognized, thereby recognizing each of the parts. In order to cope with the variation of the components in the image and to separate the recognition target image of one model into each component,
The misrecognition rate, the misjudgment rate, and the overrecognition rate can be reduced by treating the data structure as a data structure and checking the degree of coincidence of components in the structure.

In the image recognition apparatus shown in FIG. 1 or FIG. 4, the template registration means 2 performs pattern matching of each component to be recognized with a registered template instead of the above-described registration method. When a comparison with the recognition threshold value of each component is performed, only a component having a low matching rate can be registered as a template in a data structure including a component group. That is, instead of a template (a data structure for each model) as shown in FIG.
For example, as shown in FIG. 8, the template is made to have a data structure of each component center (a data structure of imageα, a data structure of imageβ, and a data structure of imageγ), and the data structure of each component is image.
A template may be registered for α, imageβ, and imageγ. In this case, for a part for which a high matching rate is stably obtained (in the example of FIG. 2, im
(for age β), the number of registered templates is small. For example, as shown in FIG. 9, when only noise-biased parts are affected and cannot be obtained stably, many templates are prepared to prevent excessive recognition. .

As described above, when a data structure as shown in FIG. 8 is used as a template, the number of templates composed of each component group having a data structure with one model as a tag is not uniform. By registering a new template as a template only for a part having a low matching rate by each pattern matching, it is possible to reduce the capacity of the image storage device without increasing the number of templates. By reducing the number of recognition candidates, the recognition time can be reduced.

In the image recognition apparatus shown in FIG. 1 or FIG. 4, the template registration means 2 matches the recognition target image and the template when the matching rate of the result of pattern matching is close to other recognition target candidates. The result is prepared as a template, and the product identification unit 3 performs matching of the recognition target itself and matching of the image after matching of the recognition target itself in the subsequent pattern matching, and the recognition target having a similar shape due to the matching of the recognition target itself. After the identification from the group, the matching of the image after the matching of the recognition target itself may be performed for further confirmation.

For example, assuming that FIG. 10 is a recognition target image for a certain model, FIG. 11 is a template for the model, and FIG. 12 is a template for a different model, first, the recognition target image of FIG. 11, FIG.
And each of the templates is subjected to pattern matching.
Note that these pattern matchings are performed on a binarized image, and the matching is performed by setting an object in the image to 1 (black in the figure) and setting a background to 0 (white in the figure), and forming an exclusive OR (ex-OR) image. This is done by taking a match (leaving the unmatched part as 1) and then determining the degree of match. Specifically, assuming that the number of pixels of the image to be recognized is u, the number of pixels of the template image is v, and the number of pixels of the image after matching is w, the degree of coincidence x is obtained by the following equation.

[0042]

X = 100 × (u + v−w) / (u + v) “%”

In Equation 1, for similar images, (u +
v) >> w, and a large S / N cannot be obtained. If the matching degree is low, the result image obtained by pattern matching between the recognition target image and the template, that is, the result image in FIG. Register more,
The result image of another model, that is, the result image of FIG. 14 is matched with the template (the result image of FIG. 13), and the matching degree is obtained from the above equation to perform recognition (the image after the matching is obtained). The used confirmation routine can be performed).

For example, when the optical distance between the production line and the image pickup means 1 changes over time or the like, or when the product is rotated around an axis perpendicular to the optical system of the product itself, the matching rate of pattern matching decreases. Since there is no great difference in the logo of a product or the like in the degree of the difference in specifications, there is no difference in the matching rate with other recognition targets. Therefore, by using the image after the matching as shown in FIG. 13 as a template, it is possible to perform pattern matching with the S / N improved with the matching portion removed. That is, in pattern matching with a similar image, there are many unmatched portions in the image, but there are many matched portions and the S / N is poor. Therefore, it is possible to improve the S / N by preparing a template of a mismatched portion that would not match as a confirmation routine. Also, in the case of over-recognition where a certain mismatched part is always obtained due to changes over time, etc., a pattern matching method is used that detects the correct shift using the repetition error of the image of the mismatched part as a template and performs a pattern matching that is highly resistant to highly accurate changes over time as a confirmation routine. Becomes possible.

Further, when using the image after the matching as a template as described above to perform the pattern matching with the S / N improved by removing the matching part, the template registration means 2 performs matching of the recognition target itself. The matching rate of subsequent image matching detects the frequency at which continuously high values are obtained, and when the frequency exceeds a certain value, the input image to be recognized is registered as a new template. The template may be deleted.

For example, if the recognition target image is as shown in FIG. 10 and the template for the model is that of FIG. 11, the pattern matching result image shown in FIG. If it is obtained, delete the thing of FIG. 11 from the template image,
The recognition target image shown in FIG. 10 may be registered as a template.

As described above, when the matching rate at the time of pattern matching using an unmatched part as a template is high and the frequency is high, the image to be recognized, which is the current input image, is different from the initially registered template. It is determined that the image is obtained as a template, and this is newly registered as a template, whereby the confirmation routine using the image after matching as described above can be omitted, and the recognition speed can be increased. Further, by deleting the template used so far, the capacity of the storage device can be deleted. Further, such a template refresh can be automatically performed. The automatic updating of the template may be performed by limiting the frequency.

When the template registration means 2 and the product identification means 3 obtain a matching image by executing pattern matching, the template registration means 2 performs labeling of the matching image, detects the feature amount of each component in the matching image, and detects the feature amount in the matching image. By comparing the feature amounts of each part in the recognition target image, the parts in the recognition target image are selected as new recognition target parts to create a new data structure consisting of a part group, and only the selected parts , Pattern matching may be performed on the recognition target parts in the images of the recognition target group having a similar shape.

For example, the image to be recognized is shown in FIG. 10, and a matching image with a new object to be recognized as shown in FIG. 12 is shown in FIG. 14, labeling is performed in the resulting image, and the area and the center of gravity are detected. It is possible to determine which component in the new recognition target is a different component from these detected values, and create a data structure using the new model name as a tag. Note that parts having a high matching rate that are not detected in the above procedure can be shared as they are, and only the detected parts can be registered as members of a new data structure.

FIG. 15 shows an example of a new data structure created in this way. In FIG. 15, the parts detected in the above procedure are assumed to be image1′γ, image1α and image1′α are image file names of the same part, and image1β and image1 ′.
By registering the image file name of the same part as β, and making it possible to refer to the image of the same part in both cases, it is possible to delete a huge amount of image files.

As described above, it is determined which component in the recognition target has a low matching rate from the result image obtained by the pattern matching, and only the component having a low matching rate is recognized from the template of the data structure to obtain a high matching rate. If the matching rate cannot be obtained, the matching part of the logo for each model is shared by creating a new data structure with the product name (B in the example of FIG. 15) of the current recognition target as a tag. Templates can be created and the capacity of the storage device can be overwhelmingly deleted. In addition, in the identification between models having shared parts, matching with improved S / N is realized by performing recognition only on non-shared parts. It becomes possible.

The template registration means 2 adds an image after matching to a data structure as a template for a part having a low matching rate of pattern matching for each part, and an image of a recognition target group having a similar shape. The components to be recognized may be selected and the number of template candidates may be reduced.

For example, when the recognition target product (logo) is only the two models shown in FIGS. 10 and 12, image1α, image1′α, and image1β are also used to improve the S / N ratio.
Since it is better not to perform pattern matching on image1 and image1'β, only image1γ and image1'γ, which are non-common parts, are used for pattern matching as templates for both, and the common part is included only for a new recognition target product. It is also possible to limit the templates used for pattern matching, such as detecting a common part using a template.

As described above, in the recognition of the recognition target having the similar shape, the common part is deleted from the template structure to reduce the number of recognition target candidates, thereby improving the S / N at a higher speed and further storing. The capacity of the device can be deleted.

The template registration means 2 captures a printed logo on the surface of the product conveyed on the production line by the image capturing means 1, and performs pattern matching between the captured logo image as a recognition target image and the template image. If the recognition target image and the template model match, register the mismatched part between the printed logo image and the template image as the deletion area, but if the recognition target image and the template model do not match Register the matching part between the printed logo image and the template image as a deletion area, and delete both of the deletion areas from the recognition target image and the template image before performing pattern matching, as a deletion area specific to the model to be recognized. It may be.

For example, when barcode information is used for production management, it is important to match the barcode information with the product model, and the recognition target image and the template image can be limited.
If it can be assumed that the models match, a non-matching part as shown in FIG.
In the preprocessing stage, the image is deleted from the recognition target image and the template image, and pattern matching is performed. If the recognition target image and the template image can be limited and it can be assumed that the models do not match, the matching part between the two as shown in FIG. Delete from the image and the template image and perform pattern matching.

As described above, the pattern matching is performed by providing the deletion area in the recognition target image and deleting the common part having a small characteristic difference and the part which is difficult to recognize and match due to the minute and complicated shape. Improve S / N, matching rate,
It is possible to reduce the erroneous recognition rate, erroneous determination rate, and excessive recognition rate.

The template registering means 2 creates a representative template, registers it first as a first template, and then stores the result of pattern matching between the object to be recognized by the product identifying means 3 and the template. It may be registered as a second template.

More specifically, for example, a logical product (AND) image of a recognition target image and a template image is used as a representative template, and A
The ND image is deleted, pattern matching is performed using the two images after deletion as a new recognition target image and a template image, and the degree of coincidence can be evaluated using Equation 1.

As described above, by registering a typical template, performing pattern matching with the template, and using the obtained result image as a true template, only the mismatched portion can be used for pattern matching. , S / N can be improved, and the capacity of the storage device for registering the template can be significantly reduced.

If a high degree of matching cannot be obtained as a result of the pattern matching a plurality of times, the template registration means 2 performs labeling in each pattern in a plurality of patterns, performs matching for each component, and performs After performing the optimal matching, an image obtained by synthesizing them may be registered as one template, and the above operation may be performed periodically.

For example, when a high matching rate is obtained by pattern matching using the template shown in FIG. 3, the number of times of recognition can be reduced. However, when a high matching rate cannot be obtained, FIG. The pattern matching is performed for each of the components described above, and those having a high matching rate obtained for each component can be combined and added to the template group shown in FIG. 3 as one template.

As described above, pattern matching is performed on registered templates, and when a high matching rate cannot be obtained for all registered templates, pattern matching is performed for each component in those templates. By creating and registering a template by combining parts with a high matching rate, it is possible to create a basic template corresponding to noise that independently affects each part in the recognition target image, and a high matching rate as a whole template Is obtained. That is, a matching rate that cannot be obtained for the entire template can be obtained by creating an optimum template for each component for each component and combining them, thereby improving the matching rate of the entire template, An erroneous recognition rate, an erroneous determination rate, and an overrecognition rate can be reduced. Thereby, it is possible to cope with the influence depending on the direction of the environmental lighting, a change in the inclination of the recognition target product, and the like.

The template registration means 2 registers, for example, an image of a common part of the recognition target images of all models as a representative template, and registers a new recognition target image and a representative template for a new recognition target image. An image of a common part with the image is generated as a new representative template image, and a result image obtained by pattern matching between the recognition target image and the template image is registered as a unique template image for the recognition target. Alternatively, the product may be identified by pattern-matching an image resulting from pattern matching with a representative template with a unique template.

When the template registration means 2 has the above configuration, a logical product (AND) image of the recognition target image and the template image is automatically updated as a representative template every time. In addition, an image obtained as a result of pattern matching by exclusive OR (ex-OR) between the recognition target image and the representative template is registered as a template unique to the recognition target product. At this time, when the representative template is updated, the unique template is also updated each time. At the time of recognition by the product identification means 3, pattern matching using the representative template and pattern matching between the resulting image and the unique template are performed. Equation 1 can be used to evaluate the matching rate.

As described above, the common part of the recognition target images of all the models is used as a representative template, and this representative template can be automatically updated for a new model. By acquiring the part that does not match the image to be recognized by pattern matching,
Image recognition can be facilitated.

Further, when the image capturing means 1 obtains a recognition target image at a correct position and detects a recognition target in the image by pattern matching, the template registration means 2 deletes the detected recognition target from the image and performs the recognition. The image obtained by subtracting the basic image from the image captured by the image capturing unit 1 may be obtained as a template, using the image from which the target has been deleted as the basic image.

For example, when the recognition target has a shadow as shown in FIG. 17, for the new recognition target, the recognition target image is obtained, and is subjected to pattern matching with the recognition target image as a template. OR), an image having only the background shadow as shown in FIG. 18 is obtained. This is added to the data structure shown in FIG. Noise can be removed by taking an exclusive OR (ex-OR) with the image.

As described above, when a new model is added to the recognition target, pattern matching between the recognition target template and the recognition target image is performed, the recognition target image is detected, and this is deleted from the input image to remove the second model. As a template, the second template is deleted from the input image at an early stage of the recognition of the product identification means, thereby eliminating the influence of shadows due to the surface shape of the product and the likelihood of erroneous recognition, erroneous determination, and excessive recognition. Can be reduced.

FIG. 19 is a diagram showing an example of a hardware configuration of the image recognition apparatus shown in FIG. 1 or FIG. Referring to FIG. 19, this image recognition device is realized by, for example, a personal computer or the like, and includes a CPU 21 that controls the entire system,
ROM 22 storing the control program of U21 and the like
And RA used as a work area or the like of the CPU 21.
M23, an imaging device 24 for imaging a printed logo on the surface of a product conveyed on the production line, and a recognition target image file 25 in which a recognition target image captured by the imaging device 24 is stored.
And a template registration file 26 in which a template is registered, and a result output device (for example, a display or a printer) 27 for outputting information on the result of product identification.

Here, the image pickup device 24 corresponds to the image pickup means 1 of FIG. 1 or FIG. Also, the CPU 21
Are the template registration means 2, the product identification means 3,
It has the function of the presentation means 4.

The functions of the CPU 21 as the template registering means 2, the product identifying means 3, the presenting means 4, and the like are implemented, for example, in the form of a software package (specifically, an information recording medium such as a CD-ROM). Can be provided.

In other words, the image recognition apparatus of the present invention causes a general-purpose computer system having an imaging device, a display, and the like to read a program recorded on an information recording medium such as a CD-ROM, and It can also be implemented in an apparatus configuration in which a microprocessor performs template registration processing and product identification processing. In this case, the program for executing the template registration process and the product identification process of the present invention (that is, the program used in the hardware system) is provided in a state recorded on a medium. The information recording medium on which the program or the like is recorded is not limited to a CD-ROM, but may be a ROM, a RAM, a flexible disk, a memory card, or the like. The program recorded on the medium is installed in a storage device incorporated in the hardware system, for example, a hard disk device, so that the program is executed and the template registration unit 2, the product identification unit 3, and the presentation unit 4 are executed. Function can be realized.

The template registration means 2 of the present invention
The program for realizing the functions of the product identification means 3 and the presentation means 4 may be provided not only in the form of a medium but also by communication (for example, by a server).

[0075]

As described above, according to the first aspect of the present invention, an image pickup means for picking up a print logo on the surface of a product conveyed on a production line, and a print picked up by the image pickup means. A template registration unit for registering a logo image as a template, and by performing pattern matching between a printed logo on the surface of the product and a template registered in the template registration unit for a product to be transported on a production line. Product identification means for performing product identification, and presentation means for presenting whether to register a logo image of a product currently in the identification space as a new template when a product identification result by the product identification means does not exceed the recognition threshold. The template registration means can register a plurality of templates for one model and one recognition target image. When a plurality of templates are registered for one model and one recognition target image, the product identification means performs pattern matching with the plurality of templates for a printed logo on the product surface. The influence of noise, false recognition rate, and excessive recognition rate can be reduced, and high-precision image recognition can be performed on a production line where the environment is easily changed. That is, it is possible to perform pattern matching with a reduced erroneous recognition rate, erroneous determination rate, and excessive recognition rate corresponding to an environmental change or the like without specially setting a unique closed optical system.

According to the second aspect of the present invention, an image pickup means for picking up a print logo on the surface of a product conveyed on a production line, and a print logo image picked up by the image pickup means are registered as a template. For template registration means and products after being transported on the production line,
Product identification means for performing product identification by performing pattern matching between the printed logo on the product surface and a template registered in the template registration means; the template registration means includes one model and one recognition target image; Multiple templates can be registered for
If the product identification result by the product identification means does not exceed the recognition threshold, the template registration means registers the logo image of the product currently in the identification space as a new template, and the product identification means is one model, one recognition target. When multiple templates are registered for an image, pattern matching with multiple templates is performed on the printed logo on the product surface, so the effects of noise due to environmental changes, false recognition rate, and excessive recognition rate , And accurate image recognition can be performed on a production line where the environment is easily changed. That is, it is possible to perform pattern matching with a reduced erroneous recognition rate, erroneous determination rate, and excessive recognition rate corresponding to an environmental change or the like without specially setting a unique closed optical system.

According to a third aspect of the present invention, in the image recognition device according to the first or second aspect, the template registering unit is configured to execute the method after the template image is captured by the image capturing unit. Then, the recognition target is divided into each component by performing labeling of the captured image, the feature amount of each component is detected, and a data structure including a component group is formed for each recognition target in which the feature amount of each component is detected. A template is prepared and a plurality of templates are prepared for each part in one model and one recognition target image. In this case, the product identification unit performs processing for each part in the product captured image. Perform pattern matching with the plurality of templates (pattern matching for each part of the template that has a structure structure for each part in the recognition target image), Since the model name is output as a tag, even if the influence from the environment is not constant in position, it is possible to respond independently to each component and perform pattern matching with reduced false recognition rate, false recognition rate, and excessive recognition rate. Becomes

According to a fourth aspect of the present invention, in the image recognition apparatus according to the first or second aspect, the template registering means stores the pattern of each component to be recognized with a registered template. When performing matching and comparing with the recognition threshold of each component, only the component having a low matching rate is registered as a template in the data structure including the component group, and the data structure is registered when the template is registered. The number of templates for each component in the template is not uniform, and whether or not registration is possible is determined based on the matching rate of each component. Therefore, it is possible to reduce the image storage capacity and the number of matching candidates, and to eliminate unnecessary routines. Pattern matching becomes possible. That is, the number of registered templates can be suppressed, and the image storage capacity can be reduced.

According to a fifth aspect of the present invention, in the image recognition apparatus according to the first or second aspect, the template registration unit determines a matching rate of a result of pattern matching between the recognition target image and the template. Is close to another recognition target candidate, a result of the matching is prepared as a template, and the product identification unit performs subsequent pattern matching by matching the recognition target itself and matching the image after the recognition target itself is matched. After performing recognition and matching from the recognition target group whose shape is similar by matching the recognition target itself, for further confirmation,
By performing matching of the image after matching of the recognition target itself (using the image after matching as a template), the S / N in which the matching part is removed is obtained.
Can be performed with improved pattern matching.

According to a sixth aspect of the present invention, in the image recognition apparatus of the fifth aspect, the template registering means has a continuously high matching rate of matching of the image after matching of the recognition target itself. When the frequency at which the value is obtained is detected and the frequency exceeds a certain value, the input image to be recognized is registered as a new template, while the previous template is deleted, so that the recognition speed can be increased. Become. Further, by deleting the template used so far, the capacity of the storage device can be deleted. Further, the template can be automatically refreshed. In other words, if the image to be recognized as an image to be recognized has many portions that do not match the model of the model, and if pattern matching using the matching image in the confirmation routine is frequently obtained with a high degree of matching, changes over time Assuming that there has been an irreversible transition in the environment due to the influence of the above, the template can be registered at the current recognition target image and the template can be updated at any time. In other words, it is possible to reduce an erroneous recognition rate and an overrecognition rate due to irreversible temporal changes such as a displacement of the system in a certain direction.

According to a seventh aspect of the present invention, in the image recognition device according to the third or fifth aspect, the product identification means obtains a matching image by executing pattern matching. Labeling is performed to detect the feature amount of each component in the matching image, and by comparing the feature amounts of each component in the matching image and the recognition target image, the components in the recognition target image are selected as new recognition target components. Create a new data structure consisting of a group of parts, perform pattern matching only on the selected parts, and select the recognition target parts in the images of the recognition target group with similar shapes. By having a common part of the template without registering the entire recognition target, the capacity of the storage device can be reduced. In addition, in recognition between recognition targets having a shared portion, only recognition of a non-shared portion is performed, thereby enabling recognition with an improved S / N.

According to an eighth aspect of the present invention, in the image recognition apparatus according to the fifth or seventh aspect, the template registering means is provided for a component having a low matching rate of pattern matching for each component. While adding the image after the matching as a template to the data structure, the recognition target parts in the image of the recognition target group having similar shapes are selected, and the number of template candidates is reduced,
In the recognition between similar recognition targets, by deleting the common part from the template, the capacity of the storage device can be reduced, the S / N can be improved, and the speed can be increased. That is, the matching can be performed at high speed by deleting the capacity of the image storage device of the template and improving the S / N.

According to the ninth aspect of the present invention, in the image recognition apparatus according to the first aspect, the template registration unit captures a print logo on a surface of a product conveyed on a production line by the image capture unit. Then, pattern matching is performed between the captured logo image as the recognition target image and the template image, and when the model of the recognition target image and the template match, the mismatched portion between the printing logo image and the template image is determined. If the recognition target image does not match the model of the template while registering it as the deletion area, the matching part between the print logo image and the template image is registered as the deletion area, and both deletion areas are unique to the recognition target model. By deleting from the recognition target image and the template image before performing pattern matching, To improve the S / N and reduce the false recognition rate, false determination rate, and excessive recognition rate by removing parts with few features in the recognition target image and parts where a high matching rate cannot be obtained in pattern matching). Can be.

According to a tenth aspect of the present invention, in the image recognition apparatus according to the first aspect, the template registering means first creates a representative template and registers it as a first template. The result of pattern matching between the recognition target and the template by the product identification means is registered as a second template, and a template is created so that a matching part of a similar recognition target can be deleted as much as possible. By detecting a mismatched portion by pattern matching with the template and performing pattern matching again on the mismatched portion, the S / N can be improved, and the capacity of the storage device for registering the template can be significantly reduced. That is, by performing pattern matching twice, S
/ N can be recognized with high accuracy.

According to the eleventh aspect of the present invention, in the image recognition apparatus according to the first, second or seventh aspect, the template registration means may obtain a high matching score as a result of a plurality of pattern matchings. Is not obtained, labeling is performed within each pattern in a plurality of patterns, matching is performed for each component, and after performing optimal matching for each component, an image obtained by combining them is registered as one template. A basic template corresponding to noise or the like that independently affects each component in the recognition target image can be created, and a high matching rate can be obtained for the entire template.

According to the twelfth aspect of the present invention, in the image recognition apparatus according to the tenth aspect, the template registration unit registers an image of a common part of the recognition target images of all the models as a representative template. For a new recognition target image, an image of a common part between the new recognition target image and the representative template image is generated as a new representative template image. The result image obtained by pattern matching is registered as a unique template image for the object to be recognized, and the product identification unit performs pattern matching of the result image obtained by performing pattern matching with the representative template with the unique template. , By removing the common part for all recognition target images, / N it is possible to automatically generate a representative template to improve. That is,
Pattern matching with improved S / N by removing a part common to all recognition targets can be performed.

According to a thirteenth aspect of the present invention, in the image recognition device according to the first aspect, the template registration unit acquires the recognition target image at a correct position by the image pickup unit, and performs image matching by pattern matching. When the recognition target in the image is detected, the detected recognition target is deleted from the image, and the image obtained by subtracting the basic image from the image captured by the image capturing unit is obtained as a template, using the image from which the recognition target has been deleted as a basic image. Although different from the recognition target image, noises such as environmental illumination and shadows formed by the surface shape of the product which can cause erroneous recognition and erroneous determination can be removed, and the accuracy of pattern matching can be improved. That is, it is possible to reduce a false recognition rate by setting a target other than a shadow due to the surface shape of the recognition target product or a recognition target component as a recognition target.

[Brief description of the drawings]

FIG. 1 is a diagram showing a configuration example of an image recognition device according to the present invention.

FIG. 2 is a flowchart illustrating a processing operation of the image recognition device of FIG. 1;

FIG. 3 is a diagram illustrating a registration example of a template.

FIG. 4 is a diagram showing another configuration example of the image recognition device according to the present invention.

FIG. 5 is a diagram showing an example of a logo attached to a product.

FIG. 6 is a diagram showing an example of a logo attached to a product.

FIG. 7 is a diagram showing a registration example of a template.

FIG. 8 is a diagram illustrating a registration example of a template.

FIG. 9 is a diagram illustrating an example of a captured logo image.

FIG. 10 is a diagram illustrating an example of a recognition target image for a certain model.

FIG. 11 is a diagram showing a template for the same model as in FIG. 10;

FIG. 12 is a diagram showing a template for a model different from FIG. 10;

13 is a diagram showing a result image as a result of pattern matching between the recognition target image in FIG. 10 and the template in FIG. 11;

14 is a diagram showing a result image as a result of pattern matching between the recognition target image in FIG. 10 and the template in FIG. 12;

FIG. 15 is a diagram illustrating a registration example of a template.

FIG. 16 is a diagram illustrating an example of a template.

FIG. 17 is a diagram illustrating an example of a recognition target.

18 is a diagram illustrating an example of a template extracted from the recognition target in FIG. 17;

19 is a diagram illustrating an example of a hardware configuration of the image recognition device of FIG. 1 or FIG. 4;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Image imaging means 2 Template registration means 3 Product identification means 4 Presentation means 21 CPU 22 ROM 23 RAM 24 Imaging device 25 Recognition image file 26 Template registration file 27 Result output device

Claims (13)

[Claims] 1. An image pickup means for picking up a printed logo on a surface of a product conveyed on a production line, a template registering means for registering a printed logo image picked up by the image pickup means as a template, and a conveyed on the production line. For the subsequent products, product identification means for performing product identification by performing pattern matching between the printed logo on the product surface and the template registered in the template registration means, and the product identification result by the product identification means Presenting means for presenting whether the logo image of the product currently in the identification space is to be registered as a new template when the recognition threshold is not exceeded, wherein the template registering means has one model, one recognition target image; And a plurality of templates can be registered, and the product identification means is provided for one model and one recognition target image. An image recognition apparatus characterized in that when a plurality of templates are registered, pattern matching with a plurality of templates is performed on a printed logo on a product surface. 2. An image pickup means for picking up a printed logo on a surface of a product conveyed on a production line, a template registering means for registering a printed logo image picked up by the image pickup means as a template, and a conveyed on the production line. And product identification means for performing product matching by performing pattern matching between the printed logo on the product surface and the template registered in the template registration means for the subsequent products. Can register a plurality of templates for one model and one recognition target image. If the product identification result by the product identification means does not exceed the recognition threshold, the template registration means Is registered as a new template, and the product identification means matches one model and one recognition target image. An image recognition apparatus characterized in that when a plurality of templates are registered, pattern matching with a plurality of templates is performed on a printed logo on a product surface. 3. The image recognition apparatus according to claim 1, wherein the template registration unit performs labeling of the captured image after the image for template registration is captured by the image capturing unit. The object is divided into components, the feature amount of each component is detected, and a data structure including a component group is created as a template for each recognition target in which the feature amount of each component is detected. A plurality of templates are prepared for each component in the target image. In this case, the product identification unit performs pattern matching with the plurality of templates for each component in the product captured image. An image recognition device characterized by the above-mentioned. 4. The image recognition apparatus according to claim 1, wherein the template registration unit performs pattern matching of each component to be recognized with a template already registered, and performs a pattern matching with a recognition threshold of each component. An image recognition apparatus characterized in that, when a comparison is made, only a component having a low matching rate is registered as a template in a data structure composed of a component group. 5. The image recognition apparatus according to claim 1, wherein the template registering unit determines that a matching rate of a result of pattern matching between the recognition target image and the template is close to another recognition target candidate. The result of matching is prepared as a template, and the product identification unit performs matching of the recognition target itself and matching of the image after matching of the recognition target itself in the subsequent pattern matching, and the shape is determined by matching of the recognition target itself. An image recognition apparatus, comprising: performing identification from a group of similar recognition targets, and then, for further confirmation, matching of the image after matching of the recognition target itself. 6. The image recognition device according to claim 5, wherein
The template registration unit detects a frequency at which the matching rate of matching of the image after matching of the recognition target itself continuously obtains a high value, and when the frequency exceeds a certain value, the input image of the recognition target is detected. An image recognizing apparatus characterized in that a new template is registered while a previous template is deleted. 7. The image recognition device according to claim 3, wherein the product identification unit performs pattern matching to obtain a matching image, performs labeling of the matching image, and performs labeling of the matching image. A new data structure consisting of a group of parts is detected by detecting the feature quantity, comparing the feature quantities of each part in the matching image and the recognition target image, and sorting out the parts in the recognition target image as new recognition target parts. And performing pattern matching only on the selected components to select the recognition target components in the images of the recognition target group having similar shapes. 8. The image recognition apparatus according to claim 5, wherein said template registering means uses, as a template, an image after matching for a part having a low matching rate of pattern matching for each part as a template. An image recognition apparatus for selecting recognition target components in an image of a recognition target group having a similar shape while reducing the number of template candidates. 9. The image recognition apparatus according to claim 1, wherein
The template registration unit captures a printed logo on the surface of a product conveyed on a production line by the image capturing unit, performs pattern matching between the captured logo image as a captured recognition target image and the template image, and When the model of the image and the template match, the part where the print logo image and the template image do not match is registered as the deletion area, while when the model of the recognition target image and the template do not match, the print logo A matching part between the image and the template image is registered as a deletion area, and both of the deletion areas are set as deletion areas specific to the recognition target model, and are deleted from the recognition target image and the template image before performing pattern matching. Image recognition device. 10. The image recognition apparatus according to claim 1, wherein said template registering means first creates a representative template, registers it as a first template, and stores a template to be recognized by said product identifying means and a template. An image recognition device for registering a result of the pattern matching of (1) as a second template. 11. The image recognition apparatus according to claim 1, wherein the template registration unit determines each of a plurality of patterns when a high degree of matching is not obtained as a result of a plurality of pattern matchings. An image recognition apparatus characterized by performing labeling in a pattern, performing matching for each component, performing optimal matching for each component, and registering an image obtained by combining them as one template. 12. The image recognition apparatus according to claim 10, wherein the template registration unit registers an image of a common part of the recognition target images of all models as a representative template, and An image of a common part between the new recognition target image and the representative template image is generated as a new representative template image, and a result image obtained by pattern matching between the recognition target image and the template image is recognized. An image recognizing apparatus, wherein the image is registered as a unique template image for an object, and wherein the product identification unit performs pattern matching with the unique template after performing pattern matching with the representative template. 13. The image recognition device according to claim 1, wherein the template registration unit detects when the image capturing unit acquires a recognition target image at a correct position and detects a recognition target in the image by pattern matching. An image recognition apparatus wherein a recognition target is deleted from an image, and an image obtained by subtracting the basic image from an image captured by an image capturing unit is obtained as a template, using the image from which the recognition target has been deleted as a basic image.