TWM563621U - A system for recognizing bread - Google Patents

Abstract

Disclosed is a system for recognizing bread. The system includes a database, an image capture module, an image processing module and an image comparison module. The database has a plurality of image models and corresponding bread information. The image capture module captures a bread image. The image processing module obtains a divided image by processing the bread image. The image comparison module compares the divided image with the image models to get the bread model which is associated with the bread and corresponding bread information from the database.

Description

Bread identification system

This creation is about an image recognition technology, in particular, about a bread identification system.

In the current merchant sales process, the product type and price are judged manually in the early stage, and the amount is manually input to settle the bill. For example, the merchant uses different price tags to be placed in front of different types of goods, and the settlement is required by the store staff. The appearance determines the type of the product, and then the amount is settled by the price of the product. However, with the advancement of technology, the process has been simplified by sensing the bar code. The merchant uses different barcodes for different products, and uses an infrared barcode scanner to check at the checkout. Read the price of the product to settle the amount. Regardless of the sales process, there are still some shortcomings in the above method.

Manually speaking, when the clerk judges the type of the product, the product type may be misjudged, or the amount corresponding to the product may be misjudged. If the merchant has a preferential activity, some products have a discount, and the clerk may forget to cause the price of the product. The above total situation may result in an incorrect checkout amount. If the bar code is used to identify the product, when scanning the bar code with the bar code machine, there may be light problems or uneven scanning surface, which may cause the scanning to fail. Moreover, the bar code on the product is not applicable to certain goods, such as bread. Store Bread, fresh bread is mostly placed barely, it is impossible to put a bar code, that is, the sales process still has to return to the manual processing.

As can be seen from the above, at present, the sales process of bread type products, especially product identification and amount settlement, still needs to be performed by manpower, which is not only time-consuming but also has the risk of identifying errors or settlement errors. To overcome this problem, the goods should be correctly identified. As well as the exact identification of the price of the goods, it is necessary to find a mechanism for identifying bread, which will become the goal of the people in the technical field.

The purpose of this creation is to propose a system for performing image recognition on different types of bread. The image identification mechanism can be used to determine the type of bread and obtain the corresponding information of the bread. This method is more conducive to the identification of the goods displayed on the bare display.

The present invention is a bread identification system comprising: a database, an image capture module, an image processing module, and an image comparison module, wherein the database prestores a plurality of image models and bread corresponding to the plurality of image models Information, the image capture module is configured to perform image capture of a bread to be tested to obtain a bread image of the bread to be tested, and the image processing module is configured to receive the bread image from the image capture module Performing image processing of the bread image, obtaining a segmented image of the bread to be tested from the bread image, and performing an alignment of the segmented image with the plurality of image models to obtain a corresponding segmented image The image model, based on the image model obtained after the comparison, obtains information about the bread of the bread to be tested from the database.

In an embodiment, before the image processing module obtains the segmented image, the method further includes performing the grayscale processing of the bread image to obtain the grayscale image of the bread image from the full color to the black and white image. .

Secondly, the image processing module further performs feature point detection of the grayscale image to obtain feature points of the bread image.

Then, the image processing module further performs the background determination of the bread image according to the feature point, so that the bread covered by the bread to be tested is the foreground portion.

Finally, the image processing module performs image segmentation on the bread image according to the foreground portion, so that the foreground portion obtained after the segmentation is the segmented image.

In another embodiment, when the image matching module does not obtain an image model corresponding to the divided image, the image matching module makes the segmented image of the unobtained image model a new image model, and adds The new image model is to the database.

In another embodiment, the bread identification system further includes a machine learning module. When the machine learning module is in the non-identification stage, the image capture module intercepts a plurality of the same bread to obtain a plurality of image data. The convolutional neural network performs machine learning based on the plurality of image data to obtain one of the plurality of image models.

In another embodiment, the bread identification system further includes a display for displaying the bread image captured by the image capturing module and the bread information obtained after the image matching module is aligned.

In addition, when the image information obtained by the image matching module is verified as being wrong, the image capturing module, the image processing module, and the image are obtained. The image recognition of the bread to be tested is re-executed like the comparison module. On the contrary, if the bread information obtained by the image matching module is verified as being correct, the bread information is transmitted to the sales terminal system.

Compared with the current manual processing of the bread sales process, the bread identification system proposed by the present invention can perform image processing on the bread image after performing image processing on the bread image, and then compare with the pre-stored plurality of image models. Identify the type of bread and obtain information about the bread of the bread, such as the price of the bread, whether it is a preferential product, etc. After that, the bread identification system can transmit the identification result to the sales terminal system of the back end, and then perform the settlement of the amount, so through the creation The bread identification system can effectively reduce the time and labor costs, and can reduce the occurrence of human error, not only help to improve the bread sales process, but also make the checkout process highly efficient and accurate.

1‧‧‧Bread Identification System

11‧‧‧Database

111‧‧‧Image Model

1111‧‧‧Image Model A

1112‧‧‧Image Model B

1113‧‧‧Image Model C

12‧‧‧Image capture module

13‧‧‧Image Processing Module

131‧‧‧Split image

14‧‧‧Image comparison module

15‧‧‧ machine learning module

16‧‧‧ display

2‧‧‧Sale terminal system

S31~S35‧‧‧Steps

S41~47‧‧‧ Process

S51~55‧‧‧ Process

1 is a system architecture diagram of the bread identification system of the present invention; FIG. 2 is a system architecture diagram of another embodiment of the bread identification system of the present invention; and FIG. 3 is a image processing module of the present invention performing bread image processing. Step diagram; Figure 4 is a flow chart of the bread identification system of the present invention in the general identification stage; Figure 5 is a flow chart of the bread identification system of the present invention in the machine learning stage; and Figure 6 is an image comparison of the creation of the present invention. After the execution of the proportion judgment Figure.

The technical content of the present invention will be described below by way of specific embodiments, and those skilled in the art can clearly understand the advantages and effects of the present invention by the contents disclosed in the present specification. However, the creation can also be implemented or applied by other different embodiments.

In view of the fact that bread sales need to be judged manually by the clerk, it is easy to have problems such as misjudgment or wrong amount. If the barcode reading method is used, it may not be suitable for the bread that is normally placed, so the creation is for bread. The sales method proposes to identify the type of bread through image recognition, thereby improving efficiency and accuracy.

Please refer to FIG. 1 , which is a system architecture diagram of the bread identification system of the present invention. As shown in the figure, the bread identification system 1 of the present invention includes a database 11 , an image capturing module 12 , an image processing module 13 and an image. Align the module 14.

The database 11 is configured to pre-store a plurality of image models 111 and bread information corresponding to the plurality of image models 111. The image models 111 can be stored in the database 11 through machine learning, and can be used for subsequent bread-time identification comparison. Of course, other new image models can be gradually added to the database 11 in the future.

Each of the image models 111 described above corresponds to a bread style, and thus each image model 111 has a corresponding bread information. Accordingly, the plurality of image models 111 are different types of bread base images, followed by The image comparison process will judge the style of the bread to be tested. The aforementioned bread information may include any quantifiable or specific Describe the information, such as the name of the bread, the amount of bread, the main ingredients of the bread, the weight of the bread, or whether it is accompanied by a discount.

The image capturing module 12 is configured to perform image capturing of the bread to be tested, thereby obtaining a bread image of the bread to be tested. Specifically, the image capturing module 12 may be a device such as a CCD image sensor or a camera for capturing images, but is not limited thereto.

The image processing module 13 is configured to receive the bread image captured by the image capturing module 12 and perform image processing on the bread image to obtain a segmented image of the bread to be tested from the bread image. In addition, for image processing, the image processing module 13 can perform programs such as grayscale processing, feature point detection, front and back scene or image segmentation on the bread image, and the obtained segmented image can be used for subsequent comparison.

The image comparison module 14 is connected to the database 11 and the image processing module 13. After the image matching module 14 receives the segmented image transmitted from the image processing module 13, the image matching module 14 reads the database 11 Taking a plurality of image models 111 of the memory, that is, performing an alignment of the divided images with the plurality of image models 111, when the comparison is successful, indicating that the divided images correspond to one of the plurality of image models 111 That is, the image model corresponding to the image to be segmented, and based on the image model obtained after the comparison, the bread information about the bread to be tested can be further obtained from the database 11.

In addition, if the image matching module 14 does not obtain the image model corresponding to the divided image, the image matching module 14 can set the divided image of the corresponding image model as a new image model, and add the new image model. New image model to database 11.

As can be seen from the above description, after the bread to be tested is obtained through the image capturing module 12 of the bread identification system 1 , the image is processed and analyzed by the image processing module 13 to obtain the portion of the bread image to be tested. That is, the image is divided, and the complex image is compared with the module 14 to obtain a corresponding image model and bread information. Accordingly, when selling bread in a store, it is possible to determine the bread by image recognition, which not only reduces the misjudgment of human resources, but also helps to improve the speed of the checkout process through identification technology.

Please refer to FIG. 2, which is a system architecture diagram of another embodiment of the bread identification system of the present invention. As shown in the figure, the database 11 , the image capturing module 12 , the image processing module 13 , and the image matching module 14 in the bread identification system 1 of the present embodiment are the same as those described in FIG. 1 . In the present embodiment, the bread identification system 1 of the present invention includes a machine learning module 15 and a display 16.

The machine learning module 15 can capture a plurality of the same type of bread through the image capturing module 12 to obtain a plurality of image data in a non-identification stage, and the plurality of image data can be executed by a Convolutional Neural Networks (CNN). Machine learning, thereby obtaining one of a plurality of image models 111 in the database 11. That is to say, in order to improve the accuracy of the bread identification system 1 in bread identification, it is necessary to obtain a better image of each bread through machine learning of a large number of samples before performing the identification, and can quickly compare in the future judgment. What is the possible image model of the bread to be tested.

The display 16 is configured to display the bread shadow captured by the image capturing module 12 The bread information obtained after the image comparison module 14 is compared. In short, in addition to displaying the bread and the related information of the bread to be tested, the display 16 can also be used for review by the person or the customer through the display of the above contents, so that the image is compared. The bread information obtained by the group 14 is the wrong one after the review, that is, the image may be misplaced, and the image capturing module 12, the image processing module 13, and the image matching module 14 may be re-executed. The image of the bread to be tested is identified to obtain the correct message of the bread to be tested. In addition, when the bread information obtained by the image matching module 14 is corrected, the relevant bread information is transmitted to the sales terminal system 2, To perform the final checkout process.

The display 16 can be designed as a touch screen or with a keyboard mouse control device to confirm the final identification result, such as identification failure to be re-identified, unrecognizable, new image model added or checkout.

Please refer to FIG. 3, which is a step diagram of performing bread image processing for the image processing module of the present invention. As shown in the figure, the steps S31 to S35 may be a preferred implementation step of the image processing, but the video processing module 13 receives the image capturing module. Image processing performed after 12 captured bread images.

In step S31, grayscale processing is performed on the bread image. This step converts the full color into black and white to obtain a grayscale image of the bread image.

In step S32, feature point detection is performed on the grayscale image of the bread image to obtain feature points of the bread image. The detection of the feature points is obtained from the bread image, and the shape of the bread can be derived therefrom.

In step S33, the foreground image determination before the bread image is performed according to the feature point, so that the coverage range of the bread to be tested is determined as the foreground portion. In short, this step is to find out the range of the bread to be tested in the bread image through the feature points, and set the range to be the foreground portion, which is advantageous for the subsequent segmentation of the image.

In step S34, image segmentation is performed on the bread image according to the foreground portion, so that the foreground portion obtained after the segmentation is the segmented image. This step illustrates how the above accurately identifies the extent of the bread to be tested and segments it for subsequent comparison.

In step S35, image filtering of the divided image is performed. The segmented image may be mixed with noise. The noise may be due to light or angle caused by other unnecessary image information, so the noise filtering can be performed on the segmented image to make the segmented image more Optimize to avoid misjudgment when subsequent comparisons are performed.

Of course, in order to obtain an excellent bread image, the image capturing module 12 of FIG. 1 can assist the appropriate light source when performing image capturing, thereby improving the quality of the bread image.

Referring again to FIG. 2, as previously described, the bread identification system of the present invention can establish the image model 111 through the machine learning module 15. In short, the bread identification system can perform two processes, namely performing image recognition. The general identification phase and the machine learning phase in which the image model is pre-established.

Please refer to Figure 4, which is a flow chart of the bread identification system of the creation in the general identification stage. As shown in the figure, in the process S41, the image capturing module captures a bread image. In process S42, the image processing module executes the The image processing of the bread image, as in the third step, includes the bread image for grayscale processing, feature point detection, establishing front and back scenes, and image segmentation to obtain a segmented image, and then filtering the segmented image with noise. In the process S43, the image comparison module executes the comparison of the segmented image and the image model of the bread image. If the comparison is successful, the process proceeds to the process S44, and the corresponding image model is obtained, that is, the bread to be tested is successfully identified. .

If the corresponding image model is not obtained after the comparison, the process proceeds to the process S45, and the next step is the process S46, the image comparison module causes the bread image of the corresponding image model to be a new image model, and finally, the process proceeds. S47, the new image model is stored in the database, which can be used for comparison of the subsequent bread identification system.

Please refer to Figure 5, which is a flow chart of the bread identification system of the creation in the machine learning stage. As shown in the figure, the process can be started by a machine learning module. In the process S51, a plurality of bread images are captured by the image capturing module, and the plurality of bread images refer to a certain number of the same bread. In the process S52, image processing of the plurality of bread images is performed by the image processing module. In the process S53, the machine learning module performs machine learning through the convolutional neural network, and the process is to repeatedly accumulate the identification of the plurality of bread images of the same bread, and to find out the bread from the repeated accumulation learning process. Better image. In the process S54, an image model is obtained. In the process S55, the image model is stored in the database.

In addition to creating an image model, the machine learning module can improve the quality of the image model through the machine learning module, thereby reducing the misjudgment. In particular, because the same bread cannot be 100% identical, to avoid Due to the misunderstanding caused by the slight difference in appearance of the same bread, machine learning can be performed through the machine learning module to establish an image model of the allowable error value range.

Please refer to Fig. 6, which is a schematic diagram of the execution ratio comparison after the image comparison of the creation. As shown in the figure and in conjunction with FIG. 2, the segmentation image 131 is generated by the image processing module 13, and the image model A 1111, the image model B 1112, and the image model C 1113 are plurals in the database 11 shown in FIG. Image model 111. When the image matching module 14 performs the comparison between the divided image 131 of the bread to be tested and the plurality of image models 111, the percentage of similarity between each image model and the segmented image 131 is respectively a% of the image model A 1111, and the image is Model B 1112 is b%, image model C 1113 is c%, and a plus b plus c is 100.

More specifically, the image matching module 14 first compares the image images according to the segmented image 131 and obtains the similarity of each image model, and then shaves the image model below a certain value (ie, obviously does not look after the comparison). ), wherein the image model A, the image model B, and the image model C may have a certain degree of similarity. Then, the image model A, the image model B, and the image model C obtain the respective proportions of the percentage values according to the degree of similarity, that is, the aforementioned a% , b% and c%, the image model with the highest percentage value is regarded as the image model corresponding to the bread to be tested, according to which the possible image model can be quickly and accurately found.

If the image model A and the image model B have the highest percentage values and the values are very close, the image matching module 14 transmits the image model A and the image model B together with the obtained bread image to the display 16 for reconfirmation. In this case, of course, if one of the image models (A or B) is selected, the machine learning module 15 can also learn about the image.

In summary, the bread identification system proposed by the present invention can convert the bread sales checkout type, which requires a lot of manpower and time, into an automated operation through the introduction of image recognition, thereby reducing the possibility of human error determination. The identification process of efficiency and high precision, in addition to image recognition, can be combined with the corresponding data of bread, and can also be combined with the checkout process, which can make the bread identification system of this creation more widely used.

The above-described embodiments are merely illustrative of the principles of the invention and its effects, and are not intended to limit the present invention. Anyone who is familiar with the art can modify and change the above-mentioned implementation form without violating the spirit and scope of this creation. Therefore, the scope of protection of this creation should be as listed in the scope of the patent application described later.

Claims (10)

A bread identification system, comprising: a database, pre-stored a plurality of image models and bread information corresponding to a plurality of image models; and an image capturing module, which performs image capturing of a bread to be tested to obtain the to-be-tested An image processing module for receiving a bread image from the image capturing module to perform image processing of the bread image, and obtaining a segmented image of the bread to be tested from the bread image; and an image ratio For the module, performing the comparison between the segmented image and the plurality of image models to obtain an image model corresponding to the segmented image, and obtaining the bread to be tested from the database according to the image model obtained after the comparison Bread information. The bread identification system of claim 1, wherein before the image processing module obtains the segmented image, the method further comprises performing grayscale processing of the bread image to obtain the bread image from full color to black and white. The steps of the grayscale image of the bread image. The bread recognition system of claim 2, wherein the image processing module further performs feature point detection of the grayscale image to obtain feature points of the bread image. The bread identification system of claim 3, wherein the image processing module further performs the front view determination of the bread image according to the feature point, so that the bread covered area is the foreground portion. The bread identification system of claim 4, wherein the The image processing module performs image segmentation on the bread image according to the foreground portion, so that the foreground portion obtained after the segmentation is the segmented image. The bread identification system of claim 1, wherein when the image matching module does not obtain an image model corresponding to the divided image, the image matching module divides the image model that does not obtain the corresponding image. The image is a new image model and the new image model is added to the database. The bread identification system of claim 1 further includes a machine learning module, wherein the machine learning module intercepts a plurality of the same bread through the image capturing module to obtain a plurality of image data. And using the convolutional neural network to perform machine learning based on the plurality of image data, thereby obtaining one of the plurality of image models. The bread identification system of claim 1, further comprising a display for displaying the bread image captured by the image capturing module and the image matching module The bread information. The bread-recognition system of claim 8, wherein the image capture module, the image processing module, and the image information obtained by the image comparison module are reviewed The image comparison module re-executes image recognition of the bread to be tested. The bread identification system of claim 8, wherein the bread information obtained by the image comparison module is correct after being reviewed, and the bread information is transmitted to the sales terminal system.