A method and system for identifying and locating a target after scattering

Abstract

The present invention provides a method and system for post-scattering target recognition and positioning, which comprises: collecting a speckle image containing a target object to generate a data set; inputting the generated data set into a pre-trained U-net image reconstruction network, and using the pre-trained U-net image reconstruction network to generate a reconstructed image; importing the generated reconstructed image into a pre-trained recognition and positioning network, and using the YOLO algorithm to recognize and locate the target object in the imported reconstructed image. The method solves the technical problem of difficulty in imaging and observing objects through scattering media in the related art.

Description

Technical Field

[0001] The present invention relates to the technical field of imaging through scattering media, and in particular to a method and system for identifying and locating a target after scattering. Background Art

[0002] When light passes through strong scattering media such as thick fog, heavy rain, clouds, and biological tissues, the wavefront phase of the incident light changes due to the interaction between particles, causing the originally orderly light field of the outgoing light to become chaotic, ultimately making it difficult to image and observe the object.

[0003] At present, most imaging technologies through scattering media use numbers or letters as target objects for post-scattering reconstruction. However, real scenes are complex and changeable, making it difficult to reconstruct after scattering. Moreover, most reconstructions only predict the speckle image and do not contain other physical information of the object.

[0004] In order to obtain a clear image of an object after light passes through a scattering medium, people have proposed many scattering imaging technologies, such as the early ballistic light scattering imaging technology. With the emergence of a large number of wavefront shaping devices, scattering imaging technology has achieved a huge breakthrough, and many new scattering imaging technologies have been born, such as optical phase conjugate scattering imaging technology, optical transmission matrix scattering imaging technology, and scattering imaging technology based on optical memory effect. Researchers have achieved focusing or imaging through scattering media through these technologies. However, the above methods all have problems such as cumbersome processes and large amounts of calculations, and have strict requirements on the environment.

[0005] Therefore, the prior art needs to be further developed. Summary of the invention

[0006] The purpose of the present invention is to overcome the above technical deficiencies and provide a method and system for identifying and locating a scattered target, so as to solve the technical problem in the related art that it is difficult to image and observe an object after passing through a scattering medium.

[0007] To achieve the above technical objectives, the present invention adopts the following technical solutions: a method for post-scattering target recognition and positioning is provided, comprising: collecting a speckle image containing a target object to generate a data set; inputting the generated data set into a pre-trained U-net image reconstruction network, and generating a reconstructed image using the pre-trained U-net image reconstruction network; importing the generated reconstructed image into a pre-trained recognition and positioning network, and using the YOLO algorithm to recognize and locate the target object in the imported reconstructed image.

[0008] Furthermore, the pre-trained U-net image reconstruction network includes a dense connection module, and the dense connection module improves the ability of the U-net image reconstruction network to store feature information by capturing data at different levels in the speckle image.

[0009] Furthermore, the mathematical model of the pre-trained U-net image reconstruction network is specifically:

[0010] O P =F(I P ),

[0011] I P ′=F -1 (O P ′),

[0012] Among them, O P is the output of the scatter imaging unit, I P is the input of the scattering imaging unit, F is the positive function from the input light field to the output light field, I P ′ is the input of the reverse process, O P ′ is the output of the inverse process, F -1 It is the inverse of F.

[0013] Furthermore, the method for training the U-net image reconstruction network includes: using a deep learning model to train the F function so that the inverse input I P ′ and positive input I P The characteristics are the same.

[0014] Furthermore, the method for training the U-net image reconstruction network also includes using a preset structural similarity function to compare the original image of the target object with the reconstructed image of the target object to obtain a similarity score.

[0015] Furthermore, the loss function of the YOLO algorithm is:

[0016]

[0017] Among them, K is the output feature, S 2 is the number of cells, B is the number of anchors on each cell, α * is the weight, Indicates whether the k-th output feature, the i-th cell, and the j-th anchor box are positive samples. If they are positive samples, then is 1, otherwise is 0; t p is the prediction vector, t gt is the ground-truth vector; is the weight used to balance the output feature maps of each scale.

[0018] Furthermore, it is characterized in that a preset structural similarity function is used to predict the bounding box of the target object to determine the position of the target object.

[0019] Furthermore, the category of the target object is predicted using a preset structural similarity function to determine the category of the target object.

[0020] A post-scattering target recognition and positioning system comprises: an acquisition unit, the acquisition unit is used to acquire a speckle image containing a target object to generate a data set; a scattering imaging unit, the scattering imaging unit is used to input the data set generated by the acquisition unit into a pre-trained U-net image reconstruction network, and generate a reconstructed image using the pre-trained U-net image reconstruction network; and a recognition and positioning unit, the recognition and positioning unit is used to import the reconstructed image generated by the scattering imaging unit into the pre-trained recognition and positioning network, and use the YOLO algorithm to recognize and locate the target object in the imported reconstructed image.

[0021] A computer-readable storage medium stores computer-readable instructions. When the computer-readable instructions are executed on a communication device, the above method is executed.

[0022] Beneficial effects:

[0023] 1. The post-scattering target recognition and positioning method of the present invention proposes a neural network model to solve the problem of complex objects being difficult to image in scattering imaging.

[0024] 2. The post-scattering target recognition and positioning method of the present invention increases the learning ability and feature information storage capacity of the image reconstruction network by adding densely connected modules and improving some image reconstruction network layers, so that complex objects can be clearly reconstructed and the imaging quality is improved.

[0025] 3. The post-scattering target recognition and positioning method of the present invention realizes the recognition and positioning of the reconstructed object through the neural network YOLO, thereby increasing the amount of physical information of the reconstructed object. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a flow chart of a method for post-scattering target recognition and positioning adopted in an embodiment of the present invention;

[0027] Figure 2 Schematic diagram of the U-net image reconstruction network structure of the post-scattering target recognition and positioning system adopted in an embodiment of the present invention;

[0028] Figure 3a It is a speckle image dataset used when training the U-net image reconstruction network in an embodiment of the present invention;

[0029] Figure 3b is the original image of the target object corresponding to the speckle image data set used when training the U-net image reconstruction network in an embodiment of the present invention;

[0030] Figure 3c is a reconstructed image corresponding to a speckle image dataset used when training a U-net image reconstruction network in an embodiment of the present invention;

[0031] Figure 4a It is the original image of the target object used when training the U-net image reconstruction network in an embodiment of the present invention;

[0032] Figure 4b is a reconstructed image generated when training a U-net image reconstruction network in an embodiment of the present invention;

[0033] Figure 4c is a schematic diagram of SSIM evaluation comparison results of training a U-net image reconstruction network according to an embodiment of the present invention;

[0034] Figure 5a It is an original picture of a real scene in which a post-scattering target recognition and positioning system used in an embodiment of the present invention collects a speckle image;

[0035] Figure 5b is a speckle image corresponding to a real scene collected by a post-scattering target recognition and positioning system adopted in an embodiment of the present invention;

[0036] Figure 5c is a reconstructed image corresponding to a real scene generated by a post-scattering target recognition and positioning system adopted by an embodiment of the present invention;

[0037] Figure 6a is a random speckle image collected by the post-scattering target recognition and positioning system adopted in the embodiment of the present invention;

[0038] Figure 6b is a reconstructed image corresponding to the random speckle image generated by the post-scattering target recognition and positioning system adopted in the embodiment of the present invention;

[0039] Figure 7a is a random speckle image collected by the post-scattering target recognition and positioning system adopted in the embodiment of the present invention;

[0040] Figure 7b It is a schematic diagram of the recognition and positioning of a reconstructed image by the post-scattering target recognition and positioning system adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0042] Embodiment 1:

[0043] like Figure 1 A method for identifying and locating a scattered target, comprising:

[0044] S100 collects speckle images containing target objects and generates a data set;

[0045] In specific practice, an image acquisition device is used to detect a region containing a target object, thereby obtaining a speckle image containing the target object. A collection of speckle images containing the target object is a data set.

[0046] Preferably, the speckle images containing the target object are all collected by the detector.

[0047] S200 inputs the generated data set into a pre-trained U-net image reconstruction network, and generates a reconstructed image using the pre-trained U-net image reconstruction network;

[0048] In specific practice, the pre-trained U-net image reconstruction network is an improved U-net image reconstruction network, which can generate a reconstructed image through a single speckle image.

[0049] Specifically, the first half of the U-Net image reconstruction network is downsampling, and the second half is upsampling. Its structure is similar to a symmetrical U-shape. The upper half of the U-Net image reconstruction network must carefully adjust the size of the convolution kernel and the size of the padding (the space between the element border and the element content, usually called the inner margin) to ensure that the image size is restored to the original size in the end.

[0050] It should be noted, however, that the pre-trained U-net image reconstruction network includes a dense connection module, which improves the ability of the U-net image reconstruction network to store feature information by capturing data at different levels in the speckle image.

[0051] Specifically, the pre-trained U-net image reconstruction network consists of a convolutional layer, a densely connected module, a maximum pooling layer and a connection layer. The pre-trained U-net image reconstruction network adopts a U-shaped structure to connect low-level features with high-level features.

[0052] In practice, the mathematical model of the pre-trained U-net image reconstruction network is:

[0053] O P =F(I P ),

[0054] Among them, O P is the output of the scatter imaging unit, I P is the input of the scattering imaging unit, and F is a positive function from the input light field to the output light field.

[0055] Since the optical path is reversible, ignoring the absorption of light by the scattering medium, the mathematical model from the output light field to the input light field is as follows:

[0056] I P ′=F -1 (O P ′),

[0057] Among them, I P ′ is the input of the reverse process, O P ′ is the output of the inverse process, F -1 It is the inverse function of F. Through the above steps, we can obtain the ideal mathematical model of the imaging problem through scattering media.

[0058] It should be noted that since the speckle image has a corresponding relationship with the original image of the target object, the deep learning model can be used to train the F function so that the inverse input I P ′ and positive input I P The features are the same.

[0059] Specifically, the process of training the U-net image reconstruction network is:

[0060] Firstly, the detector configured by the system is used to collect multiple groups of target objects and their corresponding speckle images as training sets, and then the speckle images of some target objects are collected as test sets.

[0061] Among them, it is necessary to first test simple objects as target objects and conduct imaging experiments on simple objects through scattering media, in order to make full preparations for complex object experiments. In order to collect training sets, simple objects are tested at different distances and three groups of experiments are conducted, such as Figure 3a to Figure 3c As shown in the figure, the backpacks are placed at different positions from the detector, and data is collected for the backpacks at different positions. 8,000 original object images and 8,000 corresponding speckle images are collected for each group, and these data are used as the test set. Among them, the image size is 256*256 pixel color original images. The color original images can simulate the scene environment realistically, and the color images have richer information.

[0062] System configuration such as Figure 2 As shown, the system configuration is used to acquire experimental image data. Figure 2 This is a schematic diagram of the improved U-net network structure. The dense connection module can improve the network's ability to store feature information. Figure 1 As shown, a display 1 is connected to a detector 2 , a scattering medium 3 is placed in front of the detector 2 , and a target object 4 is placed under natural light 5 .

[0063] Preferably, the color of the final reconstruction result is changed into a thermal imaging form and a grayscale form.

[0064] Since the cv2.imread function in Python reads images in BGR format (RGB color mode is a color standard in the industry, representing R (red), G (green), and B (blue) respectively), the color of the final displayed image will change.

[0065] Then, the collected training set is put into the improved U-net image reconstruction network. After the training is completed, the test set is put into the improved U-net to reconstruct the target image;

[0066] In practice, when training the U-net image reconstruction network, 8,000 original images of the target object and 8,000 corresponding speckle images were selected as the training set, and 1,000 random speckle images were used as the test set. The Adam optimizer was used for a total of 30 trainings, each time 1,000 original object images and speckle images were selected for processing, and the learning rate was set to 0.0005 each time. The training took 40 minutes. The improved U-net network was run in the Python 3.9 environment.

[0067] Finally, the quality of the imaging is determined by comparing the original image of the object with the reconstructed image using SSIM (structural similarity).

[0068] Specifically, SSIM is an indicator to measure the similarity between two images. The specific evaluation method is that given two images x and y, the structural similarity of the two images can be calculated as follows:

[0069]

[0070] Among them, μ x is the mean value of x, μ y is the mean value of y, is the variance of x, is the variance of y, σ xy is the covariance of x and y. c1=(k1L) 2 , c2=(k2L) 2It is a constant used to maintain stability, L is the dynamic range of pixel values, k1=0.01, k2=0.03.

[0071] In order to evaluate the imaging performance of the improved U-net network, SSIM is used to compare the original image of the object with the reconstructed image to determine the imaging quality.

[0072] like Figures 4a to 4c As shown in the figure, the parameter corresponding to the ssim score is the SSIM evaluation comparison result. It can be seen that the similarity scores between the reconstructed image of the target object reconstructed by the U-net image reconstruction network and the original image of the target object are all above 0.81, and the highest reaches 0.9. The distances between the three target objects and the detector are different. The target objects are placed at 5m, 7m, and 10m away from the detector respectively. Experiments have proved that the improved U-net image reconstruction network has better reconstructed image quality for simple objects.

[0073] like Figures 5a to 6b As shown in the figure, the experimental background is transferred to a complex background (a road in a real scene), and the collected random speckle image is reconstructed to clearly restore the original target object. It is proved that the improved U-net image reconstruction network can be used for imaging experiments of complex objects through scattering media.

[0074] By testing the improved U-net image reconstruction network, it was found that it has powerful image reconstruction capabilities. However, clear reconstructed images alone cannot be used in many fields. More physical information needs to be extracted from the reconstructed target objects.

[0075] In specific practice, it is necessary to improve the existing U-net image reconstruction network and build a neural network that can image complex objects through scattering media. This is used for image reconstruction of complex objects through scattering media. Specifically, the improved U-net image reconstruction network consists of a convolutional layer, a densely connected module, a maximum pooling layer and a connection layer. The improved U-net image reconstruction network can significantly improve the quality of image reconstruction.

[0076] Specifically, the dense connection mechanism in the U-net image reconstruction network uses dense connections in the convolutional layer or the encoder and decoder parts. The characteristics of the dense connection mechanism in the U-net image reconstruction network are: there is a direct connection between any two layers, which means that the input of each layer of the network is the layer before the union of all network outputs, and the feature map learned by this layer is directly passed as input to all the layers behind it, which can not only alleviate the gradient vanishing problem, enhance feature propagation, encourage feature reuse, and greatly reduce the amount of parameters, thereby improving the quality of image reconstruction.

[0077] In specific practice, the method for training the U-net image reconstruction network also includes using a preset structural similarity function to compare the original image of the target object with the reconstructed image of the target object to obtain a similarity score.

[0078] S300 imports the generated reconstructed image into a pre-trained recognition and positioning network, and uses the YOLO algorithm to recognize and locate the target object in the imported reconstructed image.

[0079] In practice, the loss function of the YOLO algorithm is:

[0080]

[0081] Among them, K is the output feature, S 2 is the number of cells, B is the number of anchors on each cell, α* is the weight, Indicates whether the k-th output feature, the i-th cell, and the j-th anchor box are positive samples. If they are positive samples, then is 1, otherwise is 0; t p is the prediction vector, t gt is the ground-truth vector; is the weight used to balance the output feature maps of each scale.

[0082] Preferably, the default value of α* is α box =0.05,α obj =0.7,α cls =0.3.

[0083] Preferably, The default value of is [4.0, 1.0, 0.4], which corresponds to output feature maps of sizes 80*80, 40*40, and 20*20, respectively.

[0084] Specifically, the YOLO algorithm is a target detection algorithm based on a neural network. It takes an image as input and can directly output the location and category information of the target object.

[0085] Specifically, the YOLO algorithm is a real-time object detection algorithm that can simultaneously detect multiple targets in an image in a single forward pass. Compared with traditional target detection methods, YOLO has the advantages of fast and real-time performance, and can accurately identify and locate multiple target objects in images or videos.

[0086] In practice, the object detection problem is transformed into a regression problem. It divides the input image into fixed-size grids and predicts the location and category probability of the object bounding box in each grid. This means that YOLO only needs to perform one forward propagation on the entire image to simultaneously predict the location and category of all objects.

[0087] It should be noted that the specific workflow of YOLO is:

[0088] Grid division: The input image is divided into grids of fixed size. Each grid is responsible for detecting the target object in the area.

[0089] Bounding box prediction: For each grid, multiple bounding boxes are predicted, each of which is represented by a set of bounding box coordinates (x, y, width, height).

[0090] Category prediction: For each bounding box, predict the category probability of the object. YOLO uses multi-label classification to handle the detection of multiple categories of objects.

[0091] Non-Maximum Suppression: The non-maximum suppression algorithm is used to suppress bounding boxes with a lot of overlap, and only the bounding boxes with the highest confidence are retained.

[0092] Before using YOLO for target detection, you need to prepare an appropriate dataset. In addition to the reconstructed image obtained by the U-net image reconstruction network mentioned above, you also need to prepare the following:

[0093] Labeling tool: Select a suitable labeling tool to manually label the location and category information of the target object. Commonly used labeling tools include LabelImg, RectLabel, CVAT, etc. Install and learn to use the labeling tool, and prepare the image file for labeling.

[0094] Image annotation: Use the annotation tool to open the image file and annotate the target objects one by one. Use rectangular boxes or polygons to annotate the target objects according to their locations, and assign corresponding category labels to each annotated box. Ensure that the locations and categories of the target objects are accurately annotated, and save the annotation information.

[0095] Annotation file format: Save the annotation information in a specific format. Common formats include PASCAL VOC format (XML file) and YOLO format (TXT file). Make sure that each image has a corresponding annotation file that contains the location and category information of the target.

[0096] Dataset division: According to the requirements, the entire dataset is divided into training set, validation set and test set. Usually, the training set is used for model training, the validation set is used for model tuning and selecting hyperparameters, and the test set is used to evaluate the performance of the model.

[0097] Dataset preprocessing: Before using the dataset for training, some preprocessing operations may be required, such as image resizing, data augmentation (such as flipping, rotation, scaling, etc.), normalization, etc. These operations help improve the robustness and generalization ability of the model.

[0098] In practice, the YOLO algorithm is used to identify and locate target objects in reconstructed images. The basic steps of using YOLO for target detection and inference include:

[0099] Model preparation: Select the YOLO version that suits your needs and obtain the corresponding model architecture and weight files. Make sure you have the corresponding deep learning framework (TensorFlow, PyTorch) and dependent libraries.

[0100] Model loading: Load the selected YOLO model into memory, including the model's architecture definition and pre-trained weight parameters. Depending on the deep learning framework used, the model can be loaded through the corresponding function or class.

[0101] Input data preparation: Convert the image to be detected into an input format acceptable to the model. Usually, the image needs to be resized and normalized to match the input requirements of the model.

[0102] Forward propagation: The preprocessed image is input into the YOLO model for forward propagation. The model will perform operations such as convolution and pooling on the image to generate the position and category prediction results of the detection box.

[0103] Prediction result parsing: Parse the output of the model to extract information such as the location, category label, and confidence score of the target. Typically, the output of the YOLO model is a list of multiple bounding boxes, each with an associated category label and confidence score.

[0104] Post-processing operations: Post-processing operations are performed as needed to filter redundant bounding boxes and improve the accuracy of detection. Common post-processing steps include non-maximum suppression (NMS) to remove highly overlapping bounding boxes and retain the bounding boxes with the highest confidence.

[0105] Target visualization or output: Visualize or output detected targets to files or other media as needed. You can draw bounding boxes and category labels on the image, or save the detection results in formats such as JSON, XML, etc.

[0106] In specific practice, a preset structural similarity function is used to predict the bounding box of the target object and determine the position of the target object.

[0107] In specific practice, a preset structural similarity function is used to predict the category of the target object and determine the category of the target object.

[0108] like Figure 7a to Figure 7b As shown in the figure, the reconstructed target object is accurately located and identified. It can be seen that the prediction box accurately determines the specific location of the target object, and accurately identifies the target object category above the prediction box. It can also be concluded that the quality of the reconstructed image is proved from another perspective.

[0109] In specific practice, it is necessary to collect objects in real scenes and their corresponding speckle images as data sets to train and test the U-net image reconstruction network, wherein people or vehicles in real scenes are taken as target objects, a piece of frosted glass is selected as the scattering medium, natural light is used as the illumination light source, and a data set of speckle images containing the target objects is obtained through the detector.

[0110] This embodiment adopts a post-scattering target recognition and positioning system, including: an acquisition unit, the acquisition unit is used to acquire a speckle image containing a target object to generate a data set; a scattering imaging unit, the scattering imaging unit is used to input the data set generated by the acquisition unit into a pre-trained U-net image reconstruction network, and use the pre-trained U-net image reconstruction network to generate a reconstructed image; an identification and positioning unit, the identification and positioning unit is used to import the reconstructed image generated by the scattering imaging unit into the pre-trained identification and positioning network, and use the YOLO algorithm to identify and locate the target object in the imported reconstructed image.

[0111] This embodiment provides a computer-readable storage medium, in which computer-readable instructions are stored. When the computer-readable instructions are executed on a communication device, the above method is executed.

[0112] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0113] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, and this embodiment will not be described in detail here.

[0114] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0115] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0116] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for identifying and locating a scattered target, characterized in that: include: Acquire a speckle image containing a target object to generate a data set; Input the generated data set into a pre-trained U-net image reconstruction network, and use the pre-trained U-net image reconstruction network to generate a reconstructed image; Import the generated reconstructed image into the pre-trained recognition and positioning network, and use the YOLO algorithm to recognize and locate the target object in the imported reconstructed image; The pre-trained U-net image reconstruction network includes a dense connection module, which improves the ability of the U-net image reconstruction network to store feature information by capturing data at different levels in the speckle image; The mathematical model of the pre-trained U-net image reconstruction network is as follows: ， ， in, is the output of the scatter imaging unit, is the input of the scatter imaging unit, is a positive function from input light field to output light field, is the input of the inverse process, is the output of the inverse process, yes The inverse of The method of training the U-net image reconstruction network includes: using a deep learning model to train the F function so that the inverse input With positive input The characteristics are the same; The method for training the U-net image reconstruction network also includes using a preset structural similarity function to compare the original image of the target object with the reconstructed image of the target object to obtain a similarity score.

2. The method for post-scattering target recognition and positioning according to claim 1, characterized in that: The preset structural similarity function is used to predict the bounding box of the target object and determine the location of the target object.

3. The method for post-scattering target recognition and positioning according to claim 1, characterized in that: The category of the target object is predicted using a preset structural similarity function to determine the category of the target object.

4. A post-scattering target recognition and positioning system, characterized in that: The method for identifying and locating a scattered target according to any one of claims 1 to 3 comprises: An acquisition unit, the acquisition unit is used to acquire a speckle image containing a target object to generate a data set; A scattering imaging unit, wherein the scattering imaging unit is used to input the data set generated by the acquisition unit into a pre-trained U-net image reconstruction network, and generate a reconstructed image using the pre-trained U-net image reconstruction network; The recognition and positioning unit is used to import the reconstructed image generated by the scattering imaging unit into a pre-trained recognition and positioning network, and use the YOLO algorithm to recognize and locate the target object in the imported reconstructed image.

5. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed on the communication device, the method according to any one of claims 1 to 3 is executed.

Images