Fabric image processing device and method

Through the fabric image processing device, the physical deformation parameters of fabric are predicted using neural networks and linear regression models, which solves the problem of cumbersome measurement in textile technology and improves the efficiency and effect of fabric simulation in three-dimensional virtual space.

CN119006973APending Publication Date: 2024-11-22NUNOX TECHNOLOGIES CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410604671.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2023-05-19
Filing Date
2024-05-15
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

In textile technology, measuring the physical deformation parameters of fabrics (such as warp bending, weft bending, warp stretching, weft stretching, etc.) requires complicated detection processes, resulting in simulated fabric drooping and simulating the fabric in the three-dimensional virtual space and Ineffective exercise.

Method used

A fabric image processing device is adopted to combine the neural network model and linear regression model through the image capture circuit, memory and processor, and use the fabric top view and side view images to perform backing algorithms to generate silhouette diagrams and update model parameters, thereby Predict physical deformation parameters and avoid direct measurement.

Benefits of technology

It saves time and manpower in measuring physical deformation parameters, improves the efficiency and effect of simulating virtual fabrics in three-dimensional virtual space, and can quickly identify new physical deformation parameters for simulation of virtual fabrics.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006973A_ABST
    Figure CN119006973A_ABST
Patent Text Reader

Abstract

The invention provides a fabric image processing device. The fabric image processing device comprises an image acquisition circuit, an internal memory and a processor, the image capturing circuit is used for capturing a first top view image and a first side view image of the first fabric. The processor executes the following operations: updating a first neural network model and a first linear regression model according to the first overlooking silhouette and the corresponding first physical deformation parameter, and updating a second neural network model and a second linear regression model according to the first side-looking silhouette and the corresponding first physical deformation parameter; and respectively inputting the first overlook silhouette picture and the first side view silhouette picture into a first neural network model and a second neural network model to generate a combination vector, and updating a third linear regression model according to the combination vector and the corresponding first physical deformation parameter. According to the fabric image processing device and method disclosed by the invention, the time and manpower for measuring the physical deformation parameters can be saved, and the effect of simulating the virtual fabric in the three-dimensional virtual space is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of fabric image processing, and in particular relates to a fabric image processing device and method. Background Art

[0002] In current textile technology, it is often necessary to simulate a virtual fabric (or cloth) in a virtual three-dimensional space in order to observe the drape and movement of the virtual fabric in the three-dimensional space. This often requires measuring the physical deformation parameters of the fabric (for example, bending works warp work, bending works weft work, stretch warp elongation, stretch weft elongation or oblique elongation, etc.) before further simulation. However, to measure the physical deformation parameters of the fabric, a specific testing machine is usually required to test the fabric, which takes a lot of time and manpower. Therefore, how to quickly know the physical deformation parameters of the fabric without measuring the physical deformation parameters of the fabric is a problem that technicians in this field are eager to solve. Summary of the invention

[0003] The present invention provides a fabric image processing device to solve at least one of the above problems.

[0004] One aspect of the present invention discloses a fabric image processing device, which includes an image capture circuit, a memory, and a processor. The image capture circuit is used to capture a first top view image and a first side view image of a first fabric. The memory is used to store a first physical deformation parameter corresponding to the first fabric. The processor is connected to the image capture circuit and the memory, and is used to execute a first neural network model, a second neural network model, a first linear regression model, a second linear regression model, and a third linear regression model, wherein the processor performs the following operations: respectively executing a background removal algorithm on the first top view image and the first side view image to generate a first top view silhouette image and a first side view silhouette image; updating the first neural network model and the first linear regression model according to the first top view silhouette image and the corresponding first physical deformation parameter, and updating the second neural network model and the second linear regression model according to the first side view silhouette image and the corresponding first physical deformation parameter; respectively inputting the first top view silhouette image and the first side view silhouette image into the first neural network model and the second neural network model to generate a first output vector and a second output vector; and concatenating the first output vector and the second output vector to generate a combined vector, and updating the third linear regression model according to the combined vector and the corresponding first physical deformation parameter.

[0005] In some embodiments, the first neural network model is connected in series before the first linear regression model, wherein the step of updating the first neural network model and the first linear regression model according to the first top-view silhouette image and the corresponding first physical deformation parameter includes: using the first top-view silhouette image and the first physical deformation parameter as a training sample and a training label, respectively, and using the training sample and the training label to update the parameters of the first neural network model and the first linear regression model.

[0006] In some embodiments, the second neural network model is connected in series before the second linear regression model, wherein the step of updating the second neural network model and the second linear regression model according to the first side-view silhouette image and the corresponding first physical deformation parameter includes: using the first side-view silhouette image and the corresponding first physical deformation parameter as a training sample and a training label, respectively, and using the training sample and the training label to update the parameters of the second neural network model and the second linear regression model.

[0007] In some embodiments, the third linear regression model is simultaneously connected to the first neural network model and the second neural network model, wherein the step of updating the third linear regression model according to the combined vector and the corresponding first physical deformation parameter includes: using the combined vector and the corresponding first physical deformation parameter as a training sample and a training label respectively, and using the training sample and the training label to update the parameters of the third linear regression model.

[0008] In some embodiments, the image capture circuit captures a second top view image and a second side view image corresponding to a second fabric, the processor executes the background removal algorithm on the second top view image and the second side view image, respectively, to generate a second top view silhouette image and a second side view silhouette image, and the processor converts the second top view silhouette image and the second side view silhouette image into a second physical deformation parameter through the updated first neural network model, the updated second neural network model, and the updated third linear regression model.

[0009] Another aspect of the present disclosure discloses a fabric image processing method, comprising: capturing a first top view image and a first side view image of a first fabric, and executing a background removal algorithm on the first top view image and the first side view image to generate a first top view silhouette image and a first side view silhouette image; updating a first neural network model and a first linear regression model according to the first top view silhouette image and the corresponding first physical deformation parameter, and updating a second neural network model and a second linear regression model according to the first side view silhouette image and the corresponding first physical deformation parameter; inputting the first top view silhouette image and the first side view silhouette image into the first neural network model and the second neural network model to generate a first output vector and a second output vector, respectively; and concatenating the first output vector and the second output vector to generate a combined vector, and updating a third linear regression model according to the combined vector and the corresponding first physical deformation parameter.

[0010] In some embodiments, the first neural network model is connected in series before the first linear regression model, wherein the step of updating the first neural network model and the first linear regression model according to the first top-view silhouette image and the corresponding first physical deformation parameter includes: using the first top-view silhouette image and the first physical deformation parameter as a training sample and a training label, respectively, and using the training sample and the training label to update the parameters of the first neural network model and the first linear regression model.

[0011] In some embodiments, the second neural network model is connected in series before the second linear regression model, wherein the step of updating the second neural network model and the second linear regression model according to the first side-view silhouette image and the corresponding first physical deformation parameter includes: using the first side-view silhouette image and the corresponding first physical deformation parameter as a training sample and a training label, respectively, and using the training sample and the training label to update the parameters of the second neural network model and the second linear regression model.

[0012] In some embodiments, the third linear regression model is simultaneously connected to the first neural network model and the second neural network model, wherein the step of updating the third linear regression model according to the combined vector and the corresponding first physical deformation parameter includes: using the combined vector and the corresponding first physical deformation parameter as a training sample and a training label respectively, and using the training sample and the training label to update the parameters of the third linear regression model.

[0013] In some embodiments, the fabric image processing method further includes: capturing a second top view image and a second side view image corresponding to a second fabric, executing the background removal algorithm on the second top view image and the second side view image to generate a second top view silhouette image and a second side view silhouette image, and converting the second top view silhouette image and the second side view silhouette image into a second physical deformation parameter through the updated first neural network model, the updated second neural network model and the updated third linear regression model.

[0014] The fabric image processing device disclosed in the present invention utilizes a combination of two neural network models and three linear regression models to train a model that can predict physical deformation parameters using the silhouette of the fabric. Thus, as long as the new silhouette corresponding to the new fabric is input into the two trained neural network models and one of the linear regression models, new physical deformation parameters for simulation can be generated. In this way, it is not necessary to measure various deformation variables of the new fabric in advance (which is more difficult to obtain), and the drape and movement of the virtual fabric can be directly simulated in the three-dimensional virtual space. This will save time and manpower for measuring physical deformation parameters separately, and greatly improve the effect of simulating virtual fabrics in the three-dimensional virtual space. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1A It is a schematic diagram showing the simulation of real fabrics in current textile technology.

[0016] Figure 1B is a schematic diagram illustrating a fabric image processing device according to some embodiments of the present disclosure.

[0017] Figure 2 FIG. 4 is a flowchart illustrating a fabric image processing method according to some embodiments of the present disclosure.

[0018] Figure 3A FIG. 4 is a schematic diagram illustrating a first top view image and a first side view image captured according to some embodiments of the present disclosure.

[0019] Figure 3B FIG. 4 is a schematic diagram illustrating a first top view silhouette image and a first side view silhouette image generated according to some embodiments of the present disclosure.

[0020] Figure 4A FIG. 4 is a schematic diagram illustrating training a first neural network model and a first linear regression model according to some embodiments of the present disclosure.

[0021] Figure 4B is a schematic diagram illustrating a trained second neural network model and a second linear regression model according to some embodiments of the present disclosure.

[0022] Figure 4CFIG. 4 is a schematic diagram illustrating training a third linear regression model according to some embodiments of the present disclosure.

[0023] Figure 5 A schematic diagram illustrating the use stages of a first neural network model, a second neural network model, and a third linear regression model in some embodiments.

[0024] [Explanation of symbols]

[0025] PDP: Physical Deformation Parameter

[0026] BWR, BWR': bending work in the warp direction

[0027] BWF, BWF': Weft bending work

[0028] SWR, SWR': tensile elongation in the warp direction

[0029] SWF, SWF': Weft tensile elongation

[0030] SO, SO': oblique tensile elongation

[0031] SS: Simulation software

[0032] SD: Analog Display

[0033] 3DS: Three-dimensional virtual space

[0034] 3DF: Virtual Fabric

[0035] 100: Fabric image processing device

[0036] 110: Image capture circuit

[0037] 120: Memory

[0038] 130: Processor

[0039] NNM1: first neural network model NNM2: second neural network model LRM1: first linear regression model LRM2: second linear regression model LRM3: third linear regression model S210-S240: steps

[0040] PIC1: First Overhead Image

[0041] PIC2: First side view image

[0042] FB: First Fabric

[0043] PLK: Round Plate

[0044] ND1, ND2: pointer

[0045] BS: Bracket

[0046] BW1: First top view silhouette

[0047] BW2: First side view silhouette

[0048] BR1, BR2: Black area

[0049] WR1, WR2: White area

[0050] VT1, VT2, VT1', VT2': output vector Output

[0051] y: first physical deformation parameter

[0052] CV: Combination Vector

[0053] PIC3: Second top view

[0054] PIC4: Second side view image

[0055] BW3: Second top view silhouette

[0056] BW4: Second side view silhouette

[0057] PDP': Second physical deformation parameter DETAILED DESCRIPTION

[0058] The following will describe multiple embodiments of the present invention with reference to the accompanying drawings. For the purpose of clarity, many practical details will be described together in the following description. However, it should be understood that these practical details should not be used to limit the present invention. That is, in some embodiments of the present invention, these practical details are not necessary. In addition, in order to simplify the drawings, some conventional structures and elements will be depicted in a simple schematic manner in the drawings; and repeated elements may be represented by the same number.

[0059] Example 1

[0060] Reference Figure 1A , Figure 1A It is a schematic diagram for simulating real fabrics in textile technology. In order to provide a close-to-real fabric simulation effect, physical deformation parameters PDP corresponding to the fabric (e.g., a piece of cloth or a piece of clothing) are usually required. Based on this, through simulation software SS (e.g., Scanatic TMA 3D simulation software such as DC Suite software can simulate a virtual 3D virtual space 3DS, and simulate various drapes and movements of a virtual fabric 3DF in the 3D virtual space 3DS according to the physical deformation parameter PDP. Thus, a simulation display SD (for example, a screen display, a touch display or a head mounted display) can display various drapes and movements of the virtual fabric 3DF in the 3D virtual space 3DS.

[0061] Generally speaking, a complicated measurement process is required to measure the physical deformation parameter PDP of the fabric, which can be the bending works warp work BWR, bending works weft work BWF, stretch warp elongation SWR, stretch weft elongation SWF or oblique stretch elongation SO, etc.

[0062] In detail, the warp bending work BWR is the work required to bend the fiber in the warp direction, and the unit of the warp bending work BWR is the product of gram-force, bending length and bending angle (gf×mm×rad). The weft bending work BWF is the work required to bend the fiber in the weft direction, and the unit of the weft bending work BWF is also the product of gram-force, bending length and bending angle.

[0063] In addition, the warp tensile elongation SWR is the elongation measured by the warp direction of the fiber under a constant load of 500 grams of force at a steady rate (CRE) according to the breaking strength bar method in the national standard CNS 12915 L3233-2010, and the unit of the warp tensile elongation SWR is percentage (%). The weft tensile elongation SWF is the elongation measured by the weft direction of the fiber under a constant load of 500 grams of force at a steady rate according to the breaking strength bar method in the national standard CNS 12915 L3233-2010, and the unit of the weft tensile elongation SWF is also percentage. The oblique tensile elongation SO is the elongation measured by the fiber in the oblique direction (the direction with an angle of 45 degrees with the weft direction) at a constant load of 500 grams of force at a steady rate according to the breaking strength bar method in the national standard CNS12915 L3233-2010, and the unit of the oblique tensile elongation SO is also a percentage. In other words, the warp bending work BWR, the weft bending work BWF, the warp tensile elongation SWR, the weft tensile elongation SWF and the oblique tensile elongation SO are each a numerical value.

[0064] The above-mentioned warp tensile elongation SWR, weft tensile elongation SWF and oblique tensile elongation SO are all calculated by the following formula (1).

[0065] (tensile displacement - clamping distance) / clamping distance × 100% (1)

[0066] The clamping distance is the distance from the starting clamping position to the ending clamping position on the fabric, and the stretching displacement is the stretching length of the fabric under a fixed load of 500 g.

[0067] However, as can be seen from the above, because the physical deformation parameter PDP often requires applying a quantitative force to the fabric and measuring the various deformation quantities produced by the fabric, in order to measure the warp bending work BWR, weft bending work BWF, warp tensile elongation SWR, weft tensile elongation SWF and oblique tensile elongation SO, the entire measurement process is complicated and time-consuming.

[0068] In view of the fact that the current general fabric simulation device may have poor simulation efficiency due to the complicated and time-consuming measurement process of the physical deformation parameter PDP. To solve the above problem, the present disclosure proposes a fabric detection technology, which first converts the top view image and the side view image corresponding to the fabric into a top view silhouette image and a side view silhouette image, and trains a neural network (NN) model and a linear regression model through the top view silhouette image, the side view silhouette image and the pre-measured physical deformation parameter PDP corresponding to the fabric.

[0069] In this way, when one wants to simulate a new fabric, one can obtain in advance a new side view image and a new top view image corresponding to the new fabric to convert the new side view image and the new top view image into a new top view silhouette image and a new side view silhouette image, and use the trained neural network model and linear regression model to convert the new top view silhouette image and the new side view silhouette image into new physical deformation parameters.

[0070] Therefore, various drapes and movements of a new virtual fabric in the three-dimensional virtual space 3DS can be simulated according to the new physical deformation parameters through the above detection method. For the new fabric, only a new top view silhouette image and a new side view silhouette image are needed to quickly identify the new physical deformation parameters to simulate the new fabric, which will save the complicated and time-consuming measurement process of the new physical deformation parameters. The above technology of the present disclosure is specifically described by taking the following embodiments as examples.

[0071] Refer to Figure 1B , Figure 1B FIG. 1 is a schematic diagram of a fabric image processing device 100 according to some embodiments of the present disclosure. Figure 1BAs shown, in this embodiment, the fabric image processing device 100 includes an image capture circuit 110 , a memory 120 , and a processor 130 . The processor 130 is coupled to the image capture circuit 110 and the memory 120 .

[0072] In some embodiments, the fabric image processing device 100 can be established by a computer, a server or a processing center. In some embodiments, the image capture circuit 110 can be a photographic circuit (e.g., a digital single-lens reflex camera (DSLR), a digital video camera (DVC) or a near-infrared camera (NIRC)) for photographing fabric (e.g., a whole piece of cloth). In some embodiments, the memory 120 can be implemented by a memory unit, a flash memory, a read-only memory, a hard disk or any equivalent storage component. In some embodiments, the processor 130 can be implemented by a processing unit, a central processing unit or a computing unit.

[0073] In this embodiment, the memory 120 stores the physical deformation parameters corresponding to the fabric in advance. In some embodiments, a detection machine (e.g., a machine for detecting the deformation (bending and stretching) of the fabric) can be used to perform stretching and bending detection on the fabric to generate the physical deformation parameters, and the detection machine can directly transmit the physical deformation parameters corresponding to the fabric to the memory 120 for storage. It is worth noting that although the physical deformation parameters corresponding to one fabric are used as an example here, in actual operation, the memory 120 can store multiple physical deformation parameters corresponding to multiple fabrics.

[0074] In this embodiment, if Figure 1B As shown, the processor 130 executes the first neural network model NNM1, the second neural network model NNM2, the first linear regression model LRM1, the second linear regression model LRM2 and the third linear regression model LRM3 based on corresponding software or firmware instruction programs. In some embodiments, the first neural network model NNM1 and the second neural network model NNM2 can both be used to perform a convolutional neural network (CNN) algorithm. In some embodiments, the first neural network model NNM1 and the second neural network model NNM2 can both be composed of a plurality of convolutional layers.

[0075] In some embodiments, the first linear regression model LRM1, the second linear regression model LRM2, and the third linear regression model LRM3 can all be regression output layers. In some embodiments, the memory 120 can store the parameters of the first neural network model NNM1, the second neural network model NNM2, the first linear regression model LRM1, the second linear regression model LRM2, and the third linear regression model LRM3, respectively, wherein these parameters can be average values ​​obtained from past training experience, manually given preset values, or random values.

[0076] The following further describes how to train a neural network model and a linear regression model based on a top view image corresponding to a fabric, a side view image conversion, and pre-measured physical deformation parameters corresponding to the fabric through an embodiment.

[0077] Refer to Figure 2 , Figure 2 FIG. 4 is a flowchart illustrating a fabric image processing method according to some embodiments of the present disclosure. Figure 1B The components in the fabric image processing device 100 are used to perform steps S210 to S240 in the fabric image processing method. Figure 2 As shown, first, in step S210, a first top view image and a first side view image of a first fabric (the fabric used for training is referred to as the first fabric herein) are captured, and a matting algorithm is performed on the first top view image and the first side view image, respectively, to generate a first top view silhouette image and a first side view silhouette image.

[0078] In some embodiments, the image capture circuit 110 can shoot the first fabric in a side view and a top view to generate a first top view image and a first side view image respectively. In some embodiments, the background removal algorithm can be an algorithm for identifying the outline of an object such as a k-means algorithm. In some embodiments, the image capture circuit 110 can shoot the first fabric placed in a green screen scene or other scenes in a side view and a top view to generate a first top view image and a first side view image respectively. In some embodiments, the processor 130 can remove the background and crop (remove a portion of the background area) the first top view image and the first side view image (for example, a pixel size of 4232×3204) by a background removal algorithm to generate a first top view silhouette image and a first side view silhouette image (for example, a pixel size of 224×224), wherein the first top view silhouette image and the first side view silhouette image can both be black and white images, wherein the white area of ​​the black and white image can be the object area of ​​the first fabric, and the black area of ​​the black and white image can be the background area.

[0079] The following is an actual example of the first top view image, the first side view image, the first top view silhouette image, and the first side view silhouette image. Figure 3A , Figure 3A 2 is a schematic diagram illustrating a first top view image PIC1 and a first side view image PIC2 captured in some embodiments of the present disclosure. Figure 3A As shown in Figures 3A1 and 3A2, the first top view image PIC1 and the first side view image PIC2 are images of the first fabric FB captured in the green screen scene. The first top view image PIC1 is a top view image captured from the -Z direction in the green screen scene, and the first side view image PIC2 is a side view image captured from the Y direction in the green screen scene.

[0080] In the first top view image PIC1, the circular plate PLK can be set above the first fabric FB (toward the Z direction), and the pointer ND1 can be set above the first fabric FB and the circular plate PLK in the horizontal direction (along the XY plane). In the first side view image PIC2, ​​the first fabric FB can be placed on the bracket BS (the first fabric FB will drape on the bracket BS, and the circular plate PLK is to block the non-draping part of the first fabric FB), and the pointer ND2 can be set below the first fabric FB in the vertical direction (along the Z direction). In addition, the pointer ND1 can be parallel to the weft direction of the fibers of the first fabric FB, and the pointer ND2 can be parallel to the warp direction of the fibers of the first fabric FB. In other words, the pointer ND1 can indicate the weft direction of the fibers of the first fabric FB, and the pointer ND2 can indicate the warp direction of the fibers of the first fabric FB.

[0081] Refer to Figure 3B , Figure 3B Schematic diagrams showing the first top view silhouette image BW1 and the first side view silhouette image BW2 generated in some embodiments of the present disclosure. Figure 3B As shown in Figures 3B1 and 3B2, Figure 3A The first top view silhouette image BW1 and the first side view silhouette image BW2 generated by the first top view image PIC1 and the first side view image PIC2 are both black and white images. In the first top view silhouette image BW1, the white area WR1 is the object area of ​​the first fabric FB, and the black area BR1 is the background area. In the first side view silhouette image BW2, the white area WR2 is the object area of ​​the first fabric FB, and the black area BR2 is the background area.

[0082] Furthermore, if Figure 2As shown, in step S220, the first neural network model NNM1 and the first linear regression model LRM1 are updated according to the first top view silhouette image and the corresponding first physical deformation parameters (the physical deformation parameters corresponding to the first fabric are referred to as the first physical deformation parameters herein), and the second neural network model NNM2 and the second linear regression model LRM2 are updated according to the first side view silhouette image and the corresponding first physical deformation parameters. In some embodiments, the first neural network model NNM1 may be connected in series before the first linear regression model LRM1, and the second neural network model NNM2 may be connected in series before the second linear regression model LRM2.

[0083] In some embodiments, the processor 130 may use the first top-view silhouette image and the first physical deformation parameter as a training sample and a training label, respectively, and use the training sample and the training label to update the parameters of the first neural network model NNM1 and the first linear regression model LRM1. In some embodiments, the processor 130 may update the parameters of the first neural network model NNM1 and the first linear regression model LRM1 according to the training sample and the training label through a backpropagation algorithm.

[0084] The following is an actual example to illustrate the update of the first neural network model NNM1 and the first linear regression model LRM1. Figure 4A , Figure 4A Schematic diagram of training a first neural network model NNM1 and a first linear regression model LRM1 according to some embodiments of the present disclosure. Figure 4A As shown, the first neural network model NNM1 and the first linear regression model LRM1 can be connected to each other. Then, the first top view silhouette image BW1 can be used as a training sample, and the first physical deformation parameter y can be used as a training label. Then, the first top view silhouette image BW1 can be input into the first neural network model NNM1 to generate an output vector VT1, and the output vector VT1 can be input into the first linear regression model LRM1 to generate an output result

[0085] Finally, the first physical deformation parameter y and the output result The loss value (e.g., mean square error (MSE)) is calculated, and the back propagation algorithm is performed using the loss value to update the parameters of the first neural network model NNM1 and the first linear regression model LRM1. It is worth noting that, for the sake of convenience of explanation, a first top view silhouette image BW1 of a first fabric and a first physical deformation parameter y are used as an example. However, in actual applications, a large number of top view silhouette images and physical deformation parameters of fabrics are used in the same way to train the first neural network model NNM1 and the first linear regression model LRM1.

[0086] In some embodiments, the processor 130 may use the first side view silhouette image and the corresponding first physical deformation parameter as training samples and training labels, respectively, and use the training samples and training labels to update the parameters of the second neural network model NNM2 and the second linear regression model LRM2. In some embodiments, the processor 130 may update the parameters of the second neural network model NNM2 and the second linear regression model LRM2 according to the training samples and training labels through a back propagation algorithm.

[0087] The following is an actual example to illustrate the update of the second neural network model NNM2 and the second linear regression model LRM2. Figure 4B , Figure 4B 2 is a schematic diagram illustrating the training of the second neural network model NNM2 and the second linear regression model LRM2 in some embodiments of the present disclosure. Figure 4B As shown, the second neural network model NNM2 and the second linear regression model LRM2 can be connected to each other. Then, the first side view silhouette image BW2 can be used as a training sample, and the first physical deformation parameter y can be used as a training label. Then, the first side view silhouette image BW2 can be input into the first neural network model NNM1 to generate an output vector VT2, and the output vector VT2 can be input into the second linear regression model LRM2 to generate an output result

[0088] Finally, the first physical deformation parameter y and the output result The loss value (e.g., root mean square error) is calculated, and the back propagation algorithm is performed using the loss value to update the parameters of the second neural network model NNM2 and the second linear regression model LRM2. It is worth noting that, for ease of explanation, a first side view silhouette image BW2 of a first fabric and a first physical deformation parameter y are used as an example. However, in actual applications, a large number of side view silhouette images and physical deformation parameters of fabrics are used in the same way to train the second neural network model NNM2 and the second linear regression model LRM2.

[0089] Furthermore, if Figure 2 As shown, in step S230, the first top view silhouette image and the first side view silhouette image are respectively input into the first neural network model NNM1 and the second neural network model NNM2 to generate a first output vector and a second output vector. In other words, when the first neural network model NNM1 and the second neural network model NNM2 are trained, the first top view silhouette image and the first side view silhouette image can be re-input into the trained first neural network model NNM1 and the second neural network model NNM2 to generate a first output vector and a second output vector.

[0090] Furthermore, in step S240, the first output vector and the second output vector are concatenated to generate a combined vector, and the third linear regression model is updated according to the combined vector and the corresponding first physical deformation parameter. In some embodiments, the third linear regression model LRM3 can be connected to the first neural network model NNM1 and the second neural network model NNM2 at the same time.

[0091] In some embodiments, the processor 130 may sequentially concatenate the first output vector and the second output vector to generate a combined vector. In some embodiments, the processor 130 may use the combined vector and the corresponding first physical deformation parameter as training samples and training labels, respectively, and use the training samples and training labels to update the parameters of the third linear regression model LRM3. In some embodiments, the processor 130 may update the parameters of the third linear regression model LRM3 according to the training samples and training labels through a back propagation algorithm.

[0092] The following is an actual example to illustrate the update of the third linear regression model LRM3. Figure 4C , Figure 4C FIG. 2 is a schematic diagram illustrating the training of the third linear regression model LRM3 in some embodiments of the present disclosure. Figure 4C As shown, the first neural network model NNM1 and the second neural network model NNM2 can be simultaneously connected to the third linear regression model LRM3. Then, the first top view silhouette image BW1 and the first side view silhouette image BW2 can be used as training samples, and the first physical deformation parameter y can be used as a training label. Then, the first top view silhouette image BW1 and the first side view silhouette image BW2 can be respectively input into the first neural network model NNM1 and the second neural network model NNM2 to generate an output vector VT1' and an output vector VT2', and the output vector VT1' and the output vector VT2' are sequentially concatenated to generate a combined vector CV.

[0093] Finally, the combined vector CV can be input into the third linear regression model LRM3 to produce the output result And according to the first physical deformation parameter y and the output result The loss value is calculated, and then the back propagation algorithm is performed using the loss value to update the parameters of the third linear regression model LRM3. It is worth noting that for the sake of explanation, a first top view silhouette image BW1, a first side view silhouette image BW2, and a first physical deformation parameter y of a first fabric are used as examples. However, in actual applications, a large number of top view silhouette images, side view silhouette images, and physical deformation parameters of fabrics are used in the same way to train the third linear regression model LRM3.

[0094] In some embodiments, the processor 130 may generate another combination vector according to the fabric specification information corresponding to the first fabric, and concatenate the other combination vector and the combination vector to generate a new combination vector. Then, the processor 130 may further update the parameters of the third linear regression model LRM3 using the new combination vector and the first physical deformation parameter. In some embodiments, the fabric specification information may include fabric category data, fabric combination data, and fabric weight data.

[0095] In some embodiments, fabric category data includes weaving category, elastic category, and fabric type. For example, weaving category can be various weaving specifications such as warp knit, weft knit, plain weave, or twill weave. Elastic category can be various fabric elastic specifications such as no elastic, 4-way elastic, warpelastic, or weft elastic. Fabric category can be fabric specifications such as dobby, single jersey fleece / plush, tricot fleece / tricot brush, or lamination. In other words, weaving category, elastic category, and fabric type are each a string of data.

[0096] In some embodiments, fabric combination data may include a composition combination and a textile finishing combination. For example, the composition combination may be a fiber combination specification such as 60% nylon fiber plus 40% polyester fiber, 92% polyester fiber plus 8% spandex fiber, 50% cationic polyester (cd polyester) fiber plus 50% polyester fiber or 77% nylon fiber plus 23% polyurethane (pu) fiber. The finishing combination may be a finishing specification such as no finishing, brushing treatment plus embossing treatment or brushing treatment. In other words, the composition combination and the finishing combination are data of a string composed of multiple strings.

[0097] In some embodiments, the fabric weight data may include fabric weight and specific gravity, wherein the unit of fabric weight is gram per square meter (GSM), and specific gravity is a ratio value between densities. In other words, fabric weight and specific gravity are numerical data respectively.

[0098] In some embodiments, the processor 130 may convert the fabric category data into a first vector through a target encoding algorithm and normalization, convert the fabric combination data into a second vector through a converter-based bidirectional encoder representation algorithm and a principal component analysis algorithm, and convert the fabric weight data into a third vector through normalization. Then, the processor 130 may concatenate the first vector, the second vector, and the third vector to generate another combined vector.

[0099] It is worth noting that although the above example is based on a physical deformation parameter, in actual application, different first neural network models NNM1, second neural network models NNM2, first linear regression models LRM1, second linear regression models LRM2 and third linear regression models LRM3 can be trained for different physical deformation parameters.

[0100] Through the above steps, the training of the first neural network model NNM1, the second neural network model NNM2, the first linear regression model LRM1, the second linear regression model LRM2 and the third linear regression model LRM3 can be completed (the training phase is completed). Thus, the trained first neural network model NNM1, the second neural network model NNM2 and the third linear regression model LRM3 can be used to detect the physical deformation parameters.

[0101] The following will describe the use stages of the first neural network model NNM1, the second neural network model NNM2 and the third linear regression model LRM3.

[0102] In some embodiments, the image capture circuit 110 can re-capture a second top view image and a second side view image corresponding to a second fabric (the new fabric is referred to herein as the second fabric) (i.e., photograph the new fabric), and the processor 130 can respectively execute a background removal algorithm on the second top view image and the second side view image to generate a second top view silhouette image and a second side view silhouette image.

[0103] In some embodiments, the processor 130 can convert the second top view silhouette image and the second side view silhouette image corresponding to the second fabric into second physical deformation parameters (i.e., the detected physical deformation parameters) through the first neural network model NNM1, the second neural network model NNM2 and the third linear regression model LRM3.

[0104] The following will illustrate the use of the first neural network model NNM1, the second neural network model NNM2 and the third linear regression model LRM3 with practical examples.

[0105] Refer to Figure 5 , Figure 5 A schematic diagram is shown showing the use phases of the first neural network model NNM1, the second neural network model NNM2 and the third linear regression model LRM3 in some embodiments. Figure 5 As shown, the background removal algorithm can be performed on the second top view image PIC3 and the second side view image PIC4 to generate the second top view silhouette image BW3 and the second side view silhouette image BW4. Then, the second top view silhouette image BW3 and the second side view silhouette image BW4 can be input into the trained first neural network model NNM1 and the second neural network model NNM2 respectively. Thereby, the second physical deformation parameter PDP' can be output from the trained third linear regression model LRM3, which can be the warp bending work BWR', the weft bending work BWF', the warp tensile elongation SWR', the weft tensile elongation SWF' or the oblique tensile elongation SO', etc.

[0106] Thus, by using the trained first neural network model NNM1, the second neural network model NNM2 and the third linear regression model LRM3, the second physical deformation parameter PDP' of the second fabric can be easily detected from the second top view silhouette image BW3 and the second side view silhouette image BW4 corresponding to the second fabric. Therefore, by using the trained first neural network model NNM1, the second neural network model NNM2 and the third linear regression model LRM3, the complicated machine detection steps can be omitted and the detection cost can be reduced. Thus, since the physical deformation parameters can be easily obtained, it will be easier to use Figure 1A The simulation display SD shows various drapes and movements of the virtual fabric 3DF in the three-dimensional virtual space 3DS.

[0107] In summary, the fabric image processing device disclosed in the present invention utilizes a combination of two neural network models and three linear regression models to train a model that can predict physical deformation parameters using the silhouette of the fabric. Thus, as long as the new silhouette corresponding to the new fabric is input into the two trained neural network models and one of the linear regression models, new physical deformation parameters for simulation can be generated. In this way, it is not necessary to measure various deformation variables of the new fabric in advance (which is more difficult to obtain), and the drape and movement of the virtual fabric can be directly simulated in the three-dimensional virtual space. This will save time and manpower for measuring physical deformation parameters separately, and greatly improve the effect of simulating virtual fabrics in the three-dimensional virtual space.

[0108] Although specific embodiments of the present disclosure have been disclosed with respect to the above embodiments, the above embodiments are not intended to limit the present disclosure. Various substitutions and improvements may be performed in the present disclosure by a person of ordinary skill in the relevant art without departing from the principles and spirit of the present disclosure. Therefore, the scope of protection of the present disclosure is determined by the scope of the attached patent applications.

Claims

1. A fabric image processing device, It is characterized in that include: An image capture circuit for capturing a first top view image and a first side view image of a first fabric; a memory for storing a first physical deformation parameter corresponding to the first fabric; and A processor is connected to the image capture circuit and the memory, and is used to execute a first neural network model, a second neural network model, a first linear regression model, a second linear regression model, and a third linear regression model, wherein the processor performs the following operations: Execute a background removal algorithm on the first top-view image and the first side-view image respectively to generate a first top-view silhouette image and a first side-view silhouette image; updating the first neural network model and the first linear regression model according to the first top-view silhouette image and the corresponding first physical deformation parameter, and updating the second neural network model and the second linear regression model according to the first side-view silhouette image and the corresponding first physical deformation parameter; Inputting the first top-view silhouette image and the first side-view silhouette image into the first neural network model and the second neural network model respectively to generate a first output vector and a second output vector; and The first output vector and the second output vector are serially connected to generate a combined vector, and the third linear regression model is updated according to the combined vector and the corresponding first physical deformation parameter.

2. The fabric image processing device according to claim 1, It is characterized in that The first neural network model is connected in series before the first linear regression model, and the step of updating the first neural network model and the first linear regression model according to the first top-view silhouette image and the corresponding first physical deformation parameter includes: The first top-view silhouette image and the first physical deformation parameter are used as a training sample and a training label respectively, and the training sample and the training label are used to update the parameters of the first neural network model and the first linear regression model.

3. The fabric image processing device according to claim 1, It is characterized in that The second neural network model is connected in series before the second linear regression model, and the step of updating the second neural network model and the second linear regression model according to the first side-view silhouette image and the corresponding first physical deformation parameter includes: The first side-view silhouette image and the corresponding first physical deformation parameter are used as a training sample and a training label respectively, and the training sample and the training label are used to update the parameters of the second neural network model and the second linear regression model.

4. The fabric image processing device according to claim 1, It is characterized in that The third linear regression model is connected to the first neural network model and the second neural network model at the same time, and the step of updating the third linear regression model according to the combined vector and the corresponding first physical deformation parameter includes: The combined vector and the corresponding first physical deformation parameter are used as a training sample and a training label respectively, and the parameters of the third linear regression model are updated by using the training sample and the training label.

5. The fabric image processing device according to claim 1, It is characterized in that in The image capture circuit captures a second top view image and a second side view image corresponding to a second fabric. The processor executes the background removal algorithm on the second top view image and the second side view image respectively to generate a second top view silhouette image and a second side view silhouette image, and The processor converts the second top-view silhouette image and the second side-view silhouette image into a second physical deformation parameter through the updated first neural network model, the updated second neural network model and the updated third linear regression model.

6. A fabric image processing method, It is characterized in that include: Capturing a first top view image and a first side view image of a first fabric, and executing a background removal algorithm on the first top view image and the first side view image to generate a first top view silhouette image and a first side view silhouette image; Updating a first neural network model and a first linear regression model according to the first top-view silhouette image and a corresponding first physical deformation parameter, and updating a second neural network model and a second linear regression model according to the first side-view silhouette image and the corresponding first physical deformation parameter; Inputting the first top-view silhouette image and the first side-view silhouette image into the first neural network model and the second neural network model respectively to generate a first output vector and a second output vector; and The first output vector and the second output vector are serially connected to generate a combined vector, and a third linear regression model is updated according to the combined vector and the corresponding first physical deformation parameter.

7. The fabric image processing method according to claim 6, It is characterized in that The first neural network model is connected in series before the first linear regression model, and the step of updating the first neural network model and the first linear regression model according to the first top-view silhouette image and the corresponding first physical deformation parameter includes: The first top-view silhouette image and the first physical deformation parameter are used as a training sample and a training label respectively, and the training sample and the training label are used to update the parameters of the first neural network model and the first linear regression model.

8. The fabric image processing method according to claim 6, It is characterized in that The second neural network model is connected in series before the second linear regression model, and the step of updating the second neural network model and the second linear regression model according to the first side-view silhouette image and the corresponding first physical deformation parameter includes: The first side-view silhouette image and the corresponding first physical deformation parameter are used as a training sample and a training label respectively, and the training sample and the training label are used to update the parameters of the second neural network model and the second linear regression model.

9. The fabric image processing method according to claim 6, It is characterized in that The third linear regression model is connected to the first neural network model and the second neural network model at the same time, and the step of updating the third linear regression model according to the combined vector and the corresponding first physical deformation parameter includes: The combined vector and the corresponding first physical deformation parameter are used as a training sample and a training label respectively, and the parameters of the third linear regression model are updated by using the training sample and the training label.

10. The fabric image processing method according to claim 6, It is characterized in that Also includes: capturing a second top view image and a second side view image corresponding to a second fabric, Executing the background removal algorithm on the second top view image and the second side view image respectively to generate a second top view silhouette image and a second side view silhouette image, and The second top-view silhouette image and the second side-view silhouette image are converted into a second physical deformation parameter through the updated first neural network model, the updated second neural network model and the updated third linear regression model.