Picture generation method and system based on artificial intelligence

Abstract

The invention belongs to the technical field of image processing, and particularly relates to a picture generation method and system based on artificial intelligence, and the method comprises the steps: S1, an uploading stage; s2, a positioning segmentation stage; s3, an instruction analysis and region binding stage; s4, a conflict detection and negotiation stage; and S5, a backup and encryption stage. Through the cooperation of the structure, the picture can be revised by multiple persons synchronously, the picture is divided into a plurality of parts through a mask division mode, then persons participating in picture revising in a group photo select corresponding masks to revise the picture, compared with a traditional alternate picture revising mode, the picture revising method is more convenient and faster, the picture generation efficiency is improved based on the method, and the picture revising efficiency is improved. And meanwhile, when multiple persons edit the masks in the same area, conflicts are solved by adopting conflict classification and priority rules and a real-time interaction negotiation mode, so that the smooth operation of multi-person collaborative editing is ensured.

Description

Technical Field

[0001] The present invention belongs to the field of image processing technology, and specifically relates to an image generation method and system based on artificial intelligence. Background Art

[0002] With the continuous progress of society and the improvement of people's quality of life, taking photos to record daily life has become a widely popular lifestyle. After taking photos with their mobile phones, many users will use software to perform post-production adjustments based on their personal aesthetics and needs to enhance the artistic effect of the photos. One-click photo editing is a common function in existing photo editing software. Although one-click photo editing is efficient, its universal design cannot meet the different editing needs of everyone. Therefore, in real life, when taking group photos, due to different personal aesthetics, users will take turns using software to edit the photos according to their own needs, resulting in a longer editing process and low image generation efficiency.

[0003] To this end, the present invention provides an artificial intelligence-based image generation method and system. Summary of the Invention

[0004] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.

[0005] The technical solution adopted by the present invention to solve the technical problem is: an artificial intelligence-based image generation method described in the present invention comprises the following steps:

[0006] S1, Upload stage: After the image to be processed is standardized, it is uploaded in a format supported by the system and the original image is saved at the same time;

[0007] S2, positioning and segmentation stage: The model automatically identifies people, objects, and background in the image and generates interactive masks;

[0008] S3, instruction parsing and region binding stage: bind the parsed text instructions to the specific mask segmented in the S2 positioning and segmentation stage;

[0009] S4, conflict detection and negotiation phase: resolve conflicts among multiple people modifying the same mask;

[0010] S5, backup and encryption stage: back up the generated images and verify the identity information when storing the images.

[0011] Furthermore, the S2 positioning and segmentation stage performs recognition operations through a semantic segmentation model, and the model formula is:

[0012] M=SAM(I)

[0013] Where M is a set of segmentation masks, i.e., the masks of each region obtained after segmenting the input image. Each mask corresponds to an independent region in the image, including people, objects, and background. SAM is a semantic segmentation model, which is a model used to perform segmentation operations on the input image. I is the original input image.

[0014] Furthermore, the S2 positioning and segmentation stage supports the user to fine-tune the segmented area by clicking and describing the area in text.

[0015] Furthermore, the instruction parsing in the S3 instruction parsing and region binding stage is implemented through multimodal instruction parsing and context association, wherein the multimodal instruction parsing combines the natural language processing model with target detection to identify the region keywords and operation keywords in the instructions, and parses the location keywords in the instructions through a neural network.

[0016] Furthermore, the context association is implemented by historical instruction analysis and image content association:

[0017] Historical command analysis involves recording the user's historical commands and performing contextual association with the current command.

[0018] Image content association is reflected in combining the image segmentation mask and target detection results to analyze the association between instructions and image content;

[0019] The historical instruction analysis finds historical instructions related to the current instruction through semantic similarity calculation;

[0020] Image content association maintains continuous attention on the target area through target tracking technology, infers the attributes of the target area based on the context, optimizes instruction parsing, and processes sequential instructions through long short-term memory networks to capture contextual dependencies.

[0021] Furthermore, the multimodal instruction parsing formula is:

[0022] E=Bind(NLP(T),M)

[0023] Among them, E is the editing instruction bound to a specific area, that is, the operation instruction for a specific area of ​​the image parsed according to the user instruction; Bind is a function used to bind the parsed text instruction to a specific area in the image; NLP stands for the natural language processing model, which is used to parse the natural language instruction T entered by the user and extract the area keywords and operation keywords therein; T is the natural language instruction entered by the user; M is the segmentation mask set, that is, the image area mask previously obtained by the semantic segmentation model.

[0024] Furthermore, the context association formula is:

[0025] E* =ContextAssociation(T, H * , M)

[0026] Among them, E * The context-optimized editing instructions are the result of optimizing the original editing instructions based on historical instructions and the current image content.

[0027] ContextAssociation is a function used to combine the current instruction T with the historical instruction analysis results H * and segmentation mask set M, to perform context association and instruction optimization;

[0028] T is the natural language instruction currently input by the user;

[0029] H * It is the result of semantic analysis of historical instructions, including key information and contextual relationships in historical instructions;

[0030] M is a set of segmentation masks, i.e., the region mask of the image.

[0031] Furthermore, the S4 conflict detection and negotiation phase uses the NSGA-II algorithm to balance the command priorities of different users. Its principle is to select excellent individuals through non-dominated sorting and congestion distance, thereby finding a set of balanced solutions among multiple objectives. When processing the command priorities of different users, each user's command and its priority are regarded as a goal, and a set of solutions that balance the needs of different users are found through the NSGA-II algorithm. Its operation process includes:

[0032] Objective function: Each user's instruction and its priority are considered as the objective function. Assume there are n users, each user i proposes an instruction and the corresponding priority pi. The decision variable is the execution degree of each user's instruction or the resource allocation ratio.

[0033] Encoding scheme: Use chromosomes to represent individuals, each individual is a vector, and the elements correspond to the execution degree of each user instruction;

[0034] Fitness function: For each user i, define the objective function fi(x) to measure the execution effect or satisfaction of the instruction, with the goal of maximizing the satisfaction of all users;

[0035] NSGA-II algorithm process:

[0036] Initialization: Randomly generate an initial population, where each individual is represented as a possible instruction execution plan;

[0037] Non-dominated sorting: Perform non-dominated sorting on all individuals in the population and divide the individuals into different non-dominated layers;

[0038] Crowding distance calculation: Calculate the crowding distance for each individual in the non-dominated layer;

[0039] Selection, crossover, and mutation: Generate the next generation of individuals through tournament selection, crossover, and mutation operations;

[0040] Iteration: Repeat the above steps until the maximum number of iterations is reached.

[0041] An artificial intelligence-based image generation system, applied to the artificial intelligence-based image generation method, includes: an upload module, a processing module, a storage module and an identity authentication module.

[0042] The upload module is used to upload the image to be processed to the system and provide real-time feedback on the upload progress and upload status;

[0043] The processing module is used to segment the image and perform collaborative modification on the segmented image to generate a new image;

[0044] The storage module is used to store the output generated pictures and record the modified and generated related parameters and content to form preference data;

[0045] The authentication module is used to verify the user's identity and record all user operations, including login, logout, image upload, editing, download and storage.

[0046] Furthermore, the processing module includes a parameter setting unit and a feedback unit, and the setting unit is used for the user to set relevant parameters for image generation, and the feedback unit is used to evaluate the quality of the generated image and make modification suggestions. The basis for the modification suggestions includes clarity, color deviation and contrast.

[0047] The beneficial effects of the present invention are as follows:

[0048] 1. It enables multiple people to edit photos simultaneously. By using a split mask method, the photo is divided into several parts. Then, each person in the group photo can select a corresponding mask to edit the photo. Compared with the traditional method of taking turns to edit, this is more convenient and faster, and thus improves the efficiency of image creation. At the same time, when multiple people edit the same mask area, conflict classification and priority rules are used, as well as real-time interactive negotiation to resolve conflicts, ensuring smooth collaborative editing.

[0049] 2. A feedback unit is provided to evaluate the quality of the generated image in terms of clarity, color deviation and contrast, and to make modification suggestions to improve the artistic effect of the generated image. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The present invention will be further described below with reference to the accompanying drawings.

[0051] Figure 1 This is a method flow chart of an artificial intelligence-based image generation method and system of the present invention;

[0052] Figure 2 This is a system structure diagram of an artificial intelligence-based image generation method and system of the present invention. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] See also Figure 1-2 The present invention provides: a method and system for generating pictures based on artificial intelligence; Example 1

[0055] An artificial intelligence-based image generation method comprises the following steps:

[0056] S1, Upload stage: After the image to be processed is standardized, it is uploaded in a format supported by the system and the original image is saved at the same time;

[0057] S2, positioning and segmentation stage: The model automatically identifies people, objects, and background in the image and generates interactive masks;

[0058] The S2 positioning and segmentation stage performs recognition operations through the semantic segmentation model. The model formula is:

[0059] M=SAM(I)

[0060] Among them, M is the segmentation mask set, that is, the masks of each area obtained after segmenting the input image. Each mask corresponds to an independent area in the image, which includes people, objects and background; SAM is the semantic segmentation model, a model used to perform segmentation operations on the input image; I is the original input image. Assume that the input image I is a group photo containing three people A, B, C and the background. After processing by the semantic segmentation model SAM, the segmentation mask set M is obtained, which includes person A mask M1, person B mask M2, person C mask M3 and background mask M4. These masks M1, M2, M3 and M4 together constitute the segmentation mask set M.

[0061] The S2 positioning and segmentation stage supports users to fine-tune the segmented area through clicks and text descriptions. For example, User 1 enters "Character A wears a Christmas hat", and User 2 enters "Character A wears a birthday hat" → the system prompts in the form of a pop-up window: "Character A's hat styles conflict, please choose to keep the Christmas hat or the birthday hat."

[0062] S3, instruction parsing and region binding stage: bind the parsed text instructions to the specific mask segmented in the S2 positioning and segmentation stage;

[0063] In the S3 command parsing and region binding phase, command parsing is achieved through multimodal command parsing and context association. Multimodal command parsing combines natural language processing models with object detection to identify region keywords and operation keywords in commands. Position keywords in commands are parsed through neural networks. For example, the user enters command T: "Remove the scar on person A's face."

[0064] First, the instruction is parsed through the natural language processing model NLP to extract the regional keyword "person A" and the operation keyword "scar removal". Then, the function Bind finds the corresponding mask MA-face in the segmentation mask set M based on the regional keyword "person A", and binds the operation instruction "scar removal" to the mask to generate the editing instruction E, which is to perform the scar removal operation on the person A area.

[0065] Context association is achieved through historical command analysis and image content association:

[0066] Historical command analysis involves recording the user's historical commands and performing contextual association with the current command.

[0067] Image content association is reflected in combining the image segmentation mask and target detection results to analyze the association between instructions and image content;

[0068] The historical instruction analysis finds historical instructions related to the current instruction through semantic similarity calculation;

[0069] Image content association maintains continuous attention on the target area through target tracking technology, infers the attributes of the target area based on the context, optimizes instruction parsing, and processes sequential instructions through long short-term memory networks to capture contextual dependencies.

[0070] The multimodal instruction parsing formula is:

[0071] E=Bind(NLP(T),M)

[0072] Among them, E is the editing instruction bound to a specific area, that is, the operation instruction for a specific area of ​​the image parsed according to the user instruction; Bind is a function used to bind the parsed text instruction to a specific area in the image; NLP stands for the natural language processing model, which is used to parse the natural language instruction T entered by the user and extract the area keywords and operation keywords therein; T is the natural language instruction entered by the user; M is the segmentation mask set, that is, the image area mask previously obtained by the semantic segmentation model.

[0073] The context association formula is:

[0074] E * =ContextAssociation(T, H * , M)

[0075] Among them, E * The context-optimized editing instructions are the result of optimizing the original editing instructions based on historical instructions and the current image content.

[0076] ContextAssociation is a function used to combine the current instruction T with the historical instruction analysis results H * and segmentation mask set M, to perform context association and instruction optimization;

[0077] T is the natural language instruction currently input by the user;

[0078] H * It is the result of semantic analysis of historical instructions, including key information and contextual relationships in historical instructions;

[0079] M is the segmentation mask set, that is, the region mask of the image;

[0080] For example, suppose the user previously entered the command "Change character B into a red short-sleeved shirt", and the current user enters the command T: "Put a hat on character A";

[0081] First, analyze the historical instructions to get H * , which contains the information about the clothing modification of character B; then, the function ContextAssociation combines the current instruction T and the historical instruction analysis result H * and segmentation mask set M, generate optimized editing instructions E * That is, after considering historical instructions, when putting a hat on character A, it may automatically select a more compatible hat style based on the previous clothing modification style of character B, or ensure that the edit of character A does not conflict with the edit of character B.

[0082] S4, conflict detection and negotiation phase: resolve conflicts among multiple people modifying the same mask;

[0083]

[0084] In the S4 conflict detection and negotiation phase, the NSGA-II algorithm is used to balance the command priorities of different users. Its principle is to select excellent individuals through non-dominated sorting and congestion distance to find a set of balanced solutions among multiple objectives. When processing the command priorities of different users, each user's command and its priority are treated as a goal. The NSGA-II algorithm is used to find a set of solutions that balance the needs of different users. Its operation process includes:

[0085] Objective function: Each user's instruction and its priority are considered as the objective function. Assume there are n users, each user i proposes an instruction and the corresponding priority pi. The decision variable is the execution degree of each user's instruction or the resource allocation ratio.

[0086] Encoding scheme: Use chromosomes to represent individuals, each individual is a vector, and the elements correspond to the execution degree of each user instruction;

[0087] Fitness function: For each user i, define the objective function fi(x) to measure the execution effect or satisfaction of the instruction, with the goal of maximizing the satisfaction of all users;

[0088] NSGA-II algorithm process:

[0089] Initialization: Randomly generate an initial population, where each individual is represented as a possible instruction execution plan;

[0090] Non-dominated sorting: Perform non-dominated sorting on all individuals in the population and divide the individuals into different non-dominated layers;

[0091] Crowding distance calculation: Calculate the crowding distance for each individual in the non-dominated layer;

[0092] Selection, crossover, and mutation: Generate the next generation of individuals through tournament selection, crossover, and mutation operations;

[0093] Iteration: Repeat the above steps until the maximum number of iterations is reached.

[0094] For example, suppose there are three users A, B, and C, each of whom issues a command to modify a group photo. The priority of each command is pA=3, pB=2, and pC=1 respectively. Our goal is to maximize the overall satisfaction by adjusting the execution degree XA, XB, and XC of each command.

[0095] Objective function:

[0096] fA(x)=3·XA

[0097] fB(x)=2·XB

[0098] fC(x)=1·XC

[0099] Constraints: Assume that total resources are limited, such as XA+XB+XC≤1, and xi≥0;

[0100] NSGA-II algorithm:

[0101] Initialization: Randomly generate the initial population:

[0102]

[0103] Non-dominated sorting:

[0104] Calculate the objective function value for each individual:

[0105]

[0106] Crowding distance calculation: For individuals in F1, calculate the distance to each target and sum them up;

[0107] Environmental selection: select individuals to enter the next generation according to the order of non-dominated layers and crowding distance, such as generating [0.45, 0.35, 0.2] through crossover;

[0108] Genetic operation: crossover and mutation of selected individuals to generate new individuals;

[0109] Iteration: A set of equilibrium solutions is obtained: x = [0.5, 0.3, 0.2], which means that under resource constraints, the execution levels of instructions for users A, B, and C are 0.5, 0.3, and 0.2, respectively, which satisfies the priority of each user as much as possible.

[0110] S5, backup and encryption stage: back up the generated images and verify the identity information when storing the images.

[0111] Example 2

[0112] This embodiment provides:

[0113] The S4 conflict detection and negotiation phase also includes version management and collaborative backtracking, which supports version control and historical backtracking during multi-person collaboration. This is achieved through a Git-style version tree: each command submission generates a version node, supporting branching and merging. For example:

[0114] Version V1.0 (initial photo)

[0115] ├─Branch A: User 1 "Change character B's hairstyle" → V1.1│└─User 2 "Add snowflakes to background" → V1.2

[0116] └─Branch B: User 3 "Character C plus glasses" → V1.3 └─Merge branch A → V2.0 (final photo);

[0117] In summary, the processing of S2, S3, and S4 is as follows:

[0118]

[0119]

[0120] Example 3

[0121] This embodiment provides:

[0122] The instructions in the S2 positioning and segmentation stage also include voice instructions. For example, suppose we have a voice message "Hello, how are you?" The following is the voice-to-text conversion process:

[0123] Signal preprocessing: remove background noise from speech and determine the start and end points of speech.

[0124] Feature extraction: Calculate the MFCC features of the speech signal.

[0125] Acoustic modeling: Use DNN to map MFCC features to phoneme sequences, such as "HELLO, HO WARE YO U?".

[0126] Language modeling: Use the LSTM language model to predict the word sequence corresponding to the phoneme sequence.

[0127] Decoding: Find the most likely word sequence “Hel lo, how are you?” through beam search.

[0128] Through the above steps, the voice signal is successfully converted into text, and then it can be bound to the specific mask segmented in the S2 positioning and segmentation stage through text instructions to meet the image editing needs of special people.

[0129] Example 4

[0130] This embodiment provides:

[0131] An artificial intelligence-based image generation system is applied to an artificial intelligence-based image generation method, comprising: an upload module, a processing module, a storage module, and an identity authentication module.

[0132] The upload module is used to upload the images to be processed to the system and provide real-time feedback on the upload progress and upload status;

[0133] The processing module is used to segment the image and perform collaborative modification on the segmented image to generate a new image;

[0134] The storage module is used to store the output images and record the modified and generated parameters and content to form preference data;

[0135] The authentication module is used to verify the user's identity and record all user operations, including login, logout, image upload, editing, and download storage.

[0136] The processing module includes a parameter setting unit and a feedback unit. The setting unit is used for users to set relevant parameters for image generation. The feedback unit is used to evaluate the quality of the generated images and make modification suggestions. The basis for the modification suggestions includes clarity, color deviation and contrast.

[0137] It is important to note that if the image clarity is low, it is recommended to adjust the parameters in the image generation process, such as increasing the number of iterations or adjusting the learning rate of the optimization algorithm.

[0138] Color deviation: If the image has color deviation, it is recommended to adjust the color balance parameters or use a color correction algorithm to correct it;

[0139] Poor contrast: If the image contrast is insufficient, it is recommended to adjust the contrast enhancement parameters or use the histogram equalization method to enhance the contrast;

[0140] For example, suppose the generated image is blurry. The feedback unit calculates the image gradient and finds that the clarity is low. The system will then make the following modification suggestions:

[0141] Adjust generation parameters: Increase the number of iterations in the image generation process from the original 100 to 200 to improve image clarity.

[0142] The directions or positional relationships indicated by terms such as "center", "longitudinal", "lateral", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside" and "outside" are based on the directions or positional relationships shown in the accompanying drawings. They are merely simplified descriptions for the convenience of describing the present invention, and do not indicate or imply that the devices or elements referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the protection content of the present invention.

[0143] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

[0144] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for generating an image based on artificial intelligence, characterized in that: The following steps are involved: S1, Upload stage: After the image to be processed is standardized, it is uploaded in a format supported by the system and the original image is saved at the same time; S2, positioning and segmentation stage: The model automatically identifies people, objects, and background in the image and generates interactive masks; S3, instruction parsing and region binding stage: bind the parsed text instructions to the specific mask segmented in the S2 positioning and segmentation stage; S4, conflict detection and negotiation phase: resolve conflicts among multiple people modifying the same mask; S5, backup and encryption stage: back up the generated images and verify the identity information when storing the images.

2. The method for generating an image based on artificial intelligence according to claim 1, wherein: The S2 positioning and segmentation stage performs recognition operations through a semantic segmentation model, and the model formula is: M=SAM(I) Where M is a set of segmentation masks, i.e., the masks of each region obtained after segmenting the input image. Each mask corresponds to an independent region in the image, including people, objects, and background. SAM is a semantic segmentation model, which is a model used to perform segmentation operations on the input image. I is the original input image.

3. The method for generating an image based on artificial intelligence according to claim 2, wherein: The S2 positioning and segmentation stage supports users to fine-tune the segmentation area by clicking and describing the text.

4. The method for generating an image based on artificial intelligence according to claim 3, wherein: The instruction parsing in the S3 instruction parsing and region binding stage is achieved through multimodal instruction parsing and context association, wherein multimodal instruction parsing combines natural language processing models with target detection to identify regional keywords and operation keywords in instructions, and parses location keywords in instructions through neural networks.

5. The method for generating an image based on artificial intelligence according to claim 4, wherein: The implementation of the context association includes historical instruction analysis and image content association: Historical command analysis involves recording the user's historical commands and performing contextual association with the current command. Image content association is reflected in combining the image segmentation mask and target detection results to analyze the association between instructions and image content; The historical instruction analysis finds historical instructions related to the current instruction through semantic similarity calculation; Image content association maintains continuous attention on the target area through target tracking technology, infers the attributes of the target area based on the context, optimizes instruction parsing, and processes sequential instructions through long short-term memory networks to capture contextual dependencies.

6. The method for generating an image based on artificial intelligence according to claim 4, wherein: The multimodal instruction parsing formula is: E=Bind(NLP(T),M) Among them, E is the editing instruction bound to a specific area, that is, the operation instruction for a specific area of ​​the image parsed according to the user instruction; Bind is a function used to bind the parsed text instruction to a specific area in the image; NLP stands for the natural language processing model, which is used to parse the natural language instruction T entered by the user and extract the area keywords and operation keywords therein; T is the natural language instruction entered by the user; M is the segmentation mask set, that is, the image area mask previously obtained by the semantic segmentation model.

7. The method for generating an image based on artificial intelligence according to claim 5, wherein: The context association formula is: E * =ContextAssociation(T,H * ,M) Among them, E * The context-optimized editing instructions are the result of optimizing the original editing instructions based on historical instructions and the current image content. ContextAssociation is a function used to combine the current instruction T with the historical instruction analysis results H * and segmentation mask set M, to perform context association and instruction optimization; T is the natural language instruction currently input by the user; H * It is the result of semantic analysis of historical instructions, including key information and contextual relationships in historical instructions; M is a set of segmentation masks, i.e., the region mask of the image.

8. The method for generating an image based on artificial intelligence according to claim 1, wherein: The S4 conflict detection and negotiation phase uses the NSGA-II algorithm to balance the command priorities of different users. Specifically, it selects excellent individuals through non-dominated sorting and congestion distance to find a set of balanced solutions among multiple objectives. When processing the command priorities of different users, each user's command and its priority are treated as a goal. The NSGA-II algorithm is used to find a set of solutions that balance the needs of different users. The operation process includes: Objective function: Each user's instruction and its priority are considered as the objective function. Assume there are n users, each user i proposes an instruction and the corresponding priority pi. The decision variable is the execution degree of each user's instruction or the resource allocation ratio. Encoding scheme: Use chromosomes to represent individuals, each individual is a vector, and the elements correspond to the execution degree of each user instruction; Fitness function: For each user i, define the objective function fi(x) to measure the execution effect or satisfaction of the instruction, with the goal of maximizing the satisfaction of all users; NSGA-II algorithm process: Initialization: Randomly generate an initial population, where each individual is represented as a possible instruction execution plan; Non-dominated sorting: Perform non-dominated sorting on all individuals in the population and divide the individuals into different non-dominated layers; Crowding distance calculation: Calculate the crowding distance for each individual in the non-dominated layer; Selection, crossover, and mutation: Generate the next generation of individuals through tournament selection, crossover, and mutation operations; Iteration: Repeat the above steps until the maximum number of iterations is reached.

9. An artificial intelligence-based image generation system, applied to the artificial intelligence-based image generation method according to any one of claims 1 to 8, characterized in that: include: Upload module, processing module, storage module and authentication module, The upload module is used to upload the image to be processed to the system and provide real-time feedback on the upload progress and upload status; The processing module is used to segment the image and perform collaborative modification on the segmented image to generate a new image; The storage module is used to store the output generated pictures and record the modified and generated related parameters and content to form preference data; The authentication module is used to verify the user's identity and record all user operations, including login, logout, image upload, editing, download and storage.

10. The artificial intelligence-based image generation system according to claim 9, characterized in that: The processing module includes a parameter setting unit and a feedback unit. The setting unit is used for the user to set the relevant parameters for image generation. The feedback unit is used to evaluate the quality of the generated image and make modification suggestions. The basis for the modification suggestions includes clarity, color deviation and contrast.

Images