Haircut style determination system and haircut style determination method

Abstract

To provide a cut style determination system that determines a type of cut style of an animal by a simple method.SOLUTION: A cut style determination system includes acceptance means for accepting an input of an animal image, and determination means for determining, using a learned model, a type of cut style of an animal from the animal image input to the acceptance means. The learned model is a learned model that has learned a relation between the animal image and the type of cut style of the animal.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a haircut style determination system and a haircut style determination method, and more particularly to a haircut style determination system and a haircut style determination method that provide a determination result regarding the type of an animal's haircut style from an image of the animal. [Background technology]

[0002] Pets such as dogs, cats, and rabbits are irreplaceable to humans, and various services for pets are becoming popular and developing. Among them, grooming services that trim pet fur by cutting and brushing are a regular necessity for pets, and are of great interest to pet owners.

[0003] Pet grooming is not a simple task but requires advanced skills. For this reason, there are training institutions such as vocational schools for groomers, and certifications for groomers such as the "JKC Certified Groomer Qualification" certified by the Japan Kennel Club.

[0004] There are several types, or cut styles, of pet grooming, just like human hair, such as lion cut, summer cut, teddy bear cut, etc.

[0005] As for the type of haircut style, if the grooming salon that performed the grooming informs the owner of that style, the owner will be able to know it. However, in cases where this is not the case, pet owners may have a need to know the type of haircut style their pets have in a simple way.

[0006] Therefore, there is a need for a simple method for determining the type of haircut style.

[0007] Patent Document 1 discloses a hair simulation device for simulating hair color, hairstyle, and the like, but does not disclose how to determine the type of haircut style for a pet. [Prior art documents] [Patent documents]

[0008] [Patent Document 1] JP 2012-168855 A Summary of the Invention [Problem to be solved by the invention]

[0009] Therefore, an object of the present invention is to provide a cutting style determination system and a cutting style determination method that can determine the type of cutting style in a simple manner. [Means for solving the problem]

[0010] As a result of intensive research into solving the above problem, the inventors discovered that the above problem could be solved by a trained model that had learned the relationship between images of animals and the types of haircut styles of the animals, and thus completed the present invention.

[0011] That is, the present invention relates to the following [1] to [9]. [1] A haircut style determination system comprising a reception means for receiving input of an image of an animal, and a determination means for determining the type of haircut style of an animal from the image of the animal input to the reception means using a trained model, wherein the trained model is a trained model that has learned the relationship between the image of the animal and the type of haircut style of the animal. [2] A haircut style determination system comprising a reception means for receiving input of an image of an animal, and a determination means for using a trained model to determine the type of haircut style of the animal from the image of the animal input to the reception means, wherein the trained model is a trained model that learns using an image of the animal and a label related to the type of haircut style of the animal as training data, and takes an image of the animal as input and outputs a determination of the type of haircut style of the animal. [3] The haircut style determination system according to [1] or [2], wherein the image of the animal accepted by the accepting means is an image of the animal's face taken from the front. [4] The haircut style determination system according to [3], wherein the face image of the animal accepted by the accepting means is a face image of the animal including the ears of the animal. [5] The haircut style judgment system according to [2], wherein the images of the animal used as the training data are images of the animal within a predetermined period of time since being trimmed. [6] The haircut style judgment system according to [2], wherein the images of animals used as the training data have a uniform resolution. [7] A method for generating a trained model that determines the type of haircut style of an animal from an image of the animal, the method comprising inputting an image of the animal and a label relating to the type of haircut style of the animal into a computer as training data and having an artificial intelligence learn the trained model. [8] A haircut style determination method comprising the steps of: preparing an image of an animal; inputting the image into a trained model; and using the trained model, a computer outputting a judgment of the type of haircut style of the animal from the input image of the animal, wherein the trained model is a trained model that has learned the relationship between the image of the animal and the type of haircut style of the animal. [9] A haircut style determination method comprising the steps of preparing an image of an animal, inputting the image into a trained model, and using the trained model by a computer to output a judgment of the type of haircut style of the animal from the input image of the animal, wherein the trained model is trained using an image of the animal and a label related to the type of haircut style of the animal as training data, and the input is an image of the animal and the output is a judgment of the type of haircut style of the animal. Effect of the Invention

[0012] The present invention makes it possible to provide a haircut style determination system and a haircut style determination method that determine the type of haircut style in a simple manner. In addition, by being able to determine the type of haircut style in a simple manner, it is also possible to propose hair and skin care products and care methods that are suitable for the haircut style. [Brief description of the drawings]

[0013] [Figure 1] FIG. 13 is a diagram showing an example of a suitable animal face image. [Diagram 2] FIG. 13 is a diagram showing an example of a suitable animal face image. [Diagram 3] (A) An image of an example of a bear cut on a toy poodle, (B) an image of an example of an afro cut on a toy poodle, (C) an image of an example of a mushroom cut on a toy poodle, and (D) an image of an example of a facial varicut on a toy poodle. [Figure 4] (A) An image of an example of a schnauzer cut, (B) an image of an example of a bear cut, and (C) an image of an example of a mohawk cut on a miniature schnauzer. [Diagram 5] 1 is a schematic diagram showing an embodiment of a haircut style determination system of the present invention. [Figure 6] FIG. 2 is a flowchart showing an example of a flow of cutting style determination by the cutting style determination system of the present invention. [Figure 7] This is an example of an animal image. [Figure 8] (A) An example image where the lines are drawn, and (B) an example image where the lines are not drawn. [Figure 9] (A) An example image where the cut surfaces are aligned, and (B) an example image where the cut surfaces are not aligned. [Figure 10] (A) An example image with left-right balance, (B) an example image with imbalance. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] <Cutting style judgment system> The haircut style determination system of this embodiment is a haircut style determination system comprising a reception means for receiving an input of an image of an animal, and a determination means for determining the type of haircut style of an animal from the image of the animal input to the reception means using a trained model, and is characterized in that the trained model is a trained model that has learned the relationship between the image of the animal and the type of haircut style of the animal.

[0015] [Method of reception] The receiving means is a means for receiving input of an image of an animal. Examples of animals include mammals such as dogs, cats, and rabbits, with dogs and cats being preferred. The method of receiving an image may be any method, such as scanning, inputting image data, transmitting, or capturing an image on the spot. The format of the image is not particularly limited, and may be a still image or a video. The trimming of an animal is for the whole body, and the part of the animal shown in the image is not particularly limited, but the image of the animal is preferably an image showing the face of the animal, more preferably a photograph of the animal's face taken from the front as in FIG. 1, and even more preferably a photograph showing the animal's face in a large size. In addition, an image of the face of an animal that shows up to the ears as in FIG. 7(A) is particularly preferred, rather than an image trimmed to show only the vicinity of the muzzle as in the circular frame of FIG. 7(B) or to show only the vicinity of the eyes as in the square frame of FIG. 7(C). An example of such a photograph is a photograph such as a photograph of a human driver's license. Images such as those used for animal health insurance cards as in FIG. 2 are also preferred. The image may be black and white, grayscale, or color. Images that do not show the entire face of an animal, images whose shape has been edited using image editing software, images that show multiple animals, images where the face is so small that the eyes and ears cannot be made out, or images that are unclear are not acceptable. Images that have been normalized or have a uniform resolution are preferable.

[0016] [Judgment means] The determination means is means for determining the type of haircut style of an animal from the image of the animal input to the reception means by using a trained model. In this embodiment, the trained model is a trained model that has learned the relationship between the image of the animal and the type of haircut style of the animal.

[0017] The trained model described above can be generated, for example, by supervised learning or unsupervised learning. In the case of supervised learning, the training data can be, for example, images of animals and data on the types of haircut styles of the animals. The training data can be obtained by attaching a tag corresponding to the type of haircut style of the animal to the image of the animal.

[0018] The types of cut styles include, for example, the cuts determined by the JKC show clip and trimming competitions, specifically, puppy clip (under 15 months old), puppy clip II (second), continental clip or English saddle clip, lion clip, modern clip, Scandinavian clip, terrier clip, etc. Others include summer cut, lion cut, teddy bear cut, mohawk cut, afro cut, peanut cut, schnauzer cut, asymmetrical cut, lamb clip, boot cut of the legs, pants cut of the buttocks, full body clipper, Mickey cut, ornamental hair style, top knot, full coat, flare cut, twin tail, braided style, schnauzer cut, etc.

[0019] The format of the animal image used as the training data is not particularly limited. The trimming of the animal is for the whole body, and the part of the animal shown in the image is not particularly limited. However, the image of the animal is preferably an image showing the face of the animal, more preferably a photograph of the animal's face taken from the front, and even more preferably a photograph showing the animal's face large as in FIG. 1. In addition, an image of the face of the animal showing up to the ears as in FIG. 7(A) is particularly preferable, compared to an image trimmed to show only the area around the muzzle as in the circular frame of FIG. 7(B) or to show only the area around the eyes as in the square frame of FIG. 7(C). An example of such a photograph is a photograph such as a photograph for a human driver's license. As in FIG. 2, an image used for an animal's health insurance card may be used. The image may be black and white, grayscale, or color. Images that do not show the entire face of the animal, images whose shape has been edited with image editing software, images that show multiple animals, images in which the face is so small that the eyes and ears cannot be distinguished, or images that are unclear are not preferable. The images are preferably of animals within a predetermined period of time since being trimmed, normalized, or have a uniform resolution. It is more preferable that the resolution of the images used in the training data and the images received by the receiving means are uniform.

[0020] As the trained model, artificial intelligence (AI) is preferable. Artificial intelligence (AI) is software or a system that imitates the intellectual work performed by the human brain using a computer, specifically, computer programs that understand natural language used by humans, perform logical inference, and learn from experience. AI may be either general-purpose or specialized, and may be any of deep neural networks, convolutional neural networks, etc., and publicly available software may be used.

[0021] To generate a trained model, artificial intelligence is trained. Either machine learning or deep learning can be used for the training, but deep learning is preferred. Deep learning is an advanced version of machine learning, and is characterized by its ability to automatically find features.

[0022] The learning method for generating the trained model is not particularly limited, and publicly available software can be used. For example, DIGITS (the Deep Learning GPU Training System) published by NVIDIA can be used. In addition, training may be performed using a known support vector machine method published in, for example, "Introduction to Support Vector Machines" (Kyoritsu Shuppan).

[0023] The haircut style judgment system and haircut style judgment method of the present invention may include, in addition to the judgment means for judging the type of haircut style, a trimming result judgment means for judging the quality of the trimming of the haircut style and the score judgment. The trimming result judgment means is a judgment means for judging the quality of the trimming of the animal from the image of the animal input to the receiving means by using a trained model, and the trained model is a trained model that has learned the relationship between the image of the trimmed animal and the quality of the trimming of the animal. Specifically, it is preferable that the trained model is a trained model that learns using the image of the trimmed animal and the label related to the quality of the trimming of the animal as teacher data, inputs the image of the animal, and outputs the quality of the trimming of the animal. The trimming judgment is preferably performed after the judgment of the haircut style, and outputs the quality of the haircut style and the score result together with the judgment of the haircut style. The trimming result judgment means is configured in the same manner as the judgment means (trained model) 11 of the embodiment described later, for example. The quality of trimming is, for example, a judgment result that the trimming is good or bad. Specific examples of judgment criteria include whether the trimming is symmetrical, whether the outline is clear, whether each part is sharp, and whether the cut surfaces are even. In all of the examples given here, the former are good examples, and the latter are bad examples. Other criteria include whether the trimming is three-dimensional, whether the cuts are well-balanced, and whether the connections between the flat surfaces are smooth and even. In addition, as a criterion for judging the trimming, for example, the criterion for the trimming competition of the All Japan Animal Specialist Education Association can be used. FIG. 8 shows an example of an image with (A) a clear line and an example of an image with (B) an unclean line. In addition, FIG. 9 shows an example of an image with (A) a clear cut and an example of an unclean cut. FIG. 10 shows (A) an example image where the left and right sides are balanced, and (B) an example image where the left and right sides are not balanced.

[0024] The animal images may be scored for each individual criterion, or may be scored based on a combination of these criteria, or may be judged as "good" or "bad." That is, each image used as training data may be tagged with a score according to the above criteria, or tagged according to whether it is good or bad, and used as training data. The quality of the trimming may be judged, for example, by an expert in trimming or a person with knowledge of the scoring method used in a trimming competition or the like.

[0025] [output] When the determination means receives an image of an animal as input information, it determines the type of haircut style of the animal using the trained model. The output format is not particularly limited, and the judgment can be output, for example, by displaying types of haircut styles such as "lion cut" and "summer cut" on a computer screen, or by presenting multiple types of haircut styles such as "there is a 70% chance that this child's haircut style is a lion cut, and a 20% chance that it is a summer cut," or by displaying the probabilities along with the types of haircut styles. The hair cutting style determination system of the present invention may further include an output means for receiving the determination result from the determination means and outputting the determination result.

[0026] Hereinafter, one embodiment of the haircut style determination system of the present invention will be described with reference to FIG.

[0027] In Fig. 5, terminal 40 is a terminal used by a user. Examples of users include pet owners who want to know the types of haircut styles for their pets, and hairdressers who want to know the types of haircut styles for their customers' pets and use them as a topic of conversation with their customers. Examples of terminal 40 include personal computers, smartphones, and tablet terminals. Terminal 40 includes a processing unit such as a CPU, a storage unit such as a hard disk, ROM, or RAM, a display unit such as a liquid crystal panel, an input unit such as a mouse, keyboard, or touch panel, and a communication unit such as a network adapter.

[0028] A user accesses the server from the terminal 40, and inputs and transmits an image (photo) of an animal, as well as information such as the type, breed, sex, and weight of the animal. In determining the type of haircut style, it is preferable that the image is taken after the animal has been trimmed and before the fur grows. For example, an image taken within one month of the trimming is preferable, within two weeks is more preferable, and within one week is even more preferable. In addition, when using this haircut style determination system, the user may take a photo or video of the target animal using a smartphone camera on the spot, and input and transmit the photo. For example, the user takes a photo of the face of the target animal according to instructions displayed on the screen of the terminal 40, and when an appropriate photo is taken, transmits the photo to the server. At this time, the server may be provided with a separate photo-taking assistant means consisting of an image determination program, and the photo-taking assistant means may be configured to determine whether the photo is suitable for haircut style determination, such as whether the entire face of the animal is captured and the photo is taken from the front of the animal's face, and to transmit the determination result to the user through an interface or terminal. Moreover, the user can receive the result of the haircut style determination made by the server by accessing the server through the terminal 40.

[0029] In the present embodiment, the server is configured by a computer, but may be any device as long as it has the functions according to the present invention. The server may be a server on a cloud.

[0030] The storage unit 10 is composed of, for example, a ROM, a RAM, a hard disk, etc. The storage unit 10 stores an information processing program for operating each part of the server, and in particular, stores a determination means (trained model) 11.

[0031] The determination means (trained model) 11 receives an image of a target animal input by a user, and outputs a determination of the type of haircut style of the animal included in the image. The determination means (trained model) 11 in this embodiment is configured to include, for example, a deep neural network or a convolutional neural network.

[0032] The processing calculation unit 20 executes haircut style judgment using the judgment means (trained model) 11 stored in the storage unit.

[0033] The interface section (communication section) 30 includes a receiving means 31 and an output means 32, and receives images of animals and other information from the user's terminal, and outputs the result of the determination of the type of haircut style to the user's terminal.

[0034] With the haircut style determination system of this embodiment, a user can easily obtain a determination of the type of haircut style of their pet by uploading photos, videos, etc. of the pet to a server.

[0035] In this embodiment, a configuration has been described in which the determination means and reception means are stored in a server and connected to a user's terminal via a connection means such as the Internet or a LAN, but the present invention is not limited to this, and may also be configured in such a way that the determination means, reception means, and interface unit are stored in a single server or device, or in such a way that a separate terminal is not required for use by the user.

[0036] A flowchart of haircut style determination based on one embodiment of the haircut style determination system of the present invention is shown in Fig. 6. A user inputs basic information such as the type and breed of the target animal into the reception means and uploads an image of the animal (step S1). The processing calculation unit of the server uses the determination means (trained model) to determine the type of haircut style of the animal from the uploaded image (step S2). The output means outputs the derived determination result by displaying it on a screen, for example, and presents it to the user (step S3).

[0037] <Other embodiments> Another embodiment of the present invention is a haircut style determination system comprising a reception means for receiving an input of an image of an animal, and a determination means for using a trained model to determine the type of haircut style of an animal from the image of the animal input to the reception means, characterized in that the trained model is a trained model that learns using an image of an animal and a label related to the type of haircut style of the animal as training data, and that takes the image of the animal as input and outputs a determination of the type of haircut style of the animal. This embodiment is characterized in that the trained model is trained using images of animals and labels related to the types of haircut styles of the animals as training data, and the input is the image of the animal and the output is a determination of the type of haircut style of the animal.Other points are the same as those described above.

[0038] <How to determine cut style> The haircut style determination method of the present invention is a haircut style determination method having the steps of preparing an image of an animal, inputting the image into a trained model, and having a computer use the trained model to output a haircut style determination of the animal from the input image of the animal, characterized in that the trained model is a trained model that has learned the relationship between the image of the animal and the type of haircut style of the animal.

[0039] Another aspect of the present invention is a haircut style determination method comprising the steps of preparing an image of an animal, inputting the image into a trained model, and using the trained model by a computer to output a judgment of the type of haircut style of the animal from the input image of the animal, characterized in that the trained model is a trained model that learns using an image of an animal and a label related to the type of haircut style of the animal as training data, and that takes the image of the animal as input and outputs a judgment of the type of haircut style of the animal.

[0040] The animal images and the trained model are the same as those in the haircut style determination system of the present invention described above. EXAMPLES

[0041] [Example 1] We prepared facial photographs of toy poodles with bare cuts (an example is the image in Fig. 3(A)), afro cuts (an example is the image in Fig. 3(B)), mushroom cuts (an example is the image in Fig. 3(C)) or facial clipper cuts (an example is the image in Fig. 3(D)). These facial photographs were cropped (cut out) to include the entire head. Using these facial photos, we performed deep learning and generated a trained model. There were 60 facial photos used as training data for each cut. Each image was tagged with "0" if it was a bare cut, "1" if it was an afro cut, "2" if it was a mushroom cut, and "3" if it was a facial haircut, and used for learning. The learning method was transfer learning using MobileNet as the artificial intelligence (neural network), and Pytorch was used as the machine learning library (deep learning library). In addition, 35 photographs of the face of a toy poodle were used for each cut as data for evaluation during the learning process. The evaluation results during the learning process are shown in Table 1.

[0042] [Table 1]

[0043] Next, using the trained model obtained, we conducted verification using 57 photos of a toy poodle for each cut, which were different from the photos used for the training data and verification data. The accuracy rate was 93.4%. The results are shown in Table 2.

[0044] [Table 2]

[0045] [Example 2] We prepared facial photographs of miniature schnauzers with a schnauzer cut (an example is the image in Figure 4(A)), a bear cut (an example is the image in Figure 4(B)), or a mohawk cut (an example is the image in Figure 4(C)). These facial photographs were cropped (cut out) to include the entire head. Using these facial photos, we performed deep learning and generated a trained model. There were 60 facial photos used as training data for each cut. Each image was tagged with "0" if it was a Schnauzer cut, "1" if it was a Bear cut, and "2" if it was a Mohawk cut, and used for learning. The learning method was transfer learning using MobileNet as the artificial intelligence (neural network), and Pytorch was used as the machine learning library (deep learning library). In addition, 35 facial photographs of miniature schnauzers were used for each cut as data for evaluation during the learning process. The evaluation results during the learning process are shown in Table 3.

[0046] [Table 3]

[0047] Next, using the trained model, we conducted verification using 55 photos of miniature schnauzers for each cut, which were different from the photos used for the training data and verification data. The accuracy rate was 89.7%. The results are shown in Table 4.

[0048] [Table 4]

Claims

1. A haircut style determination system comprising: a receiving means for receiving an input of an image of an animal; and a determining means for determining a type of haircut style of the animal from the image of the animal input to the receiving means using a trained model, A haircut style determination system, characterized in that the trained model is a trained model that has learned the relationship between images of animals and types of haircut styles of the animals.

2. A haircut style determination system comprising: a receiving means for receiving an input of an image of an animal; and a determining means for determining a type of haircut style of the animal from the image of the animal input to the receiving means using a trained model, A haircut style determination system characterized in that the trained model is trained using images of animals and labels related to the types of haircut styles of the animals as training data, and the input is an image of an animal and the output is a determination of the type of haircut style of the animal.

3. 3. The haircut style determining system according to claim 1, wherein the image of the animal accepted by said accepting means is an image of the animal's face photographed from the front.

4. 4. The haircut style determining system according to claim 3, wherein the face image of the animal accepted by said accepting means is a face image of the animal including the ears of the animal.

5. 3. The haircut style determination system according to claim 2, wherein the images of the animal used as the training data are images of the animal within a predetermined period of time since being trimmed.

6. 3. The haircut style determination system according to claim 2, wherein the images of animals used as the training data have a uniform resolution.

7. A method for generating a trained model that determines the type of haircut style of an animal from an image of the animal, the method comprising inputting an image of the animal and a label relating to the type of haircut style of the animal into a computer as training data, and allowing an artificial intelligence to learn the trained model.

8. Providing an image of an animal; A haircut style determination method comprising: inputting the image into a trained model; and outputting a type of haircut style of the animal from the input image of the animal using the trained model by a computer. A haircut style determination method, characterized in that the trained model is a trained model that has learned the relationship between an image of an animal and a type of haircut style of the animal.

9. providing an image of an animal; A haircut style determination method comprising: inputting the image into a trained model; and outputting a type of haircut style of the animal from the input image of the animal using the trained model by a computer. A haircut style determination method, characterized in that the trained model is a trained model that learns using images of animals and labels related to the types of haircut styles of those animals as training data, takes an image of an animal as input, and outputs a determination of the type of haircut style of that animal.

Images