3D Model Generation Method and Device

Abstract

The present application discloses a three-dimensional model generation method and apparatus. The method includes: performing a face replacement operation on a first rendered image according to a user's original image to obtain a first face-replaced image, and performing a face replacement operation on a second rendered image to obtain a second face-replaced image, where the first rendered image is generated by rendering a reference human body model under specific lighting, and the second rendered image is generated by rendering the reference human body model without lighting. Then, a target human body model is generated according to the first face-replaced image, the second face-replaced image, and the reference human body model. By using the method of the embodiment of the present application, the influence of unknown face light and shadow in the user's original image on the face model is reduced, and the clarity and accuracy of the facial features of the generated target human body model are improved.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method and device for generating a three-dimensional model. Background Art

[0002] The rise of artificial intelligence (AI) and the development of machine learning have led to the emergence of a variety of two-dimensional (2D) highly realistic face-swapping products. The popularity of AI has also spurred the development of three-dimensional (3D) technology, making it possible to create personalized, highly realistic 3D human models.

[0003] However, the current process for generating 3D human models typically uses a frontal facial photo of the user to create a facial model. Textures for the 3D facial model are then generated based on the user's facial photo. Finally, the generated 3D facial model and textures are fused onto the user's chosen body shape. While this can produce a 3D human model that resembles a real person, the lighting and texture of the facial texture are mixed with those of the original photo, making the facial features of the generated 3D human model less clear and accurate. Summary of the Invention

[0004] An embodiment of the present application provides a three-dimensional model generation method and device, which obtains an image generated by unilluminated rendering of a baseline human body model and an image generated by specific lighting rendering, then performs a face-swap operation on the user's original image with the two images to obtain two face-swapped images, and finally generates a target human body model based on the two face-swapped images and the baseline human body model. This process reduces the influence of unknown light and shadow on the user's original image's face on the facial model, and improves the clarity and accuracy of the facial features of the generated target human body model.

[0005] In a first aspect, a three-dimensional model generation method is provided, which includes: performing a face-changing operation on a first rendered image according to an original image of a user to obtain a first face-changing image, and performing a face-changing operation on a second rendered image to obtain a second face-changing image, wherein the first rendered image is generated by rendering a reference human body model with specific lighting, and the second rendered image is generated by rendering the reference human body model without lighting; generating a target human body model according to the first face-changing image, the second face-changing image and the reference human body model.

[0006] In an embodiment of the present application, when generating a target human body model, a first face-swapped image is obtained by swapping the face of a first rendered image with the user's original image, and then a second face-swapped image is obtained by swapping the face of a second rendered image with the user's original image. Then, a target human body model is generated based on the first face-swapped image, the second face-swapped image, and a reference human body model. Since the first rendered image is generated by rendering with specific lighting, the target human body mesh generated based on the first face-swapped image not only avoids being affected by unknown light and shadow features but also can ensure a better three-dimensional effect. The second rendered image is generated by rendering without lighting, so the facial texture generated based on the second face-swapped image can avoid being affected by the light and shadow features on the user's original face image. The entire process improves the clarity and accuracy of the facial features of the finally generated target human body model.

[0007] In a possible implementation, the first rendered image includes shadow features.

[0008] In a possible implementation, the face-swapping operation includes: positioning the rendered image to identify the position of the face in the rendered image; segmenting the rendered image according to the face position to obtain a first face image; replacing the facial features in the first face image with the user's original image to obtain a second face image, and fusing the second face image with the rendered image to obtain a face-swapped image. When the rendered image is the first rendered image, the face-swapped image is the first face-swapped image, and when the rendered image is the second rendered image, the face-swapped image is the second face-swapped image.

[0009] In a possible implementation, generating a target human body model based on the first face-swapped image, the second face-swapped image, and a reference human body model includes: obtaining a facial three-dimensional model according to the first face-swapped image; deforming the facial mesh of the reference human body based on the facial three-dimensional model and the reference human body model to obtain a target human body mesh; and performing facial texture synthesis based on the second face-swapped image and the target human body mesh to obtain the target human body model.

[0010] In a possible implementation, obtaining a facial three-dimensional model corresponding to the user's original image according to the first face-swapped image includes: matching the first face-swapped image with basis vectors in multiple directions in a three-dimensional face model to obtain coefficients corresponding to the basis vectors in multiple directions of the first face-swapped image, where the basis vectors are basic elements constituting a vector space; and obtaining the facial three-dimensional model corresponding to the user's original image based on the basis vectors in multiple directions and the coefficients corresponding to the basis vectors in multiple directions.

[0011] In a possible implementation, face texture synthesis is performed based on the second face-swapped image and the target human body mesh to obtain a target human body model, including: obtaining a planar projection image of the target human body mesh, and determining a first correspondence between multiple patches of the planar projection image and multiple patches of the second face-swapped image; obtaining a texture map corresponding to the target human body mesh; determining the positions of multiple patches of the planar projection image corresponding to the UV texture mapping coordinates according to the texture map, and determining the positions of multiple patches of the second face-swapped image corresponding to the UV texture mapping coordinates according to the first correspondence; determining the pixel values of the positions of the VU texture mapping coordinates corresponding to multiple patches of the second face-swapped image according to the pixel values of multiple patches of the second face-swapped image, to obtain a face texture map; and replacing the texture map corresponding to the target human body mesh with the face texture map to obtain the target human body model.

[0012] In a possible implementation, the first rendered image and the second rendered image are a single image based on a frontal face or multiple images based on multi-angle side faces.

[0013] In the embodiments of the present application, using a single frontal face rendered image to obtain a face-swapped image for the subsequent target human body model generation process can improve the model generation efficiency; using multiple side face rendered images at different angles to obtain a face-swapped image for the subsequent target human body model generation process can provide more face regions, so that the texture information of the texture map in the side face region is more. And it can avoid the stretching of the generated face texture in the side face region, reduce the resource consumption for processing the stretching situation, and improve the model generation accuracy at the same time.

[0014] In a second aspect, a three-dimensional model generation device is provided. The device includes an image processing unit and a model generation unit, where,

[0015] The image processing unit is configured to perform a face-swapping operation on the first rendered image to obtain a first face-swapped image and perform a face-swapping operation on the second rendered image to obtain a second face-swapped image according to the user's original image, where the first rendered image is generated by rendering a reference human body model under specific lighting, and the second rendered image is generated by rendering the reference human body model without lighting;

[0016] The model generation unit is configured to generate a target human body model according to the first face-swapped image, the second face-swapped image, and the reference human body model.

[0017] In a possible implementation, the image processing unit is specifically configured to, during the face-swapping operation:

[0018] Locate the rendered image and identify the face position in the rendered image;

[0019] Segment the rendered image according to the face position to obtain a first face image;

[0020] The facial features in the first facial image are replaced according to the user's original image to obtain a second facial image, and the second facial image is fused with the rendered image to obtain a face-swapped image, wherein when the rendered image is the first rendered image, the face-swapped image is the first face-swapped image, and when the rendered image is the second rendered image, the face-swapped image is the second face-swapped image.

[0021] In one possible implementation, the facial features replaced in the face-changing operation do not include skin features.

[0022] In a possible implementation, the model generation unit is specifically configured to:

[0023] Acquire a three-dimensional facial model according to the first face-changing image;

[0024] Deforming the reference human face mesh according to the three-dimensional face model and the reference human body model to obtain a target human body mesh;

[0025] Facial texture synthesis is performed based on the second face-swapped image and the target human body mesh to obtain a target human body model.

[0026] In a possible implementation, the model generation unit is further specifically configured to:

[0027] Matching the first face-swapped image with basis quantities in multiple directions in the three-dimensional face model to obtain coefficients corresponding to the basis quantities in the first face-swapped image in the multiple directions, where the basis quantities are basic elements constituting the vector space;

[0028] A three-dimensional facial model corresponding to the original image of the user is obtained according to the basis quantities in multiple directions and the coefficients corresponding to the basis quantities in multiple directions.

[0029] In a possible implementation, the model generation unit is further specifically configured to:

[0030] Acquire a planar projection image of the target human body mesh, and determine a first correspondence between a plurality of face patches of the planar projection image and a plurality of face patches of the second face-swapped image;

[0031] Get the texture map corresponding to the target human body mesh;

[0032] Determining positions of the plurality of face patches of the planar projection image corresponding to UV texture mapping coordinates according to the texture map, and determining positions of the plurality of face patches of the second face-swapped image corresponding to UV texture mapping coordinates according to the first corresponding relationship;

[0033] Determine the pixel values of the corresponding VU texture map coordinate positions according to the pixel values of the plurality of face patches of the second face-swapped image, and obtain a face texture map;

[0034] The texture map corresponding to the target human body mesh is replaced with the face texture map to obtain the target human body model.

[0035] In a possible implementation, the first rendered image and the second rendered image are single images based on a frontal face or multiple images based on multi-angle profile faces.

[0036] In a third aspect, an embodiment of the present application provides an electronic device, which includes a communication interface and a processor. The communication interface is used for the device to communicate with other devices, such as the transceiver of data or signals. Exemplarily, the communication interface can be a transceiver, a circuit, a bus, a module, or other types of communication interfaces, and the other device can be a network device. The processor is used to call a set of programs, instructions, or data to execute the method described in the first aspect above. The device may further include a memory for storing the programs, instructions, or data called by the processor. The memory is coupled to the processor, and when the processor executes the instructions or data stored in the memory, the method described in the first aspect above can be implemented.

[0037] In a fourth aspect, an embodiment of the present application further provides an electronic device, characterized in that the communication device includes a processor, a transceiver, a memory, and computer-executable instructions stored on the memory and executable on the processor. When the computer-executable instructions are run, the communication device executes the method in the first aspect or any possible implementation manner in the first aspect.

[0038] In a fifth aspect, an embodiment of the present application further provides a computer-readable storage medium, in which computer-readable instructions are stored. When the computer-readable instructions are run on a computer, the computer executes the method in the first aspect or any possible implementation manner in the first aspect.

[0039] In a sixth aspect, an embodiment of the present application provides a chip system, which includes a processor and may further include a memory for implementing the method in the first aspect or any possible implementation manner in the first aspect. The chip system may be composed of chips or may include chips and other discrete devices.

[0040] Optionally, the chip system further includes a transceiver.

[0041] In a seventh aspect, an embodiment of the present application further provides a computer program product, including instructions, which when run on a computer, cause the computer to execute the method in the first aspect or any possible implementation manner in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments.

[0043] Figure 1ASchematic diagram of a process for generating a three-dimensional human body model provided by an embodiment of the present application;

[0044] Figure 1B Schematic diagram of a process for generating a three-dimensional human body model provided by an embodiment of the present application;

[0045] Figure 2A Flowchart of a three-dimensional model generation method provided by an embodiment of the present application;

[0046] Figure 2B Schematic diagram of a process for obtaining a rendered image provided by an embodiment of the present application;

[0047] Figure 3A Flowchart of a face swapping operation provided by an embodiment of the present application;

[0048] Figure 3B Schematic diagram of a CNN face conversion network structure provided by an embodiment of the present application;

[0049] Figure 3C Schematic diagram of a face swapping implementation process provided by an embodiment of the present application;

[0050] Figure 3D Schematic diagram of a face swapping process provided by an embodiment of the present application;

[0051] Figure 4A Flowchart of a process for generating a target human body model provided by an embodiment of the present application;

[0052] Figure 4B Schematic diagram of a process for obtaining a three-dimensional face model through 3DMM provided by an embodiment of the present application;

[0053] Figure 4C Schematic diagram of a process for obtaining a target human body mesh provided by an embodiment of the present application;

[0054] Figure 4D Schematic diagram of a process for generating a facial texture map provided by an embodiment of the present application;

[0055] Figure 4E Schematic diagram of a facial texture synthesis process provided by an embodiment of the present application;

[0056] Figure 5 Block diagram of a three-dimensional model generation device provided by an embodiment of the present application;

[0057] Figure 6 Schematic diagram of the structure of an electronic device provided by an embodiment of the present application.[[ID=�7]] Detailed implementation manners

[0058] In the description, claims and drawings of the present application, terms such as "first", "second", "third" and "fourth" are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or modules is not limited to the listed steps or modules, but optionally further includes steps or modules not listed, or optionally further includes other steps or modules inherent to these processes, methods, products or devices.

[0059] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in connection with the embodiments can be included in at least one embodiment of the present application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive of other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0060] "Plurality" means two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships can exist. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0061] The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application.

[0062] The method of the embodiment of the present application can be implemented on a terminal. The terminal can be referred to as terminal equipment, user equipment (UE), mobile station (MS), mobile terminal (MT), etc. The terminal equipment can be a mobile phone, a tablet computer (Pad), a computer with wireless transceiver function, a virtual reality (VR) terminal device, an augmented reality (AR) terminal device, a wireless terminal in industrial control, a wireless terminal in self-driving, a wireless terminal in remote medical surgery, a wireless terminal in smart grid, a wireless terminal in transportation safety, a wireless terminal in smart city, a wireless terminal in smart home, and so on.

[0063] The method of the embodiment of the present application can be specifically applied to the fields related to the generation of high-fidelity digital humans in virtual reality and augmented reality scenarios. For example, augmented reality (AR) navigation, AR games, or AR clothing try-on, etc. The specific application process can refer to Figure 1A , Figure 1A which is a schematic diagram of the process of generating a three-dimensional human body model provided by the embodiment of the present application. As Figure 1A shown, the terminal 111 acquires the user's original image 110, for example, receives it from other devices or obtains it by shooting with an image acquisition component integrated in itself. Then, the terminal 111 performs a face swapping operation on the acquired other images or parameters and the user's original image 110, and further obtains a three-dimensional face model and a face texture map based on the face-swapped image, and finally obtains the output three-dimensional human body model 112.

[0064] In the process of acquiring the three-dimensional human body model described above, since the user's original image usually includes some shadows formed by unknown light. Specifically, please refer to Figure 1B , Figure 1B which is a schematic diagram of the process of generating a three-dimensional human body model provided by the embodiment of the present application. As Figure 1B shown, the original face image 120 includes shadows formed by unknown light, and the three-dimensional model 121 generated based on this image also includes these shadows, which makes the facial features of the generated three-dimensional model unclear or the authenticity low, reducing the quality of the three-dimensional model.

[0065] Based on the above description, please refer toFigure 2A , Figure 2A is a flowchart of a three-dimensional model generation method provided by an embodiment of this application. As Figure 2A shown, the method includes the following steps:

[0066] 201. The terminal performs a face replacement operation on the first rendered image based on the user's original image to obtain a first face-replaced image, and performs a face replacement operation on the second rendered image to obtain a second face-replaced image, where the first rendered image is generated by rendering a reference human body model under specific lighting, and the second rendered image is generated by rendering the reference human body model without lighting;

[0067] 202. The terminal generates a target human body model based on the first face-replaced image, the second face-replaced image, and the reference human body model.

[0068] The user's original image refers to an image obtained by photographing a person with a camera, or a video stream obtained by shooting with a camera, and then segmenting to obtain a video image. These user's original images include human faces. Since the acquisition condition of the user's original image is natural light, there are unknown-angle and unknown-intensity lighting and shadow features on these images.

[0069] Obtaining a rendered image is a process of generating a two-dimensional image by performing geometric transformation, projection transformation, perspective transformation, and window clipping on a three-dimensional model, and then through the obtained material and light and shadow information. The shading model describes how the color of a rendered object changes based on coefficients such as surface orientation, view direction, and lighting. In other words, the shading model can control the input data constituting the material to create the final appearance. The reference human body model is the basis for generating the target human body model. By performing corresponding transformations on the face model of the reference human body model, the target human body model can be obtained.

[0070] The reference human body model is rendered under specific lighting to obtain the first rendered image. The specific lighting can be ambient light, parallel light, point light, or other light. The reference human body model irradiated by specific lighting is a kind of lighting shading model, which uses direct and indirect lighting, as well as reflected highlights. That is to say, on the first rendered image generated by irradiating the reference human body model with specific lighting and then rendering, it includes the color image information formed by the reference human body model reflecting the specific lighting, and the shadow information that may be generated when the specific lighting irradiates the surface of the reference human body model.

[0071] For the second rendered image, it is an image generated by rendering the reference human body model without light. The reference human body model without light illumination is a non-lighting shading model that only outputs the self-emission of the model color. Therefore, the second rendered image only includes the texture color of the self-emission of the model itself and does not include any shadow information.

[0072] In addition, assuming that a specific light source is an ideal point light source, then an obvious boundary shadow may be formed after the target human body model is illuminated. Assuming that a specific light source is an area light source with a certain size, then a shadow with the model boundary may be formed after the target human body model is illuminated. Corresponding to the first rendered image, the boundary between the shadow-rendered image and the human body model-rendered image can also be identified by a certain method, and then it can be determined whether the obtained rendered image is generated by rendering with a specific light source, and further distinguished as the first rendered image or the second rendered image. The method used can be, for example, a convolutional neural network model or other methods for determining that the image includes shadows.

[0073] Since the first rendered image contains shadow information, it can make geometric features appear on the image, such as the distances, areas, and angles between the facial features, thereby highlighting the three-dimensional effect of the facial features of the reference human body model. Then, when obtaining the three-dimensional facial model based on the first rendered image, a more three-dimensional three-dimensional model can also be obtained.

[0074] The second rendered image does not contain shadow information and only includes the texture color of the self-emission of the model itself. Then, when obtaining the texture information by obtaining the gray difference between adjacent pixel points based on the second rendered image (specifically, the second face-swapped image obtained after performing a face-swapping operation on the second rendered image in the embodiments of the present application, and the face-swapping operation does not affect the image texture information), the influence caused by the shadow color information can be avoided, and the accuracy of the obtained facial texture map can be improved.

[0075] Optionally, the conditions that specific lighting needs to meet may include: (1) including a front-direction light source. If the specific lighting does not include a front light source, the rendering of the facial features of the entire face will be overall dark and lack a three-dimensional sense. (2) The rendered image has no large-area shadows. If the first rendered image rendered by the specific lighting includes a large-area shadow effect, it may interfere with the geometric features of the image and is not conducive to subsequent face swapping operations and face reconstruction. Therefore, for condition (1) that specific lighting needs to meet, it can be adjusted before rendering the image, and for condition (2) that needs to be met, it can be adjusted in real time during the process of generating the rendered image so that the first rendered image meets the feature of having no large-area shadows. That is to say, by controlling the characteristics of the specific lighting, such as the irradiation distance and irradiation angle of the specific lighting, the light and shadow colors in the first rendered image can be controlled, and then the corresponding geometric features can be controlled. The second rendered image is generated by rendering the reference human body model without lighting, that is to say, when generating the second rendered image according to the reference human body model, it only includes the self-luminous color of the reference human body model and does not include the light and shadow colors formed by any external light irradiating on the reference human body model. Please refer to Figure 2B , Figure 2B which is a schematic diagram of the process of obtaining a rendered image provided by an embodiment of the present application. As shown in (a) in Figure 2B , the first rendered image 211 is generated by rendering the reference human body model 210 with parallel light, and as shown in (b) in Figure 2B , the second rendered image 212 is generated by rendering the reference human body model 210 without lighting. Optionally, the rendered image can be a frontal face image, and the complete facial features of the user can be displayed in this frontal face image. Or the collected rendered images can be multiple profile face images. Usually, the side light of the reference human body model within the range of ±30° to 90° from the center of the face is used for facial lighting, and facial images are rendered from three perspectives of left, middle, and right (typical values are left 45°, 0°, and right 45°) to obtain a multi-angle rendered image of the side light of the reference model. These profile face images combined can display the complete facial features of the user, and at the same time, these profile face images can provide more facial features, such as profile textures, jaw lines, etc.

[0076] After obtaining the rendered image, the AI face swapping algorithm can be used to perform a face swapping operation on the rendered image and the user's original image to generate a face swapped image, so as to generate a target human body model according to the face swapped image subsequently.

[0077] Please refer to Figure 3A , Figure 3A which is a flowchart of a face swapping operation provided by an embodiment of the present application. As shown in Figure 3A , the face swapping process of the first rendered image may include the following steps:

[0078] In 2011, locate the first rendered image and identify the position of the face in the first rendered image;

[0079] In 2012, segment the first rendered image based on the face position to obtain the first face image;

[0080] In 2013, replace the face features in the first face image according to the user's original image to obtain the second face image;

[0081] In 2014, fuse the second face image with the rendered image to obtain the face-swapped image.

[0082] For the first rendered image and the second rendered image, the face-swapping operation process is the same. In this embodiment of the application, the first rendered image is taken as an example for specific description. In the face-swapping process, it is actually the replacement of face features. Therefore, during the face-swapping operation, it is necessary to first extract face features and train to obtain a face conversion network, and then input the first face image obtained by segmenting the first rendered image and the user's original face image into the face conversion network to complete the conversion from the first face image to the second face image. Among them, face features can include geometric features and characterization features of the face. Geometric features are specifically the distances, areas, and angles between the facial features of the face, and the characterization feature is to pass the face image through a neural network to obtain a feature vector with a specific dimension, and this vector can well represent the face data.

[0083] Taking the face-swapping algorithm of Convolutional Neural Networks (CNN) as an example, the first face image segmented from the first rendered image can be obtained through the CNN face localization algorithm first, and then the first face image is replaced with the second face image according to the CNN face conversion network. For details, please refer to Figure 3B , Figure 3B is a schematic diagram of the CNN face conversion network structure. As shown in Figure 3B , this network is mainly composed of an encoder network and a decoder network. First, the face conversion network is trained by inputting a real face image A, so that the encoder network in the face conversion network can perform feature extraction processing on the face image A to obtain face feature A, and the decoder network can restore face feature A to obtain the original face image A. In the test stage, the original face image B is input into the trained face conversion network. The encoder network in it can perform feature extraction on the original face image B to obtain face feature B. The decoding network includes face feature A. Face feature A is used to replace face feature B, and the image after replacing the face feature is restored to obtain the face-swapped image B including face feature A, that is, the face swapping of the original face image B is realized.

[0084] In step 2013 of the embodiment of the present application, specifically, the original face image is trained to obtain a decoder network corresponding to the original face image, and then the first face image obtained by segmenting the first rendered image is input into the decoder network, so that the decoding network corresponding to the original face image reconstructs the first face image to obtain a second face image. The second face image actually includes the facial features of the original face image. Finally, the second face image is fused with other features of the first rendered image. The other features can be, for example, hairstyle features, earring features, neck features, etc. For example, please refer to Figure 3C , Figure 3C which is a schematic diagram of a face swapping implementation process provided by the embodiment of the present application. As shown in Figure 3C , the corresponding facial features obtained by encoding the original face image A, abbreviated as facial features A, are input into the decoder network B, and a face image B reconstructed according to the face image A can be obtained. Its corresponding facial features such as facial features, hair, etc. are all similar to those of the face image B (determined by the corresponding decoder network), and only the mouth curvature is similar to that of the face A. Similarly, the facial features obtained by encoding the original face image B, abbreviated as facial features B, are input into the decoder network A, and a face image A reconstructed according to the face image B can be obtained. Its corresponding facial features, hairstyle, etc. are all similar to those of the face A, and only the mouth curvature is similar to that of the face B. This process completes the face swapping between the original face image A and the original face image B. Finally, the face obtained after face swapping is fused with other features in the original image, such as being fused with the background, clothing, etc., to complete the face swapping operation.

[0085] Specifically, since the facial texture is located on the human face skin and the two are closely combined, in the embodiment of the present application, in order to avoid the possible influence of the shadow features included in the original face image on the generation of facial texture and the generation of the 3D human face model, during the face swapping process, the skin features on the human face can be not replaced. That is to say, when replacing the face image in the first rendered image or the second rendered image with the original face image, the facial skin features in the first rendered image and the second rendered image are not replaced. Then, the facial skin rendered under specific illumination is maintained in the first face-swapped image obtained after face swapping according to the first rendered image, and the facial skin without illumination rendering is maintained in the second face-swapped image obtained after face swapping according to the second rendered image.

[0086] According to the foregoing description, the first rendered image can be a rendered image obtained by rendering the front face of the reference human body model, or multiple rendered images obtained by rendering the side faces of the reference human body model at multiple different angles. Similarly, when performing the face swapping operation, it may be a face swapping of a single front face rendered image, or a face swapping of multiple side face rendered images. For specific reference, please refer to Figure 3D , Figure 3D which is a schematic diagram of a face swapping process provided by the embodiment of the present application. As shown in Figure 3DAs shown in (a) therein, when the first rendered image (or the second rendered image) is a front-face rendered image, only this one rendered image needs to be subjected to a face-swapping operation with the user's original face image to obtain a first face-swapped image (the one corresponding to the second rendered image is the second face-swapped image). Or as Figure 3D As shown in (b) therein, when the first rendered image (or the second rendered image) is multiple side-face rendered images, these multiple rendered images need to be subjected to a face-swapping operation with the user's original face image to obtain multiple first face-swapped images (or multiple second face-swapped images). In the subsequent processing, if one first face-swapped image is obtained, a target human body mesh is obtained based on this one face-swapped image; if multiple first face-swapped images are obtained, a target human body mesh is obtained based on these multiple face-swapped images. Similarly, if one second face-swapped image is obtained, a facial texture is obtained based on this one face-swapped image; if multiple second face-swapped images are obtained, a facial texture is obtained based on these multiple second face-swapped images. During the process of generating a target human body model by the terminal, obtaining a target human body mesh based on a single or multiple first face-swapped images and obtaining a facial texture based on a single or multiple second face-swapped images can be combined arbitrarily. For example, a target human body network can be obtained based on a single first face-swapped image, and then a facial texture can be obtained based on multiple second face-swapped images. Finally, the obtained target human body mesh and facial texture are synthesized to obtain the final target human body model.

[0087] In the above process, using a single front-face rendered image to obtain a face-swapped image for the subsequent target human body model generation process can improve the model generation efficiency; using multiple side-face rendered images with different angles to obtain face-swapped images for the subsequent target human body model generation process can provide more face regions, making the texture information in the side-face region of the texture map more. And it can avoid the stretching situation of the generated facial texture in the side-face region, reduce the resource consumption for dealing with the stretching situation, and at the same time improve the model generation accuracy.

[0088] It can be seen that in the embodiment of the present application, when generating a target human body model, a first face-swapped image is obtained by performing a face-swapping operation on the first rendered image and the user's original image, then a second face-swapped image is obtained by performing a face-swapping operation on the second rendered image and the user's original image, and then a target human body model is generated based on the first face-swapped image, the second face-swapped image and a reference human body model. Among them, since the first rendered image is generated by rendering with specific lighting, the target human body mesh generated based on the first face-swapped image not only avoids being affected by unknown light and shadow features but also can ensure a better three-dimensional effect. The second rendered image is generated by rendering without lighting, so the facial texture generated based on the second face-swapped image can avoid the influence of the light and shadow features on the user's original face image. The entire process improves the clarity and accuracy of the facial features of the finally generated target human body model.

[0089] Optionally, the process of generating the target human body model based on the first face-swapped image, the second face-swapped image, and the reference human body model can be specifically described. For example, Figure 4A As shown, generating the target human body model based on the first face-swapped image, the second face-swapped image, and the reference human body model includes the following steps:

[0090] 2021. Obtain the three-dimensional face model according to the first face-swapped image;

[0091] 2022. Perform reference human body face mesh deformation based on the three-dimensional face model and the reference human body model to obtain the target human body mesh;

[0092] 2023. Perform face texture synthesis according to the second face-swapped image and the target human body mesh to obtain the target human body model.

[0093] Specifically, the first face-swapped image is a two-dimensional human face image including specific lighting. First, face three-dimensional reconstruction needs to be performed according to this two-dimensional human face image. For example, self-supervised monocular 3D face reconstruction (Self-Supervised Monocular 3D Face Reconstruction, MGCnet), or reconstruction algorithms such as PRnet can be used for face reconstruction. Taking MGCnet as an example, this algorithm mainly uses a deep learning network to estimate the parameters of the three-dimensional deformable human face model (3D Morphable Models, 3DMM) to obtain the three-dimensional face model. Please refer to Figure 4B , Figure 4B which is a schematic diagram of the process of obtaining the three-dimensional face model through 3DMM provided by the embodiment of the present application. As shown in Figure 4B , first, model training is performed according to the 3D face model dataset to obtain a deformable face model, then face feature analysis is performed on the first face-swapped image to determine the coefficients corresponding to the first face-swapped image on each basis vector relative to the deformable face model, including the face shape coefficient and the face texture coefficient, and then the shape and texture of the deformable face model are adjusted according to these coefficients to obtain the three-dimensional face model corresponding to the finally output first face-swapped image.

[0094] Specifically, the deformable face model is a general model that represents a face with a fixed number of points. Its core idea is that a face can be matched one by one in three-dimensional space, and a general model can be obtained by linearly adding the weighted orthogonal bases of multiple faces. In the three-dimensional space we are in, each point (x, y, z) is actually obtained by linearly adding the basis vectors in three directions of the three-dimensional space, (1, 0, 0), (0, 1, 0), and (0, 0, 1), with the weights being x, y, and z respectively. Similarly, each three-dimensional face can be represented in the basis vector space composed of all the faces in a database, and solving the model of any three-dimensional face is actually equivalent to solving the problem of the coefficients of each basis vector. The basic attributes of a face include shape and texture, and each face can be represented as a linear superposition of a shape vector and a texture vector.

[0095] After obtaining the three-dimensional model of the user's face according to the first face-swapping image, it is further necessary to perform benchmark human face mesh deformation based on the three-dimensional model of the face and the benchmark human body model, that is, to restore the three-dimensional model of the face to the benchmark human body model to obtain the target human body mesh. The specific process can be referred to Figure 4C , Figure 4C which is a schematic diagram of the process of obtaining the target human body mesh provided by the embodiment of the present application. As Figure 4C shown, first, the three-dimensional model of the user's face is restored to the benchmark human body model through rough registration to obtain a preliminary human body model, and then the overall face radial basis function (Radial Basis Function, RBF) deformation and the overall face non-rigid iterative closest point (Non-rigid iteration closest point, NICP) deformation are performed on the preliminary human body model. The RBF deformation can achieve rapid convergence of the deformation result, and the NICP deformation can obtain a more accurate deformation result. After completing the overall face deformation, the overall eye deformation is performed, and the overall eye RBF deformation can also be used. Then, detailed deformation is performed, including RBF deformation of details such as double eyelids. Then, the overall smoothing of the face and eyes is performed, and RBF deformation is performed on the transition zone. Finally, the model position of the eyeball is adjusted. This process gradually and accurately deforms the benchmark human body model (mainly the benchmark face model) to the three-dimensional model of the user's face, realizing seamless connection between the face and other parts of the head while maintaining the geometric topology of the benchmark human body model.

[0096] Finally, obtain the facial texture of the user based on the second face-swapped image, and realize the facial texture synthesis of the target human body mesh, so as to obtain the target human body model. First, the planar projection image of the target human body mesh can be obtained. The planar projection image is a 2D image. The target human body mesh is projected onto the 2D plane according to a certain pose and projection matrix, and can coincide with the second face-swapped image (which is also a 2D image). In addition, there are several patches on the target human body mesh for dividing the sub-regions of the human body mesh (the geometric shapes corresponding to the sub-regions divided by the patches can be triangles, parallelograms, etc.). These patches can be correspondingly mapped onto the planar projection image, and the planar projection image and the second face-swapped image can have a one-to-one correspondence of sub-regions based on the patches (that is, the second face-swapped image is also divided into patches, and the patches of the second face-swapped image correspond one-to-one to the patches of the planar projection image). This kind of correspondence is called the first correspondence. Further, reference can be made to Figure 4D , Figure 4D which is a schematic diagram of the process of generating a facial texture map provided by an embodiment of the present application. Project the target human body mesh into the UV texture mapping coordinates to generate a texture map. In fact, it is also to map the planar projection map of the target human body mesh into the UV coordinates to generate a texture map. Then, determine the positions of multiple patches of the planar projection image corresponding to the UV texture mapping coordinates according to the texture map, and then determine the positions of multiple patches of the second face-swapped image corresponding to the UV texture mapping coordinates according to the first correspondence obtained above; determine the pixel values of the positions of the VU texture mapping coordinates corresponding to multiple patches of the second face-swapped image according to the pixel values of multiple patches of the second face-swapped image. For example, directly use the pixel values of multiple patches of the second face-swapped image as the pixel values of the corresponding UV texture mapping coordinate positions. Finally, complete the filling (or replacement) of the pixel values at all UV coordinate positions of the original texture map to obtain the final facial texture map, as shown in (1) of Figure 4E ; finally, replace the texture map corresponding to the target human body mesh with the facial texture map, that is, synthesize the target human body mesh ( Figure 4E in (2)) with the facial texture map to obtain the target human body model ( Figure 4E in (3)).

[0097] Figure 5 A three-dimensional model generation device 500 provided by an embodiment of the present application can be used to execute the above-mentioned Figures 2A to 4E three-dimensional model generation method and specific embodiments applied to a terminal. The device includes an image processing unit 501 and a model generation unit 502.

[0098] An image processing unit 501 is configured to perform a face swapping operation on a first rendered image based on a user's original image to obtain a first face-swapped image, and perform a face swapping operation on a second rendered image to obtain a second face-swapped image, where the first rendered image is generated by rendering a reference human body model under specific lighting, and the second rendered image is generated by rendering the reference human body model without lighting; a model generation unit 502 is configured to generate a target human body model based on the first face-swapped image, the second face-swapped image, and the reference human body model.

[0099] In a possible implementation manner, the image processing unit 501 is specifically configured to, during the face swapping operation:

[0100] Locate the rendered image, and identify the face position in the rendered image; segment the rendered image according to the face position to obtain a first face image; replace the face features in the first face image according to the user's original image to obtain a second face image, and fuse the second face image with the rendered image to obtain a face-swapped image, where when the rendered image is the first rendered image, the face-swapped image is the first face-swapped image, and when the rendered image is the second rendered image, the face-swapped image is the second face-swapped image.

[0101] In a possible implementation manner, the model generation unit 502 is specifically configured to:

[0102] Obtain a three-dimensional face model according to the first face-swapped image; perform reference human body face mesh deformation based on the three-dimensional face model and the reference human body model to obtain a target human body mesh; perform face texture synthesis based on the second face-swapped image and the target human body mesh to obtain a target human body model.

[0103] In a possible implementation manner, the model generation unit 502 is further specifically configured to:

[0104] Match the first face-swapped image with basis vectors in multiple directions in a three-dimensional face model, and obtain coefficients corresponding to the basis vectors in multiple directions for the first face-swapped image, where the basis vectors are basic elements that make up a vector space; obtain a three-dimensional face model corresponding to the user's original image based on the basis vectors in multiple directions and the coefficients corresponding to the basis vectors in multiple directions.

[0105] In a possible implementation manner, the model generation unit 502 is further specifically configured to:

[0106] Obtain a planar projection image of the target human body mesh, and determine a first correspondence between multiple patches of the planar projection image and multiple patches of the second face-swapped image; obtain the texture map corresponding to the target human body mesh; determine the positions of multiple patches of the planar projection image corresponding to the UV texture mapping coordinates according to the texture map, and determine the positions of multiple patches of the second face-swapped image corresponding to the UV texture mapping coordinates according to the first correspondence; determine the pixel values of the positions of the VU texture mapping coordinates corresponding to multiple patches of the second face-swapped image according to the pixel values of multiple patches of the second face-swapped image, and obtain a face texture map; replace the texture map corresponding to the target human body mesh with the face texture map to obtain a target human body model.

[0107] In a possible implementation manner, the first rendered image and the second rendered image are a single image based on a frontal face or multiple images based on multi-angle profile faces.

[0108] Optionally, the above image processing unit 501 may be a Central Processing Unit (CPU), or may be a graphics processing unit (GPU), or may be a combination of a CPU and a GPU, and can be used for image processing. The above model generation unit 502 may also be a CPU, or a GPU, or a combination of a CPU and a GPU. The present application does not make specific limitations.

[0109] Optionally, the three-dimensional model generation device 500 may further include a transceiver unit 503, and the transceiver unit 503 may be an interface circuit or a transceiver. It is used to obtain data or receive instructions from other electronic devices.

[0110] Optionally, the three-dimensional model generation device 500 may further include a storage module (not shown in the figure), and the storage module may be used to store data and / or signaling. The storage module may be coupled to the image processing unit 501 and the model generation unit 502, or may also be coupled to the transceiver unit 503. For example, the image processing unit 501 may be used to read data and / or signaling in the storage module, so that the face-swapping operation process in the foregoing method embodiments is executed.

[0111] As Figure 6 shown, Figure 6 shows a schematic hardware structure diagram of an electronic device in an embodiment of the present application. The structure of the three-dimensional model generation device 500 may refer to the structure Figure 6 shown. The electronic device 800 includes: a memory 801, a processor 802, a communication interface 803, and a bus 804. Among them, the memory 801, the processor 802, and the communication interface 803 are communicatively connected to each other through the bus 804.

[0112] The memory 801 can be a Read Only Memory (ROM), a static storage device, a dynamic storage device, or a Random Access Memory (RAM). The memory 801 can store a program. When the program stored in the memory 801 is executed by the processor 802, the processor 802 and the communication interface 803 are used to execute each step of the XX network training method of the embodiments of the present application.

[0113] The processor 802 can be a general-purpose CPU, a microprocessor, an Application Specific Integrated Circuit (ASIC), a GPU, or one or more integrated circuits, and is used to execute relevant programs to implement the functions required by the image processing unit 501 or the model generation unit 502 in the 3D model generation device 500 of the embodiments of the present application, or to execute the 3D model generation method of the method embodiments of the present application.

[0114] The processor 802 can also be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the 3D model generation method of the present application can be completed by the integrated logic circuit in the hardware of the processor 802 or by instructions in software form. The above-mentioned processor 802 can also be a general-purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as being executed by the hardware decoding processor, or executed by a combination of the hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory 801, and the processor 802 reads the information in the memory 801 and combines its hardware to complete the functions required by the units included in the 3D model generation device 500 of the embodiments of the present application, or to execute the 3D model generation of the method embodiments of the present application.

[0115] The communication interface 803 uses a transceiver device such as, but not limited to, a transceiver to implement communication between the device 800 and other devices or communication networks. For example, the first rendered image and the second rendered image can be obtained through the communication interface 803.

[0116] The bus 804 may include a path for transmitting information between various components of the device 800 (e.g., the memory 801, the processor 802, the communication interface 803).

[0117] It should be understood that the transceiver unit 503 in the three-dimensional model generation device 500 is equivalent to the communication interface 803 in the electronic device 800, and the image processing unit 501 or the model generation unit 502 may be equivalent to the processor 802.

[0118] It should be noted that although Figure 6 the illustrated electronic device 800 only shows the memory, the processor, and the communication interface, in the specific implementation process, those skilled in the art should understand that the electronic device 800 also includes other devices necessary for normal operation. At the same time, according to specific needs, those skilled in the art should understand that the electronic device 800 may also include hardware devices for implementing other additional functions. In addition, those skilled in the art should understand that the electronic device 800 may also only include the devices necessary for implementing the embodiments of the present application, and do not necessarily include Figure 6 all the devices shown therein.

[0119] It should be understood that in various embodiments of the present application, the magnitudes of the sequence numbers of the above processes do not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0120] Those of ordinary skill in the art can realize that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present application.

[0121] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the systems, devices, and modules described above can refer to the corresponding processes in the foregoing method embodiments, and will not be repeated herein.

[0122] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of devices or modules can be in electrical, mechanical, or other forms.

[0123] The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical modules, that is, they can be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0124] In addition, in each embodiment of the present application, the functional modules can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0125] If the function is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in each embodiment of the present application. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs and other various media that can store program codes.

[0126] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present application can easily think of changes or substitutions, which should all be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A three-dimensional model generation method, characterized in that The method includes: Performing a face swapping operation on a first rendered image based on a user's original image to obtain a first face-swapped image, and performing a face swapping operation on a second rendered image to obtain a second face-swapped image. Among them, the first rendered image is generated by rendering a reference human body model under specific lighting, and the second rendered image is generated by rendering the reference human body model without lighting. The specific lighting is used to make the reference human body model produce reflections, forming color image information on the first rendered image; Generating a target human body model based on the first face-swapped image, the second face-swapped image, and the reference human body model, including: obtaining a three-dimensional face model according to the first face-swapped image; performing deformation on the reference human face grid based on the three-dimensional face model and the reference human body model to obtain a target human body grid; performing face texture synthesis based on the second face-swapped image and the target human body grid to obtain the target human body model.

2. The method according to claim 1, characterized in that The first rendered image includes shadow features.

3. The method according to claim 1, wherein The face swapping operation includes: Locating the rendered image and identifying the position of the human face in the rendered image; Segmenting the rendered image according to the position of the human face to obtain a first human face image; Replacing the human face features in the first human face image according to the user's original image to obtain a second human face image, and fusing the second human face image with the rendered image to obtain a face-swapped image. When the rendered image is the first rendered image, the face-swapped image is the first face-swapped image; when the rendered image is the second rendered image, the face-swapped image is the second face-swapped image.

4. The method according to claim 1, wherein The obtaining of the three-dimensional face model corresponding to the user's original image according to the first face-swapped image includes: Matching the first face-swapped image with basis vectors in multiple directions in a three-dimensional face model to obtain coefficients corresponding to the basis vectors in multiple directions of the first face-swapped image. The basis vectors are basic elements that make up a vector space; Obtaining the three-dimensional face model corresponding to the user's original image based on the basis vectors in multiple directions and the coefficients corresponding to the basis vectors in multiple directions.

5. The method according to claim 4, wherein The performing of face texture synthesis based on the second face-swapped image and the target human body grid to obtain the target human body model includes: Obtaining a planar projection image of the target human body grid and determining a first correspondence relationship between multiple patches of the planar projection image and multiple patches of the second face-swapped image; Obtaining a texture map corresponding to the target human body grid; Determining the positions of multiple patches of the planar projection image corresponding to UV texture mapping coordinates according to the texture map, and determining the positions of multiple patches of the second face-swapped image corresponding to the UV texture mapping coordinates according to the first correspondence relationship; Determining the pixel values of the positions of the UV texture mapping coordinates corresponding to multiple patches of the second face-swapped image according to the pixel values of multiple patches of the second face-swapped image to obtain a human face texture map; Replacing the texture map corresponding to the target human body grid with the human face texture map to obtain the target human body model.

6. The method according to any one of claims 1-5, characterized in that, The first rendered image and the second rendered image are single images based on a frontal face or multiple images based on multi-angle side faces.

7. A three-dimensional model generation device, characterized in that, The device includes an image processing unit and a model generation unit, where the image processing unit is configured to perform face swapping operations on a first rendered image to obtain a first face-swapped image and on a second rendered image to obtain a second face-swapped image according to a user's original image. The first rendered image is generated by rendering a reference human body model under specific lighting, and the second rendered image is generated by rendering the reference human body model without lighting. The specific lighting is used to make the reference human body model produce reflections, forming the color image information on the first rendered image; the model generation unit is configured to generate a target human body model according to the first face-swapped image, the second face-swapped image, and the reference human body model. Specifically, it is configured to: obtain a three-dimensional face model according to the first face-swapped image; perform reference human body face mesh deformation based on the three-dimensional face model and the reference human body model to obtain a target human body mesh; perform face texture synthesis based on the second face-swapped image and the target human body mesh to obtain the target human body model.

8. The device according to claim 7, characterized in that, The first rendered image includes shadow features.

9. The device according to claim 7, characterized in that, During the face swapping operation, the image processing unit is specifically configured to: locate the rendered image and identify the face position in the rendered image; segment the rendered image according to the face position to obtain a first face image; replace the face features in the first face image according to the user's original image to obtain a second face image, and fuse the second face image with the rendered image to obtain a face-swapped image. When the rendered image is the first rendered image, the face-swapped image is the first face-swapped image; when the rendered image is the second rendered image, the face-swapped image is the second face-swapped image.

10. The device according to claim 7, characterized in that, The model generation unit is also specifically configured to: match the first face-swapped image with basis vectors in multiple directions in a three-dimensional face model to obtain coefficients corresponding to the basis vectors in multiple directions for the first face-swapped image. The basis vectors are basic elements that make up a vector space; obtain the three-dimensional face model corresponding to the user's original image according to the basis vectors in multiple directions and the coefficients corresponding to the basis vectors in multiple directions.

11. The device according to claim 10, characterized in that, The model generation unit is also specifically configured to: obtain a planar projection image of the target human body mesh and determine a first correspondence relationship between multiple patches of the planar projection image and multiple patches of the second face-swapped image; obtain the texture map corresponding to the target human body mesh; determine the positions of multiple patches of the planar projection image corresponding to UV texture mapping coordinates according to the texture map, and determine the positions of multiple patches of the second face-swapped image corresponding to the UV texture mapping coordinates according to the first correspondence relationship; determine the pixel values of the positions of the UV texture mapping coordinates corresponding to multiple patches of the second face-swapped image according to the pixel values of multiple patches of the second face-swapped image to obtain a face texture map; replace the texture map corresponding to the target human body mesh with the face texture map to obtain the target human body model.

12. The device according to any one of claims 7-11, characterized in that, The first rendered image and the second rendered image are single images based on a frontal face or multiple images based on side faces from multiple angles.

13. An electronic device, characterized in that, The electronic device includes a processor and a memory, and the processor is configured to run code instructions stored in the memory to execute the method according to any one of claims 1 to 6.

14. A computer program product comprising instructions that, when run on a computer, cause the computer to execute the method according to any one of claims 1 to 6.

15. A computer-readable storage medium, characterized in that, Computer instructions are stored in the computer-readable storage medium, and when the computer instructions are run on a communication device, the communication device is caused to execute the method according to any one of claims 1 to 6.

Images