Digital human synthesis method, device, electronic equipment and storage medium

Abstract

The present disclosure provides a digital human synthesis method, device, electronic device and storage medium, which relate to the field of artificial intelligence technology. The method includes: obtaining a target image, a driving video and a target audio; extracting three-dimensional face data and three-dimensional implicit key points in the target image, as well as the audio features of the target audio; predicting the three-dimensional face data mapped by the audio features based on the three-dimensional face data and the audio features; obtaining the head posture data of the character in the driving video, and predicting the motion optical flow of the target image based on the head posture data, the three-dimensional implicit key points and the three-dimensional face data mapped by the audio features; synthesizing the video frame image of the digital human based on the motion optical flow to obtain a video of the digital human broadcasting the target audio. The present disclosure ensures that the detailed description of the digital human's face and the head swing are continuous and natural, and improves the effect and stability of the digital human broadcasting video.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a digital human synthesis method, device, electronic equipment and storage medium. Background Art

[0002] Digital humans refer to virtual characters generated through computer graphics, artificial intelligence, natural language processing, deep learning and other technologies, which can imitate the appearance, voice, expression and body movements of real people and interact with users. Enterprises are also gradually applying virtual digital humans to brand promotion and customer service, using their 24-hour uninterrupted interaction and personalized image to create a more efficient user experience.

[0003] At present, relevant technologies can use artificial intelligence algorithms to generate highly realistic digital humans from an image, mainly based on diffusion models or convolution models. However, the above diffusion models have high inference costs, low controllability in the process of generating continuous video frames, difficulty in maintaining the consistency of the previous and next frames of the subject and background of the photo, and poor stability of the generated video. The above convolution models have problems such as the lack of realism in the synthesized digital humans and unnatural head movements in the synthesized videos. Summary of the invention

[0004] In view of this, the embodiments of the present disclosure propose a digital human synthesis method, device, electronic device and storage medium to solve the problems existing in the related art such as the high inference cost of the diffusion model, the low controllability in the process of generating continuous video frames, the difficulty in maintaining the consistency of the previous and next frames of the photo subject and background, and the poor stability of the generated video; and the lack of realism of the digital human in the convolution model, and the unnatural head movement of the characters in the synthesized video.

[0005] According to a first aspect of the present disclosure, a digital human synthesis method is provided, comprising: acquiring a target image, a driving video and a target audio; extracting three-dimensional face data and three-dimensional implicit key points in the target image, and audio features of the target audio; predicting the three-dimensional face data of the audio feature map based on the three-dimensional face data and the audio features; acquiring head posture data of a person in the driving video, and predicting the motion optical flow of the target image based on the head posture data, the three-dimensional implicit key points and the three-dimensional face data of the audio feature map; synthesizing the video frame image of the digital human based on the motion optical flow to obtain a video of the digital human broadcasting the target audio.

[0006] According to a second aspect of the present disclosure, a digital human interaction device is provided, comprising: an acquisition module for acquiring a target image, a driving video and a target audio; an extraction module for extracting three-dimensional face data and three-dimensional implicit key points in the target image, and audio features of the target audio; a processing module for predicting the three-dimensional face data of the audio feature map based on the three-dimensional face data and the audio features; the processing module is also used to acquire head posture data of a person in the driving video, and predict the motion optical flow of the target image based on the head posture data, the three-dimensional implicit key points and the three-dimensional face data of the audio feature map; a synthesis module is used to synthesize the video frame image of the digital human based on the motion optical flow to obtain a video of the digital human broadcasting the target audio.

[0007] According to a third aspect of the present disclosure, there is provided an electronic device, comprising: a processor; and a memory storing a program, wherein the program comprises instructions, which, when executed by the processor, cause the processor to perform the above-mentioned digital human synthesis method.

[0008] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute the above-mentioned digital human synthesis method.

[0009] The exemplary embodiments of the present disclosure may have some or all of the following beneficial effects:

[0010] In the digital human synthesis method provided in the exemplary implementation mode of the present disclosure, a target image, a driving video and a target audio are obtained; three-dimensional face data and three-dimensional implicit key points in the target image, as well as audio features of the target audio are extracted; based on the three-dimensional face data and the audio features, three-dimensional face data of the audio feature mapping is predicted; head posture data of a person in the driving video is obtained, and the motion optical flow of the target image is predicted based on the head posture data, the three-dimensional implicit key points and the three-dimensional face data of the audio feature mapping; and the video frame image of the digital human is synthesized based on the motion optical flow to obtain a video of the digital human broadcasting the target audio. The disclosed embodiment extracts three-dimensional face data from a target image, and constructs a mapping relationship between a target audio and a digital human face based on the extracted three-dimensional face data, and utilizes the dense feature representation of the three-dimensional face data to ensure the detailed description of the face of the digital human during the playback of the target audio. In addition, the disclosed embodiment predicts the motion optical flow of the target image by driving the mapping relationship between the head posture data of the character in the video, the three-dimensional implicit key points of the target image, and the audio features and the face, and uses the motion optical flow to guide the synthesis of the digital human's spoken video, decoupling the connection between the audio data and the head swinging movement of the digital human generated based on the target image, thereby ensuring that the synthesized digital human's head swing is continuous and natural, and improving the stability of the synthesized spoken video. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Further details, features and advantages of the present disclosure are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which:

[0012] Figure 1 A flowchart of a digital human synthesis method according to an exemplary embodiment of the present disclosure is shown;

[0013] Figure 2 A schematic block diagram of a digital human interaction device according to an exemplary embodiment of the present disclosure is shown;

[0014] Figure 3 A structural block diagram of an exemplary electronic device that can be used to implement the embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0015] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0016] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0017] The term "including" and its variations used in this document are open inclusions, that is, "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments". Relevant definitions of other terms will be given in the description below. It should be noted that the concepts of "first", "second", etc. mentioned in this disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0018] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0019] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0020] Digital human refers to a virtual character image generated through computer graphics, artificial intelligence, natural language processing, deep learning and other technologies, which can imitate the appearance, voice, expression and body movements of real people and interact with users. Related technologies can use artificial intelligence algorithms to automatically generate highly realistic digital humans from an image.

[0021] Specifically, the related technology can realize the process of generating digital human through images through diffusion model or convolution model. However, the above method has the following problems:

[0022] (1) The diffusion model encodes the features of the image and integrates multimodal data (voice data, text data) for denoising, which can generate high-resolution, detail-rich images. In the process of gradual denoising, the model can capture complex textures and subtle features. At the same time, by adjusting the input parameters, it can realize digital humans with different styles, expressions, and postures. However, the inference cost of the diffusion model is high, requiring a lot of computing resources and time. At the same time, the images generated by the diffusion model are highly diverse. It is difficult to maintain the consistency of the previous and next frames of the image subject and background during the generation of continuous video frames. In addition, as the generation time increases, the content diverges more violently, and it is usually impossible to generate a stable video of more than 10 seconds.

[0023] (2) The convolutional network model usually uses facial key points as intermediate variables of the model, maps audio data to 2D facial key points, and then uses the 2D key points to guide the generation of continuous video frames. This method is easy to implement and fast. However, in this method, sparse discrete key points cannot describe the detailed characteristics of local nonlinear changes in the face, and the plane points lack depth information, which easily leads to the lack of realism in the synthesized digital human. In addition, this method couples the facial expression movements guided by audio with the shaking of the human head, which easily leads to unnatural movements such as twitching of the head of the synthesized video character.

[0024] In order to solve the above problems, the embodiments of the present disclosure provide a digital human synthesis method, device, electronic device and storage medium. The following describes the scheme of the present disclosure with reference to the accompanying drawings for illustration.

[0025] The present disclosure first provides a method for synthesizing a digital human. Figure 1 As shown, the digital human synthesis method specifically includes the following steps:

[0026] S110: Acquire target image, drive video and target audio;

[0027] S120: extracting three-dimensional face data and three-dimensional implicit key points in the target image, and audio features of the target audio;

[0028] S130: predicting three-dimensional face data of audio feature mapping based on the three-dimensional face data and the audio feature;

[0029] S140: Acquire head posture data of a person in a driving video, and predict motion optical flow of a target image based on the head posture data, three-dimensional implicit key points, and three-dimensional face data mapped by audio features;

[0030] S150: Synthesizing the video frame images of the digital human based on the motion optical flow to obtain a video of the digital human broadcasting the target audio.

[0031] The disclosed embodiment extracts three-dimensional face data from a target image, and constructs a mapping relationship between a target audio and a digital human face based on the extracted three-dimensional face data, and utilizes the dense feature representation of the three-dimensional face data to ensure the detailed description of the face of the digital human during the playback of the target audio. In addition, the disclosed embodiment predicts the motion optical flow of the target image by driving the mapping relationship between the head posture data of the character in the video, the three-dimensional implicit key points of the target image, and the audio features and the face, and uses the motion optical flow to guide the synthesis of the digital human's spoken video, decoupling the connection between the audio data and the head swinging movement of the digital human generated based on the target image, thereby ensuring that the synthesized digital human's head swing is continuous and natural, and improving the stability of the synthesized spoken video.

[0032] Next, in another embodiment, the above steps are described in more detail.

[0033] In step S110 , a target image, a driving video, and a target audio are acquired.

[0034] In the embodiment of the present application, the target image is an image used to generate a digital human character image. Exemplarily, the target image can be determined by a photo including a face input by a user.

[0035] In an embodiment of the present application, the above-mentioned target image can be determined by a photo including a face input by a user, which means that the photo including a face input by the user can be directly used as the target image, or the photo including a face input by the user can be used as an original image, and the original image can be pre-processed, and the processed original image can be used as the target image.

[0036] Exemplarily, the above process of obtaining the target image by preprocessing the original image can be implemented as follows: obtain the original image; detect the face in the original image, and perform key point detection on the detected face; align the face detected in the original image with the standard face based on the coordinates of the face key points to obtain the target image. Among them, the above process of aligning the face detected in the original image with the standard face based on the coordinates of the face key points to obtain the target image can be implemented as follows: crop the original image based on the detected face key points and preset cropping data to obtain a face cropped image; calculate the coordinates of the face key points in the face cropped image, and determine the affine transformation matrix based on the coordinates of the face key points; align the face cropped image with the standard face based on the affine transformation matrix to obtain the target image.

[0037] Specifically, the preset cropping data can be: cropping the original image with a radius of 2.3 times the distance between the mouth and eyes and the midpoint of the face as the center point, then the process can be implemented as follows: performing face detection on the original image, illustratively, the face in the original image can be detected based on a one-stage target detector (retinaNet); detecting facial key points, and cropping the original image with a radius of 2.3 times the distance between the mouth and eyes and the midpoint of the face as the center point, wherein the facial key points include the eye corner key points, the mouth corner key points to calculate the distance between the mouth and eyes and the midpoint of the face; calculating the coordinates of the facial key points on the cropped image, and determining the affine transformation matrix by pairing with the same-name points of the standard face, so as to align the cropped image with the standard face based on the affine transformation matrix to obtain the target image.

[0038] It should be noted that the above scenario is only an exemplary description, and the protection scope of the embodiments of the present application is not limited thereto. For example, the above original image or target image may also be any other image containing elements that can generate a virtual digital human.

[0039] In the embodiments of the present application, the digital human refers to a virtual character image generated by computer graphics, artificial intelligence, natural language processing, deep learning and other technologies, which can imitate the appearance, voice, expression and body movements of real humans and interact with users. The digital human can be a simple static virtual image or a digital image with real-time interaction and high intelligence.

[0040] In the embodiment of the present application, the driving video may be any video containing a human image, and is used to drive the head movement of the digital human generated based on the target image, so that the digital human imitates the head rotation trajectory of the human image in the video. For example, the driving video may be any speech video.

[0041] Since the voice of a person is strongly correlated with the person's face, especially the shape of the mouth and the movement of the mandible, but not with the movement of the head posture, the same person can say the same thing, and the head movement can be in any trajectory, and there is no mapping relationship in the data distribution. Therefore, the embodiment of the present application drives the head movement of the digital human through the above-mentioned driving video, decouples the connection between the audio data and the head swinging movement of the person, avoids coupling the facial expression movement guided by the audio with the head swinging of the person, and thus solves the problem of unnatural movements such as head twitching caused by this.

[0042] In the embodiment of the present application, the target audio is the audio data broadcasted by the digital human in the synthesized spoken video, which is used to guide the generation of the digital human's facial expressions and movements. Exemplarily, the target audio can be the audio input by the user.

[0043] In step S120, three-dimensional face data and three-dimensional implicit key points in the target image, as well as audio features of the target audio are extracted.

[0044] In the embodiment of the present application, the above three-dimensional face data is the parameters of a three-dimensional face statistical model (3D MorphableModel, 3DMM). The above 3DMM three-dimensional face statistical model is used to solve the problem of restoring a three-dimensional shape from a two-dimensional face image, and the above 3DMM parameters provide a set of high-dimensional vectors that describe the shape, texture, posture, and expression of the face.

[0045] In the embodiment of the present application, the above 3DMM parameters may include shape parameters, texture parameters and expression parameters. Among them, the shape parameters are used to describe the geometric shape of the face, including the vertex coordinate information of the face, which defines the geometric structure of the face, such as the height of the nose, the position of the mouth, etc.; the texture parameters are used to describe the surface color and texture of the face, including the color information of each vertex (such as RGB value), which defines the surface color and texture details of the face; the expression parameters are used to describe the expression changes of the face, including the vertex displacement information related to the expression, which defines the changes of the face under different expressions, such as smiling, frowning, etc.

[0046] In an embodiment of the present application, the above-mentioned three-dimensional implicit key points (3D implicit key points) refer to key points defined in an implicit manner in three-dimensional space. They are representative points in three-dimensional point clouds and are widely used in object recognition, scene reconstruction, robot operation and other fields.

[0047] In an embodiment of the present application, exemplarily, the above-mentioned extraction of three-dimensional face data and three-dimensional implicit key points in the target image, and audio features of the target audio can be implemented as follows: cutting the audio data of the target audio, and extracting the audio features of each audio data block through an audio feature extractor; extracting 3DMM parameters of the target image through a convolutional neural network; constructing a 3D convolutional network, and extracting 3D implicit key points in the target image based on the 3D convolutional network.

[0048] Exemplarily, the above-mentioned audio feature extractor can use a sequence-to-sequence model (whisper) audio feature extractor. Among them, the above-mentioned Whisper is a transformer-based encoder-decoder model that can map the spectrogram feature sequence of audio data to the word sequence of text. The embodiment of the present application uses a pre-trained whisper speech recognition model to extract features from audio, and can obtain audio features with better expression and stronger generalization, while covering multi-language features such as Chinese and English.

[0049] In step S130, based on the three-dimensional face data and the audio features, the three-dimensional face data of the audio feature mapping is predicted.

[0050] In an embodiment of the present application, after the 3DMM parameters in the target image and the audio features in the target audio are extracted through the above step S120, the 3DMM parameters corresponding to the audio features of the target audio can be predicted based on the extracted 3DMM parameters and audio features.

[0051] Specifically, take the audio data of the above-mentioned target audio cut, and extract the audio features of each audio data block through an audio feature extractor; extract the 3DMM parameters of the target image through a convolutional neural network as an example, after extracting the audio features of each audio data block and the 3DMM parameters of the target image, further, the embodiment of the present application implements the above-mentioned prediction process through the following method: split the 3DMM parameters, stack the exp parameters expressing the facial features with the audio features, input the multilayer perceptron (MLP) for prediction, and obtain the 3DMM facial parameters mapped by the corresponding audio data block. Among them, the 3DMM facial parameters mapped by the above audio data blocks integrate the voice information of the target video and the facial features of the characters in the target image, and the exp parameters in the above 3DMM refer to expression parameters, which are used to represent the expression changes of the face.

[0052] The embodiment of the present application uses the 3DMM expression parameters to construct a mapping relationship between the target audio and the face in the target image through the above process, and uses the dense feature representation of 3DMM to ensure that the facial details required by the target audio are described during the speech of the generated digital person.

[0053] In step S140, the head posture data of the person in the driving video is obtained, and the motion optical flow of the target image is predicted based on the head posture data, three-dimensional implicit key points and three-dimensional face data mapped by audio features.

[0054] In the embodiment of the present application, the head posture data is used to describe the movement trajectory of the head of the character in the driving video. Exemplarily, the head posture data of the character in the driving video can be obtained as follows: head movement posture detection is performed on the character in the driving video; the posture difference of the head movement posture between each adjacent video frame in the driving video and the angle difference of the head movement posture between each video frame and the first frame of the video are calculated to obtain the head posture data.

[0055] In the embodiment of the present application, the above-mentioned motion optical flow refers to the movement of pixels in an image sequence, which is usually represented by a two-dimensional vector, representing the displacement in the horizontal direction and the vertical direction respectively. The basic task of optical flow prediction is to estimate the position of each pixel in the first image in the adjacent image from a pair of time-related image pairs, which can be used to predict the movement between two consecutive image frames in a video.

[0056] In the embodiment of the present application, exemplarily, the above-mentioned prediction of the motion optical flow of the target image based on the three-dimensional face data of the head posture data, the three-dimensional implicit key points and the audio feature mapping can be implemented as follows: determine the three-dimensional driving key points based on the three-dimensional face data of the audio feature mapping, the head posture data and the three-dimensional implicit key points; predict the motion optical flow of the target image based on the three-dimensional driving key points. Specifically, the process can be implemented as follows: estimate the 3D implicit key points of the target image; construct a mapping function, link the 3DMM expression parameters with the 3D implicit key points, estimate the implicit key point offset corresponding to each frame of audio, apply the rotation angle and offset distance of the driving video to the 3D key points of the target image as the driving points, that is, the above-mentioned three-dimensional driving key points; calculate the heat map between the 3D feature points of the target image and the above-mentioned 3D driving key points, and predict the motion optical flow of the high-dimensional features of the target image.

[0057] In step S150, the video frame images of the digital human are synthesized based on the motion optical flow to obtain a video of the digital human broadcasting the target audio.

[0058] In the embodiment of the present application, the video frame images are video frame images of a digital person broadcasting a video generated based on a target image and a target audio. Exemplarily, after the motion optical flow is obtained, the continuous video frame images are generated based on the motion optical flow guidance to synthesize the video of the digital person broadcasting the target audio.

[0059] Exemplarily, the above-mentioned video frame image of synthesizing a digital human based on motion optical flow can be implemented as follows: extracting the three-dimensional features of the target image, resampling the three-dimensional features of the target image based on the motion optical flow, and obtaining the video frame image. Specifically, the process can be implemented as follows: resampling the three-dimensional features of the target image according to the above-mentioned motion optical flow, sending the resampled features to the feature fusion layer, decoding the fused features, and outputting the video frame image.

[0060] In the embodiment of the present application, corresponding to the preprocessing operation on the original image in step S110, after obtaining the above-mentioned video frame image, a post-processing operation can also be performed on the video frame image. Exemplarily, the post-processing operation can be implemented as follows: the video frame image is subjected to the inverse affine transformation of the affine transformation matrix in the above-mentioned preprocessing process to obtain an intermediate processed image; the intermediate processed image is subjected to Poisson fusion operation based on the above-mentioned preset cropping data. Specifically, the above-mentioned process can be implemented as follows: the inverse matrix of the affine transformation matrix in the preprocessing process is calculated, and the inverse matrix of the affine transformation matrix is ​​applied to the output video frame image; the transformed image is backfilled back to the original image using Poisson fusion and the above-mentioned preset cropping parameters.

[0061] In an embodiment of the present application, the original image is aligned to the standard input scale and posture through the above-mentioned preprocessing operation, and after the digital human video frame image is synthesized through the post-processing operation, the synthesized image is transformed back to the scale and posture of the original image and seamlessly integrated into the original image, thereby ensuring the consistency of the scale and posture of the original image and the synthesized image, and improving the quality and stability of the synthesized video.

[0062] The disclosed embodiment extracts three-dimensional face data from a target image, and constructs a mapping relationship between a target audio and a digital human face based on the extracted three-dimensional face data, and utilizes the dense feature representation of the three-dimensional face data to ensure the detailed description of the face of the digital human during the playback of the target audio. In addition, the disclosed embodiment predicts the motion optical flow of the target image by driving the mapping relationship between the head posture data of the character in the video, the three-dimensional implicit key points of the target image, and the audio features and the face, and uses the motion optical flow to guide the synthesis of the digital human's spoken video, decoupling the connection between the audio data and the head swinging movement of the digital human generated based on the target image, thereby ensuring that the synthesized digital human's head swing is continuous and natural, and improving the stability of the synthesized spoken video.

[0063] The above mainly introduces the solution provided by the embodiment of the present invention from the perspective of the method. In order to achieve the above functions, it includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of each example described in the embodiments disclosed herein, the present invention can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0064] The embodiment of the present invention can divide the digital human interaction device into functional modules according to the above method example. For example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing module. The above integrated module can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the embodiment of the present invention is schematic and is only a logical function division. There may be other division methods in actual implementation.

[0065] Accordingly, the embodiment of the present disclosure also provides a digital human interaction device. Figure 2 As shown, the digital human interaction device 200 may include an acquisition module 210, an extraction module 220, a processing module 230 and a synthesis module 240, wherein:

[0066] The acquisition module 210 can be used to acquire the target image, drive video and target audio;

[0067] An extraction module 220 may be used to extract 3D face data and 3D implicit key points in a target image, as well as audio features of a target audio;

[0068] The processing module 230 may be used to predict the three-dimensional face data of the audio feature mapping based on the three-dimensional face data and the audio feature;

[0069] The processing module 230 is further used to obtain head posture data of the character in the driving video, and predict the motion optical flow of the target image based on the head posture data, the three-dimensional implicit key points and the three-dimensional face data mapped by the audio feature;

[0070] The synthesis module 240 is used to synthesize the video frame images of the digital human based on the motion optical flow to obtain the video of the digital human broadcasting the target audio.

[0071] In an embodiment of the present application, the above-mentioned acquisition module is specifically used to: acquire the original image; detect the face in the original image, and perform key point detection on the detected face; align the face detected in the original image with the standard face based on the coordinates of the face key points to obtain the target image.

[0072] In an embodiment of the present application, the above-mentioned acquisition module is specifically used to: crop the original image based on the detected facial key points and preset cropping data to obtain a cropped face image; calculate the coordinates of the facial key points in the cropped face image, and determine the affine transformation matrix based on the coordinates of the facial key points; align the cropped face image with the standard face based on the affine transformation matrix to obtain a target image.

[0073] In an embodiment of the present application, after synthesizing the video frame image of the digital human based on motion optical flow, the above-mentioned processing module is also used to: perform an inverse affine transformation of the video frame image of the affine transformation matrix to obtain an intermediate processed image; and perform a Poisson fusion operation on the intermediate processed image based on preset cropping data.

[0074] In an embodiment of the present application, the above-mentioned processing module is specifically used to: perform head movement posture detection on the characters in the driving video; calculate the posture difference of the head movement posture between adjacent video frames in the driving video, and the angle difference of the head movement posture between each video frame and the first frame of the video, to obtain head posture data.

[0075] In an embodiment of the present application, the above-mentioned processing module is specifically used to: determine the three-dimensional driving key points based on the three-dimensional face data mapped by audio features, head posture data and three-dimensional implicit key points; and predict the motion optical flow of the target image based on the three-dimensional driving key points.

[0076] In an embodiment of the present application, the above-mentioned synthesis module is specifically used to: extract the three-dimensional features of the target image; resample the three-dimensional features of the target image based on the motion optical flow to obtain a video frame image.

[0077] In addition, the specific implementation details of the above-mentioned digital human interactive device have been described in detail in the corresponding position of the digital human synthesis method, so they will not be repeated here.

[0078] Figure 3 Schematic diagram of the structure of an electronic device in the embodiment of the present disclosure. Figure 3 , which shows a structural schematic diagram of an electronic device 300 suitable for implementing the embodiments of the present disclosure. Figure 3 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0079] like Figure 3As shown, the electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 to a random access memory (RAM) 303 to implement the digital human synthesis method of the embodiment described in the present disclosure. In RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, ROM 302, and RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0080] Typically, the following devices may be connected to the I / O interface 305: input devices 306 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 308 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 309. The communication devices 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0081] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains a program code for executing the method shown in the flowchart, thereby implementing the digital human synthesis method as described above. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 309, or installed from the storage device 308, or installed from the ROM 302. When the computer program is executed by the processing device 301, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.

[0082] It should be noted that the computer-readable medium disclosed above may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0083] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (Hyper Text Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0084] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0085] The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device:

[0086] Acquire target images, drive videos and target audio;

[0087] Extracting 3D face data and 3D implicit key points in the target image, as well as audio features of the target audio;

[0088] Based on the three-dimensional face data and audio features, predict the three-dimensional face data of the audio feature mapping;

[0089] Obtain the head posture data of the person in the driving video, and predict the motion optical flow of the target image based on the head posture data, three-dimensional implicit key points and three-dimensional face data mapped by audio features;

[0090] The video frame images of the digital human are synthesized based on the motion optical flow to obtain a video of the digital human broadcasting the target audio. Optionally, when the above one or more programs are executed by the electronic device, the electronic device may also execute other steps described in the above embodiments.

[0091] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including, but not limited to, object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0092] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0093] The units involved in the embodiments described in the present disclosure may be implemented by software or hardware, wherein the name of a unit does not, in some cases, limit the unit itself.

[0094] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0095] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0096] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other to form a technical solution.

[0097] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0098] Although the subject matter has been described in language specific to structural features and / or methodological logical actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims.

Claims

1. A digital human synthesis method, characterized in that: include: Acquire target images, drive videos and target audio; Extracting three-dimensional face data and three-dimensional implicit key points in the target image, and audio features of the target audio; Based on the three-dimensional face data and the audio feature, predicting the three-dimensional face data of the audio feature mapping; Acquire head posture data of a person in the driving video, and predict the motion optical flow of the target image based on the head posture data, the three-dimensional implicit key points and the three-dimensional face data mapped by the audio feature; The video frame image of the digital human is synthesized based on the motion optical flow to obtain a video of the digital human broadcasting the target audio.

2. The digital human synthesis method according to claim 1, characterized in that: The acquiring of the target image comprises: Get the original image; Detecting a human face in the original image, and performing key point detection on the detected human face; The face detected in the original image is aligned with a standard face based on the coordinates of the facial key points to obtain the target image.

3. The digital human synthesis method according to claim 2, characterized in that: The step of aligning the face detected in the original image with a standard face based on the coordinates of the key points of the face to obtain the target image includes: Performing image cropping on the original image based on the detected facial key points and preset cropping data to obtain a facial cropped image; Calculating the coordinates of the key points of the face in the cropped face image, and determining an affine transformation matrix based on the coordinates of the key points of the face; The cropped face image is aligned with the standard face based on the affine transformation matrix to obtain the target image.

4. The digital human synthesis method according to claim 3, characterized in that: After synthesizing the video frame image of the digital human based on the motion optical flow, the method further includes: Performing an inverse affine transformation of the affine transformation matrix on the video frame image to obtain an intermediate processed image; A Poisson fusion operation is performed on the intermediate processed image based on the preset cropping data.

5. The digital human synthesis method according to claim 1, characterized in that: The step of obtaining head posture data of a person in the driving video includes: Performing head movement and posture detection on the person in the driving video; The posture difference of the head movement posture between adjacent video frames in the driving video and the angle difference of the head movement posture between each video frame and the first video frame are calculated to obtain the head posture data.

6. The digital human synthesis method according to claim 5, characterized in that: The predicting of the motion optical flow of the target image based on the three-dimensional face data of the head posture data, the three-dimensional implicit key points and the audio feature map includes: Determine three-dimensional driving key points based on the three-dimensional face data mapped by the audio feature, the head posture data and the three-dimensional implicit key points; The motion optical flow of the target image is predicted based on the three-dimensional driving key points.

7. The digital human synthesis method according to claim 1, characterized in that: The step of synthesizing the video frame image of the digital human based on the motion optical flow comprises: Extracting three-dimensional features of the target image; A resampling operation is performed on the three-dimensional features of the target image based on the motion optical flow to obtain the video frame image.

8. A digital human synthesis device, characterized in that: include: An acquisition module is used to acquire target images, drive videos and target audio; An extraction module, used to extract three-dimensional face data and three-dimensional implicit key points in the target image, and audio features of the target audio; A processing module, configured to predict the three-dimensional face data mapped by the audio feature based on the three-dimensional face data and the audio feature; The processing module is further used to obtain head posture data of the character in the driving video, and predict the motion optical flow of the target image based on the head posture data, the three-dimensional implicit key points and the three-dimensional face data mapped by the audio feature; A synthesis module is used to synthesize the video frame image of the digital human based on the motion optical flow to obtain a video of the digital human broadcasting the target audio.

9. An electronic device, comprising: processor; and a memory storing a program, wherein the program comprises instructions for causing the processor to execute the method according to any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-7.

Images