WO2022148379A1

WO2022148379A1 - Image processing method and apparatus, electronic device, and readable storage medium

Info

Publication number: WO2022148379A1
Application number: PCT/CN2022/070336
Authority: WO
Inventors: 李益永; 黄秋实; 孙准; 井雪; 项伟
Original assignee: 百果园技术(新加坡)有限公司
Priority date: 2021-01-05
Filing date: 2022-01-05
Publication date: 2022-07-14
Also published as: CN112734632B; CN112734632A

Abstract

The present disclosure provides an image processing method and apparatus. The image processing method comprises: acquiring an image to be migrated and a reference image, the image to be migrated comprising a target object of which the posture is to be converted; the reference image comprising a reference object presenting a reference posture; acquiring a first key feature of the target object and a second key feature of the reference object; determining a posture migration matrix according to the first key feature and the second key feature; acquiring an initial image; and determining a target synthetic image according to the posture migration matrix, the image to be migrated, and the initial image. In embodiments of the present disclosure, a large number of training sample training models do not need to be acquired to obtain the target synthetic image, so that the complexity of image migration is reduced; moreover, the initial image is acquired, and the whole image to be migrated is migrated according to the posture migration matrix, the image to be migrated, and the initial image, so that all the details of the image to be migrated can be ensured to be displayed in the target synthetic image, and details are prevented from being missed.

Description

Image processing method, apparatus, electronic device and readable storage medium

CROSS-REFERENCE TO RELATED APPLICATIONS

This disclosure claims the priority of a Chinese patent application with application number 202110009523.2 and titled "Image Processing Method, Apparatus, Electronic Device, and Readable Storage Medium" filed with the China Patent Office on January 5, 2021, the entire contents of which are hereby incorporated by reference Incorporated in this disclosure.

technical field

The present disclosure belongs to the technical field of image processing, and in particular, relates to an image processing method, apparatus, electronic device and readable storage medium.

Background technique

Pose transfer means that after an image A is processed, the person P in the image A has the pose of the person H in the other image B, and a composite image C is obtained.

At present, in order to achieve pose transfer, multiple images A, multiple images B and multiple images C are used as training samples to train an image transfer model, and then the new image A and image B are processed according to the image transfer model to obtain New composite image C.

In the above pose transfer method, when training the image transfer model, a large number of training samples need to be prepared, and the training method is cumbersome. Moreover, when the above-mentioned image migration model is used for image migration, when the clothing and body shapes of the characters in the two images are quite different, the characters in the composite image C cannot keep the details of the character P in the original image A. The shapes of the characters in different perspectives and postures are quite different. In addition, it may occur that only part of the human body of the characters has been migrated, and other parts of the human body need to be processed again to achieve the migration, resulting in a cumbersome migration process.

Overview

In view of this, the present disclosure provides an image processing method, which solves the problems of cumbersome migration process and incomplete migration to a certain extent.

A first aspect of the embodiments of the present disclosure provides an image processing method, the method includes:

Acquiring an image to be migrated and a reference image; the image to be migrated includes: a target object whose posture is to be converted; the reference image includes: a reference object showing a reference posture;

acquiring the first key feature of the target object and the second key feature of the reference object;

Determine a posture transition matrix according to the first key feature and the second key feature;

get the initial image;

A target composite image is determined according to the pose transfer matrix, the image to be transferred, and the initial image.

A second aspect of the embodiments of the present disclosure provides an image processing apparatus, the apparatus includes:

a first acquisition module, configured to acquire an image to be migrated and a reference image; the image to be migrated includes: a target object whose posture is to be converted; the reference image includes: a reference object that presents a reference posture;

a second acquisition module, configured to acquire the first key feature of the target object and the second key feature of the reference object;

a first determination module, configured to determine a posture transition matrix according to the first key feature and the second key feature;

The third acquisition module is used to acquire the initial image;

The second determining module is configured to determine a target composite image according to the posture transfer matrix, the image to be transferred and the initial image.

A third aspect of the embodiments of the present disclosure provides an electronic device, the electronic device includes a processor, a memory, and a program or instruction stored on the memory and executable on the processor, the program or instruction being executed by the The processor implements the steps of the method as described in the first aspect when executed.

A fourth aspect of the embodiments of the present disclosure provides a readable storage medium, where a program or an instruction is stored on the readable storage medium, and when the program or instruction is executed by a processor, the steps of the method according to the first aspect are implemented.

In the embodiment of the present disclosure, by acquiring an image to be migrated and a reference image; the image to be migrated includes: a target object whose posture is to be converted; the reference image includes: a reference object showing a reference posture; the target object is acquired The first key feature of the reference object and the second key feature of the reference object; determine a posture transfer matrix according to the first key feature and the second key feature; obtain an initial image; The migration image and the initial image determine a target composite image. In the embodiment of the present disclosure, it is not necessary to acquire a large number of training samples to train the model to obtain the target composite image, which reduces the tediousness of image migration, and to acquire the initial image, the pose transfer matrix, the image to be migrated and the initial image are used to analyze the entire image. The image to be migrated is migrated, so it can be ensured that the details of the image to be migrated are displayed in the target composite image, preventing the omission of details.

The above description is only an overview of the technical solutions of the present disclosure. In order to understand the technical means of the present disclosure more clearly, it can be implemented according to the contents of the description, and in order to make the above-mentioned and other purposes, features and advantages of the present disclosure more obvious and easy to understand , the following specific embodiments of the present disclosure are given.

Description of drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are for purposes of illustrating preferred embodiments only and are not to be considered limiting of the present disclosure. Also, the same components are denoted by the same reference numerals throughout the drawings. In the attached image:

FIG. 1 is a flowchart of steps of an image processing method provided by an embodiment of the present disclosure;

FIG. 2 is a schematic diagram of an image processing method provided by an embodiment of the present disclosure;

3 is a schematic diagram of another image processing method provided by an embodiment of the present disclosure;

4 is a block diagram of an image processing apparatus provided by an embodiment of the present disclosure;

FIG. 5 is a structural block diagram of an electronic device provided by an embodiment of the present disclosure.

specific embodiment

Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure will be more thoroughly understood, and will fully convey the scope of the present disclosure to those skilled in the art.

Referring to FIG. 1, a flowchart of steps of an image processing method provided by an embodiment of the present disclosure is shown, and the image processing method specifically includes the following steps:

Step 101: Obtain an image to be migrated and a reference image; the image to be migrated includes: a target object whose posture is to be converted; and the reference image includes: a reference object showing a reference posture.

The image to be migrated is an image of m ₁ *n ₁ *3, where m ₁ is the width of the image to be migrated, n ₁ is the height of the image to be migrated, and 3 means that the image to be migrated is an RGB image. The reference image is an image of m ₂ *n ₂ *3, where m ₂ is the width of the reference image, n ₂ is the height of the reference image, and 3 means that the reference image is an RGB image.

In the embodiment of the present disclosure, the target object and the reference object usually refer to the human body object in the image; with reference to FIG. 2 , wherein, the image A is the image to be migrated, and the image B is the reference image; the image A to be migrated includes the target object P, The reference image B includes: the reference object H.

In the embodiment of the present disclosure, the user can select the image to be migrated and the reference image from the image memory according to requirements, and can also capture and obtain at any time, which is not limited.

In addition, the user can select a video as a reference video, then use each frame of image in the reference video as a reference image, and then process the image to be migrated based on each frame of reference image.

Step 102: Acquire a first key feature of the target object and a second key feature of the reference object.

In the embodiment of the present disclosure, the image to be migrated is represented by a dimensional vector, then the dimensional vector of the image to be migrated is: x(i*m*n+j*m+k)=x(j, k, i); where , 3≥i≥1, n≥i≥1, m≥i≥1. The reference image is also represented by a dimensional vector, then the dimensional vector of the reference image is y(i*m*n+j*m+k)=y(j, k, i); among them, 3≥i≥1, n≥ i≥1, m≥i≥1.

Specifically, each pixel in the image to be migrated can be represented by a dimensional vector or coordinate, and each pixel in the reference image can also be represented by a dimensional vector or coordinate. For example: an image has 10 rows*10 columns of pixels, then the coordinates of the pixel p in the 5th row and 5th column are represented as (5,5); the pixel p is represented by a one-dimensional vector as p(45).

In the embodiment of the present disclosure, the first key feature refers to the coordinates of multiple feature points that can mark the pose of the target object; for example, the first key feature may be the coordinates of each joint of the target object; each joint includes: shoulder joint, elbow Joints, radiocarpal joints, carpal metacarpal joints, hip joints, knee joints, ankle joints, etc. In addition, the first key feature may also be the coordinates of the main parts of the human body, for example, the parts that characterize the posture of the head, including: eyes, nose tip, temples, and the tip of the chin; the parts that characterize the posture of the arms, including: shoulder joints, elbow joints and Carpal-metacarpal joint; parts that characterize hand posture, including: knuckles and fingertips of each finger; parts that characterize leg posture, including: hip joint, knee joint, ankle joint.

In the embodiment of the present disclosure, the first key feature is a preset key feature in the target object; the second key feature is in one-to-one correspondence with the first key feature.

Specifically, when the first key feature is obtained, the second key feature can be obtained according to the first key feature, wherein the second key feature corresponds to the first key feature, for example, the first key feature includes: the shoulder joint of the target object, The coordinates of the elbow joint, radiocarpal joint, carpal metacarpal joint, hip joint, knee joint, and ankle joint in the image to be migrated, then the second key feature includes: the shoulder joint, elbow joint, radiocarpal joint, and carpal metacarpal joint of the reference object , the coordinates of the hip, knee, and ankle joints in the reference image.

In this embodiment of the present disclosure, if the target image only includes facial images, that is, when only facial poses are migrated, the first key feature is set as the coordinates of each feature point of the human face, such as eyes, nose, eyebrows, ears, mouth, etc.

In the embodiment of the present disclosure, the user can select the part to be migrated as required. For example, when only the face is migrated, only the first key feature of the face is selected, and when only the body is migrated, only the first key feature of the body is selected. key features.

Step 103: Determine a posture transition matrix according to the first key feature and the second key feature.

In this embodiment of the present disclosure, the step 103 includes: determining the coordinate value of each of the first key features and the coordinate value of each of the second key features;

The attitude transfer matrix is determined according to the coordinate value of the first key feature and the coordinate value of the second key feature, and the attitude transfer matrix is used to convert the coordinate value of the first key feature into a The coordinate value of the second key feature corresponding to the first key feature.

In the embodiment of the present disclosure, the attitude transfer matrix refers to the attitude transfer matrix required for the coordinates of the first key feature to be transferred to the coordinates of the second key feature. For example, the first key feature includes: the coordinates of the temple (a, b), the shoulder joint (c, d), the coordinates of the temple (m, n) of the second key feature, and the shoulder joint (o, p); The coordinates of the temple of the first key feature are stored as (m, n), the coordinates of the shoulder joint of the first key feature are stored as (o, p), and the coordinates of the elbow joint of the first key feature are stored as (q, r ), and so on. Among them, according to

The attitude transition matrix W is obtained. When the first key feature includes multiple (greater than or equal to three) features, the attitude transition matrix W can also be obtained in this way.

In the embodiment of the present disclosure, the coordinates of each first key feature are Px, and the coordinates of each second key feature are Py, then W=W[Px,Py] is required for the transformation from the first key feature to the second key feature The pose transfer matrix of .

Wherein, after the attitude transfer matrix W is determined, each pixel point of the image to be transferred can be transferred using the attitude transfer matrix W.

Step 104, acquiring an initial image.

In the embodiment of the present disclosure, the initial image is an initial image that needs to be input in order to complete the subsequent steps of obtaining the target composite image by using a preset method in the embodiment of the present disclosure.

In this embodiment of the present disclosure, step 104 includes: inputting the pose transfer matrix and the image to be transferred into an initial network model to obtain an initial image.

In this embodiment of the present disclosure, the initial network model may be a model trained according to data samples, wherein the data samples include: a plurality of pose transfer matrix samples for converting image samples to be migrated into reference image samples, and a plurality of image samples to be migrated and multiple targets to synthesize image samples; use these data samples to train to obtain the initial network model; then input the attitude transfer matrix and the image to be migrated into the initial network model obtained by training, that is, to obtain the initial image, using this method to obtain the initial image, is The target object in the image to be migrated adopts the initial composite image in the pose of the reference object, but the details of the initial composite image are still missing and cannot fully present all the features of the image to be migrated. The details of the image to be migrated can be completed after proceeding with the next steps.

In addition, the working principle of the initial network model can also be Z ₀ =W*x; where Z ₀ is the dimensional vector of the initial image, W is the pose transfer matrix, and x is the dimensional vector of the image to be migrated. The initial image obtained in this way has some features of the image to be migrated, but it is not very clear, and all the pixels in the image to be migrated have not been migrated. Using the initial image obtained in this way as the basis for subsequent calculations can improve the quality of the image to be migrated.

Optionally, step 104 includes: taking a preset image whose dimension vector is zero as the initial image. Wherein, the preset image can be stored in the memory, and is called when the image to be migrated is processed.

In this embodiment of the present disclosure, the dimensional vector corresponding to the initial image may also be assigned a value of zero, and then subsequent calculations are performed.

Step 105: Determine a target composite image according to the pose transfer matrix, the image to be transferred, and the initial image.

In this embodiment of the present disclosure, step 105 includes: obtaining an intermediate composite image according to a preset method, the pose transfer matrix, the image to be migrated, and the initial image; and using the intermediate composite image as a new The initial image is performed cyclically for a preset number of times according to the preset method, as well as the posture transition matrix, the image to be migrated and the initial image to obtain an intermediate composite image.

In the embodiment of the present disclosure, F(Z, Px, Py) is set to represent the target object in the image to be migrated from the gesture of the target object to the dimensional vector of the target composite image in the gesture of the reference object. Then it is required that when min∑‖F[z, P _x , P _y ]-x‖ approaches 0, the details of the image to be migrated are displayed in the target composite image. Among them, x is the dimension vector of the image to be migrated. The following solution steps for min∑‖F[z, P _x , P _y ]-x‖ approaching 0;

1) Optimize min∑‖F[z, P _x , P _y ]-x‖ to get

2) Let A=(W[P _x ,P _y ]) ^T W[P _x ,P _y ], b=(W[P _x ,P _y ]) ^T x ; then

Carry out inverse problem modeling to solve equation system AZ=b;

3) Set the solution precision e=0.0000001, then, r ₀ =b-AZ ₀ ; p ₀ =r ₀ ; if r ₀ is greater than e,

r _k =r _k-1 +α _k-1 Apk-1; p _k =r _k +β _k-1 p _k-1 ;

Wherein, let A=(W) ^T W; P ₀ =r ₀ ; r ₀ =b-AZ ₀ ; b=W ^T x;

4) After sorting out the above formula, Z _k+1 = f(b, A, Z _k ), it can be seen that the target composite image Z _k+1 depends on the pose transfer matrix W, the image to be transferred x, and the initial image Z _k 's.

In the embodiment of the present disclosure, according to the above steps 1)-step 4), the preset mode described in the embodiment of the present disclosure is specifically as follows:

Z _k+1 =Z _k +α _k P _k ;

Wherein, Z _k+1 is the first dimension vector of the intermediate composite image; Z _k is the second dimension vector in the initial image; wherein W is the attitude transition matrix;

r _k =r _k-1 +α _k-1 Apk-1; p _k =r _k +β _k-1 p _k-1 ;

Wherein, let A=(W) ^T W; P ₀ =r ₀ ; r ₀ =b-AZ ₀ ; b=W ^T x; where x is the third-dimensional vector of the image to be migrated, and Z ₀ is the initial network The model obtains the initial fourth dimension vector of the initial image.

In the embodiment of the present disclosure, the preset mode refers to obtaining the target composite image by adopting the above formula Z _k+1 =Z _k +α _k P _k .

In the embodiment of the present disclosure, the attitude transition matrix W is the above W[P _x , P _y ].

Specifically, for example, the initial image obtained above is taken as Z ₀ ; then, for the first time, the attitude transition matrix W, the image to be migrated x and the initial image Z ₀ are input into the formula corresponding to the above-mentioned preset mode, to obtain: Z ₁ =Z ₀ +α ₀ P ₀ ;

Wherein, r ₀ =b-Az ₀ ; P ₀ =r ₀ ; then α ₀ =1/A=1/(W) ^T W; P ₀ =b-Az ₀ =W ^T x-(W) ^T W· Z ₀ ; then Z ₁ =Z ₀ +(1/(W) ^T W)·(W ^T x-(W) ^T W·Z ₀ )=x/WZ ₀ . Finally, Z ₁ =x/WZ ₀ is obtained, wherein Z ₁ is the first time that the attitude transition matrix W, the image to be migrated x and the initial image Z ₀ are input into the formula corresponding to the above preset method, and the dimensional vector of the intermediate composite image is obtained; Z ₀ is the dimensional vector of the obtained initial image.

Taking the intermediate composite image Z ₁ obtained above as a new initial image, input the shift matrix W, the image to be migrated x and the new initial image Z ₁ into the formula corresponding to the above preset mode for the second time, and obtain: Z ₂ =Z ₁ +α ₁ P ₁ ; where,

r ₁ =r ₀ +α ₀ Ap ₀ ; p ₁ =r ₁ +β ₀ p ₀ ;

Wherein, let A=(W) ^T W; P ₀ =r ₀ ; r ₀ =b-AZ ₀ ; b=W ^T x; Z ₂ is obtained.

In the embodiment of the present disclosure, the dimensional vector of the above-mentioned image may be a one-dimensional vector, a two-dimensional vector, or a three-dimensional vector, which is not limited herein.

In the embodiment of the present disclosure, the preset number of times is greater than or equal to 2; when the preset number of times is 2, the final target composite image is Z ₂ . Wherein, when the final obtained target composite image Z ₂ is not clear enough in detail, the intermediate composite image as the new initial image can be cyclically executed, and the posture according to the preset method and the posture can be executed cyclically for a preset number of times. The transition matrix, the image to be migrated, and the initial image are used to obtain an intermediate composite image, until the target composite image is obtained to be a satisfactory image for the user.

In the embodiment of the present disclosure, it includes: multiple frames of reference images; the reference images include a time sequence; after step 105, the method further includes: arranging the multiple frames of the target composite images according to the time sequence to obtain a target composite video .

In the embodiment of the present disclosure, the method further includes: inputting the target composite image into the completion model to obtain the final composite image. Among them, the completion model is used to complete the missing parts in the target synthetic image. For example, when the image to be migrated input by the user is an image lacking a face or a part of limbs, the completion model completes these missing parts.

Specifically, the completion model can be trained based on a large number of images as training samples; for example, a back photo (without face photo), a legless photo, an armless photo and a corresponding full body photo are used as training samples to train the completion model.

Among them, multiple frames of reference images with a time sequence form a reference video; the user can click to upload the reference video, the reference video includes: multiple frames of reference images, and the multiple frames of reference images have a corresponding time sequence; Steps 101 to 105 are performed in sequence for each frame of image in the video, and finally a multi-frame target composite image is obtained, and the multi-frame target composite image is arranged in a time series to obtain the final target composite video.

In the embodiment of the present disclosure, the method further includes: identifying each frame of images in the reference video, selecting an image including a human object as a reference image; taking an image not including a human object as a transition image; then finally synthesizing the multi-frame target image and the transition image The images are arranged in time series to obtain the target composite video.

Wherein, the reference video includes dancing actions or other actions, which are not limited herein.

In this embodiment of the present disclosure, the step 105 includes: extracting the target object in the image to be migrated; determining a composite object according to the pose transfer matrix, the target object and the initial image; The background of the image is combined with the composite object to obtain the target composite image.

In the embodiment of the present disclosure, only the target object in the image to be migrated is migrated, and the background of the target object in the image to be migrated is not migrated. After migrating the target object in the image to be migrated, the obtained human body The object is a composite object, and then the background of the reference image and the composite object are composited. Referring to FIG. 3 , after the target object corresponding to the migration image A is migrated, the background of the reference image B is used to obtain the target composite image C.

In the embodiment of the present disclosure, referring to FIG. 2 , the entire image A to be migrated may also be migrated to obtain the target composite image C.

FIG. 4 is a block diagram of an image processing apparatus provided by an embodiment of the present disclosure. As shown in the figure, the apparatus may include:

The third acquisition module is used to acquire the initial image;

The image processing apparatus provided by the embodiment of the present disclosure has functional modules corresponding to executing the image processing method, can execute the image processing method provided by the embodiment of the present disclosure, and can achieve the same beneficial effects.

In yet another embodiment provided by the present disclosure, an electronic device is also provided. The electronic device may include: a processor, a memory, and a computer program stored on the memory and executable on the processor, the When the processor executes the program, each process of the above image processing method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, details are not repeated here. For example, as shown in FIG. 5 , the electronic device may specifically include: a processor 301 , a storage device 302 , a display screen 303 with a touch function, an input device 304 , an output device 305 , and a communication device 306 . The number of processors 301 in the electronic device may be one or more, and one processor 301 is taken as an example in FIG. 5 . The processor 301 , the storage device 302 , the display screen 303 , the input device 304 , the output device 305 and the communication device 306 of the electronic device may be connected by a bus or in other ways.

In yet another embodiment provided by the present disclosure, a computer-readable storage medium is also provided, where instructions are stored in the computer-readable storage medium, when the computer-readable storage medium runs on a computer, the computer causes the computer to execute any one of the foregoing embodiments. the image processing method.

In yet another embodiment provided by the present disclosure, there is also provided a computer program product containing instructions, which when run on a computer, causes the computer to execute the image processing method described in any one of the foregoing embodiments.

It should be noted that, in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any relationship between these entities or operations. any such actual relationship or sequence exists. Moreover, the terms "comprising", "comprising" or any other variation thereof are intended to encompass a non-exclusive inclusion such that a process, method, article or device that includes a list of elements includes not only those elements, but also includes not explicitly listed or other elements inherent to such a process, method, article or apparatus. Without further limitation, an element qualified by the phrase "comprising a..." does not preclude the presence of additional identical elements in a process, method, article or apparatus that includes the element.

Each embodiment in this specification is described in a related manner, and the same and similar parts between the various embodiments may be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, as for the system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and for related parts, please refer to the partial descriptions of the method embodiments.

The above descriptions are only preferred embodiments of the present disclosure, and are not intended to limit the protection scope of the present disclosure. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure are included in the protection scope of the present disclosure.

Claims

An image processing method, characterized in that the method comprises:

Acquiring an image to be migrated and a reference image; the image to be migrated includes: a target object whose posture is to be converted; the reference image includes: a reference object showing a reference posture;

acquiring the first key feature of the target object and the second key feature of the reference object;

Determine a posture transition matrix according to the first key feature and the second key feature;

get the initial image;

A target composite image is determined according to the pose transfer matrix, the image to be transferred, and the initial image.
The method according to claim 1, wherein the acquiring an initial image comprises:

Inputting the pose transfer matrix and the image to be transferred into an initial network model to obtain an initial image.
The method according to claim 1, wherein the acquiring an initial image comprises:

A preset image whose dimension vector is zero is used as the initial image.
The method according to claim 1, wherein the determining the target composite image according to the pose transfer matrix, the image to be transferred and the initial image comprises:

Obtain an intermediate composite image according to a preset manner, and the pose transfer matrix, the image to be transferred, and the initial image;

Taking the intermediate composite image as the new initial image, and cyclically executing the preset method for a preset number of times, as well as the attitude transition matrix, the image to be migrated, and the initial image, to obtain the intermediate composite image. step to obtain the target composite image.
The method according to claim 4, wherein the preset mode is as follows:

Z k+1 =Z k +α k P k ;

Wherein, Z k+1 is the first dimension vector of the intermediate composite image; Z k is the second dimension vector in the initial image; wherein W is the attitude transition matrix;
r k =r k-1 +α k-1 Apk-1; p k = r k +β k-1 p k-1 ;
Wherein, let A=(W) T W; P 0 =r 0 ; r 0 =b-AZ 0 ; b=W T x; where x is the third-dimensional vector of the image to be migrated, and Z 0 is the initial network The model obtains the initial fourth dimension vector of the initial image.
The method according to claim 1, wherein the first key feature is a preset key feature in the target object; the second key feature is in one-to-one correspondence with the first key feature.
The method according to claim 6, wherein the determining a posture transition matrix according to the first key feature and the second key feature comprises:

Determine the coordinate value of each of the first key features, and the coordinate value of each of the second key features;

The attitude transfer matrix is determined according to the coordinate value of the first key feature and the coordinate value of the second key feature, and the attitude transfer matrix is used to convert the coordinate value of the first key feature into a The coordinate value of the second key feature corresponding to the first key feature.
The method according to claim 7, characterized in that, comprising: multiple frames of reference images; the reference images comprise time series;

Then, after determining the target composite image according to the posture transfer matrix, the image to be transferred and the initial image, the method further includes:

Arrange multiple frames of the target composite images according to the time sequence to obtain a target composite video.
The method according to claim 1, wherein the determining the target composite image according to the pose transfer matrix, the image to be transferred and the initial image comprises:

extracting the target object in the image to be migrated;

determining a synthetic object according to the pose transfer matrix, the target object and the initial image;

Synthesize the background of the reference image and the synthetic object to obtain the target synthetic image.
An image processing device, characterized in that the device comprises:

a first acquisition module, configured to acquire an image to be migrated and a reference image; the image to be migrated includes: a target object whose posture is to be converted; the reference image includes: a reference object that presents a reference posture;

a second acquisition module, configured to acquire the first key feature of the target object and the second key feature of the reference object;

a first determining module, configured to determine a posture transition matrix according to the first key feature and the second key feature;

The third acquisition module is used to acquire the initial image;

The second determining module is configured to determine a target composite image according to the posture transfer matrix, the image to be transferred and the initial image.
An electronic device, characterized in that the electronic device includes a processor, a memory, and a program or instruction stored on the memory and executable on the processor, and the program or instruction is executed by the processor while implementing the steps of the method according to any one of claims 1-9.
A readable storage medium, characterized in that a program or an instruction is stored on the readable storage medium, and when the program or instruction is executed by a processor, the steps of the method according to any one of claims 1-9 are implemented.
A computer program product comprising computer readable code which, when run on a computing processing device, causes the computing processing device to perform the steps of the method according to any one of claims 1-9.