An image optimization method for game character and biological model design

Abstract

The invention discloses an image optimization method for game character and biological model design, which relates to the field of 3D game technology, including: classifying character design and defining components from the perspective of biological population; constructing different semantic tree models for different operations; performing data preparation and preprocessing of semantic tree expression for multiple types of three-dimensional characters from different sources; finding segmentation boundaries through search and verification of geometric features of end points and curvatures; performing pre-segmentation of components, and obtaining semantic tree model subtree segmentation results through boundary optimization; using semantic tree model subtrees as genetic genes to create novel character models; combining user selection with probability graph models as fitness judgment criteria in evolution to optimize solution space search paths. The problem of lack of structural innovation in game character models is solved through the guiding rules and optimization methods of batch segmentation marking and combined modeling in model innovation design.

Description

Technical Field

[0001] The present invention relates to the field of 3D game technology, and in particular to an image optimization method for designing game characters and biological models. Background Art

[0002] As the cornerstone of building virtual worlds, 3D models are essential components of digital design. The core mission of computer-aided modeling research is to make designers' work more efficient and produce superior designs. To achieve this goal, 3D modeling research has progressed through stages such as parameterization, characterization, and structuring. From the concrete to the abstract, from the local to the global, and from geometry to semantics, a relatively comprehensive approach to model creation and editing has been gradually developed.

[0003] However, assisting and guiding creative model design is a new issue that has only received attention in recent years. Example-based learning methods are the primary approach to solving this problem. These methods include some innovative design work based on the combination of 3D model components. Their core concepts of batch learning and equal-position permutation are suitable for man-made objects with similar structures and fixed component functions. However, for the diverse and creative 3D animated characters in games and films, data collection difficulties and a lack of structural innovation remain. As design concepts become increasingly personalized, structural diversity, such as that of 3D characters, will become increasingly evident in the design of man-made objects. Therefore, it is necessary to explore a 3D model representation that allows for structural innovation and a corresponding combination design solution. Summary of the Invention

[0004] In order to solve the above technical problems, a method for image optimization for game character and biological model design is provided. This technical solution solves the problems raised in the above background technology.

[0005] In order to achieve the above objects, the technical solution adopted by the present invention is:

[0006] An image optimization method for game character and creature model design, comprising:

[0007] Classify character designs and define components from the perspective of biological populations;

[0008] Construct different semantic tree models for different operations, and obtain the acquisition methods and model construction results of various semantic tree models;

[0009] Through experiments with role models from different sources, we analyze the applicable objects and respective advantages of each role model;

[0010] Perform data preparation and semantic tree expression preprocessing for multiple types of 3D characters from different sources;

[0011] Branch view-based segmentation and labeling finds segmentation boundaries by searching and verifying the geometric features of end points and curvature under the premise of known semantic constraints;

[0012] Skeleton view-based segmentation uses the skeleton and skin information of the animated character to pre-segment the parts, and obtains the subtree segmentation results of the semantic tree model through boundary optimization;

[0013] Using semantic tree model subtrees as genetic genes, non-allelic exchange is performed to expand the solution space and create novel role models;

[0014] Combining user selection with a probabilistic graphical model as the fitness judgment criterion in evolution, the solution space search path is optimized and the acceptability of the results is improved.

[0015] Preferably, the classification of character design and component definition from the perspective of biological populations specifically includes:

[0016] The character model parts are defined as a head, a torso, limbs, and appendages, wherein the appendages include wings, a tail, and tentacle structures;

[0017] Based on the character model components, the character models are classified into two-legged upright type, two-legged head-up type, four-legged head-up type, multi-legged head-up type, legless horizontal type and four-legged horizontal type characters;

[0018] Bipedal characters have complete upright legs and feet, with the feet touching the ground as support, and obvious upper limb structures, modeled with reference to human form;

[0019] The bipedal head-raising character has slender legs, a long neck covered with feathers, and a pair of wings on the back of the torso. It is modeled based on the morphology of birds.

[0020] A quadruped head-raising character has strong front and back limbs that touch the ground for support. The torso is parallel to the ground and at a distance from the ground. The head is clearly located above the front of the torso and extends forward. The modeling is based on the morphology of quadruped mammals.

[0021] The multi-legged head-raising character is similar to a large quadruped animal, while maintaining the characteristics of multi-legged limbs. The various accessories are increased and the modeling is based on the morphology of arthropod insects.

[0022] The legless horizontal character has a simple structure, consisting of only two parts: the torso and the head. It is modeled based on the morphology of reptiles such as fish and snakes.

[0023] The design style of the four-legged horizontal character is similar to that of the multi-legged head-raising character. The outline of the torso parts is unclear, and the legs are relatively thick. The most notable feature is that the model is close to the ground and is modeled with reference to the morphology of large reptiles.

[0024] Preferably, the method of constructing different semantic tree models for different operations and obtaining various semantic tree models and the model construction results specifically include:

[0025] Express the obvious connection relationship between the components in a role model from the perspective of general user cognition and build a basic view semantic tree model;

[0026] Express the connection relationship between the various components in a character model from the animator's cognitive perspective, determine the skeleton information and skin information of the semantic tree model, and build a skeletal view semantic tree model;

[0027] Perform consistent segmentation on a batch of role models with the same structure and construct a branch view semantic tree model;

[0028] Expand on the basic view semantic tree model, introduce an empty component symbol to describe the possible expansion area, and build the design view semantic tree model.

[0029] Preferably, the data preparation and semantic tree expression preprocessing work for multiple types of three-dimensional characters from different sources specifically includes:

[0030] Get at least one role model from the model database;

[0031] Based on the acquisition methods of various semantic tree models, various semantic tree models of different types of 3D characters are constructed;

[0032] At least one character model in the model database is input into various semantic tree models of different types of three-dimensional characters to obtain preprocessing models of the multiple types of three-dimensional characters.

[0033] Preferably, the segmentation and labeling based on the branch view, under the premise of known semantic constraints, is performed by searching and verifying the geometric features of the end points and curvatures to find the segmentation boundary, specifically including:

[0034] Using morphological simulation to perform corrosion operations, the objects in the preprocessing model are reduced in size and the parts smaller than the preset threshold are eliminated.

[0035] The output residual is the end point of the preprocessing model;

[0036] Get the parameter equations of the preprocessing model;

[0037] The arc length parameter of the preprocessing model is calculated using the arc length parameter formula;

[0038] Based on the arc length parameter of the preprocessing model, the unit tangent vector of any point on the preprocessing model is calculated using the unit tangent vector formula;

[0039] Calculate the curvature of any point on the preprocessed model using the curvature formula;

[0040] Based on the curvature of each point between the end points, the segmentation boundary is found and the pre-processed model is segmented;

[0041] The parametric equation is:

[0042]

[0043] Where, are the preprocessing model parameters, are the coordinates of each point in the preprocessing model;

[0044] The arc length parameter formula is:

[0045]

[0046] Where, is the arc length parameter of the preprocessing model, is the first derivative of x with respect to t;

[0047] The normalized tangent vector formula is:

[0048]

[0049] Where, is the unit tangent vector of any point on the preprocessed model;

[0050] The curvature formula is:

[0051]

[0052] Where, is the curvature at any point on the preprocessed model, for The derivative with respect to t.

[0053] Preferably, the segmentation based on the skeleton view utilizes the skeleton and skin information of the animated character to pre-segment the components, and the semantic tree model subtree segmentation result is obtained by boundary optimization, specifically including:

[0054] Based on the hierarchy and connection relationship of the skeletal view semantic tree model, select the skeletal subtrees that need to be split. These subtrees represent different components in the character model.

[0055] Perform weight analysis on the skin vertices around the selected bone subtree to determine which vertices are most affected by the bones in the subtree;

[0056] Perform preliminary component segmentation based on the skeleton subtree and skin weight information;

[0057] For gaps or holes in the initial segmentation results, gap filling is performed by interpolating the median values ​​of adjacent points.

[0058] The two-dimensional Gaussian distribution value of each point on the segmentation boundary is calculated using the two-dimensional Gaussian distribution formula;

[0059] The original vertical coordinate value of each point is replaced by the average value of the two-dimensional Gaussian distribution values ​​of all points in the neighborhood of each point on the segmentation boundary, and Gaussian filtering is performed on the segmentation boundary to eliminate jagged edges or sharp corners, thereby obtaining the subtree segmentation result of the semantic tree model;

[0060] The two-dimensional Gaussian distribution formula is:

[0061]

[0062] Where, is the two-dimensional Gaussian distribution value of each point on the segmentation boundary, is an adjustable smoothing parameter.

[0063] Preferably, the method of using the semantic tree model subtree as a genetic gene to perform non-allelic exchange, expand the solution space, and create a novel role model specifically includes:

[0064] S101: Randomly generate at least two character models as the initial population;

[0065] S102: Evaluate all individuals in the population according to predetermined evaluation criteria to obtain a fitness value of each individual, wherein the evaluation criteria include aesthetics and functionality;

[0066] S103: Select excellent individuals as parents based on fitness values;

[0067] S104: performing a non-equivalent exchange operation between parent individuals, exchanging the positions and contents of subtrees of different semantic tree models, and generating new child individuals;

[0068] S105: Perform random mutation operations on offspring individuals and introduce new semantic tree model subtrees as genetic genes;

[0069] S106: Replace the non-excellent individuals in the population with the newly generated offspring individuals;

[0070] S107: Determine whether the maximum number of iterations has been reached. If so, end the iteration and output the current population as a novel character model. If not, return to step S102.

[0071] Preferably, the combining of user selection and probabilistic graphical model as fitness judgment criteria in evolution, optimizing the solution space search path, and improving the acceptability of the results specifically include:

[0072] Users select and rate within a given range of role models based on their preferences and needs;

[0073] Based on the user's ratings of each role model, a rating probability graph model is constructed;

[0074] The scoring probability graph model replaces the predetermined evaluation criteria as the fitness judgment criterion in evolution to optimize the solution space search path.

[0075] Compared with the prior art, the present invention has the following beneficial effects:

[0076] Starting from the basic idea of ​​the semantic tree model, we expand its extensibility and describe the structural characteristics of three-dimensional characters from multiple angles, so that it can support the synthesis of more novel and diverse characters of various categories. We conduct in-depth exploration of the model segmentation technology that supports character creation, and establish a special method for the three-dimensional character as a special object. For complex three-dimensional character models from multiple sources, we can obtain more accurate and meaningful segmentation results, which can assist and guide users in conducting combined searches and generate a variety of new characters. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 This is a flow chart of the image optimization method for game character and creature model design of the present invention;

[0078] Figure 2 A flow chart of the method for classifying character design and defining components from the perspective of biological populations according to the present invention;

[0079] Figure 3 A flow chart of the method for constructing different semantic tree models for different operations according to the present invention;

[0080] Figure 4 This is a flow chart of the preprocessing method for preparing data and expressing semantic trees for multiple types of three-dimensional characters from different sources according to the present invention;

[0081] Figure 5 This is a flow chart of the segmentation boundary search method of the present invention;

[0082] Figure 6 Flowchart of the method for pre-segmenting components and obtaining subtree segmentation results of a semantic tree model through boundary optimization according to the present invention;

[0083] Figure 7 A flow chart of the method for creating a novel role model of the present invention;

[0084] Figure 8 This is a flow chart of the optimized solution space search path method of the present invention. DETAILED DESCRIPTION

[0085] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are merely examples, and those skilled in the art may conceive of other obvious variations.

[0086] Reference Figure 1 As shown, an image optimization method for game character and creature model design includes:

[0087] Classify character designs and define components from the perspective of biological populations;

[0088] Construct different semantic tree models for different operations, and obtain the acquisition methods and model construction results of various semantic tree models;

[0089] Through experiments with role models from different sources, we analyze the applicable objects and respective advantages of each role model;

[0090] Perform data preparation and semantic tree expression preprocessing for multiple types of 3D characters from different sources;

[0091] Branch view-based segmentation and labeling finds segmentation boundaries by searching and verifying the geometric features of end points and curvature under the premise of known semantic constraints;

[0092] Skeleton view-based segmentation uses the skeleton and skin information of the animated character to pre-segment the parts, and obtains the subtree segmentation results of the semantic tree model through boundary optimization;

[0093] Using semantic tree model subtrees as genetic genes, non-allelic exchange is performed to expand the solution space and create novel role models;

[0094] Combining user selection with a probabilistic graphical model as the fitness judgment criterion in evolution, the solution space search path is optimized and the acceptability of the results is improved.

[0095] Reference Figure 2 As shown in the figure, the classification and component definition of character design from the perspective of biological populations include:

[0096] The character model parts are defined as a head, a torso, limbs, and appendages, wherein the appendages include wings, a tail, and tentacle structures;

[0097] Based on the character model components, the character models are classified into two-legged upright type, two-legged head-up type, four-legged head-up type, multi-legged head-up type, legless horizontal type and four-legged horizontal type characters;

[0098] Bipedal characters have complete upright legs and feet, with the feet touching the ground as support, and obvious upper limb structures, modeled with reference to human form;

[0099] The bipedal head-raising character has slender legs, a long neck covered with feathers, and a pair of wings on the back of the torso. It is modeled based on the morphology of birds.

[0100] A quadruped head-raising character has strong front and back limbs that touch the ground for support. The torso is parallel to the ground and at a distance from the ground. The head is clearly located above the front of the torso and extends forward. The modeling is based on the morphology of quadruped mammals.

[0101] The multi-legged head-raising character is similar to a large quadruped animal, while maintaining the characteristics of multi-legged limbs. The various accessories are increased and the modeling is based on the morphology of arthropod insects.

[0102] The legless horizontal character has a simple structure, consisting of only two parts: the torso and the head. It is modeled based on the morphology of reptiles such as fish and snakes.

[0103] The design style of the four-legged horizontal character is similar to that of the multi-legged head-raising character. The outline of the torso parts is unclear, and the legs are relatively thick. The most notable feature is that the model is close to the ground and is modeled with reference to the morphology of large reptiles.

[0104] The main design style defines the design goal of the character. The number of feet is defined as the number of feet, which is easy to understand. The form gives the composition mechanism of the head and body: the upright type is defined as the body is vertical, and the head and neck are perpendicular to the center of the torso and located at the top; the horizontal type is defined as the body is horizontal, the head and neck are located in front of the body, and the body touches the ground; the head-up type is defined as the body is horizontal or slightly tilted, and the head and neck are located in front of the body.

[0105] Reference Figure 3 As shown, different semantic tree models are constructed for different operations, and the acquisition methods and model construction results of various semantic tree models are specifically as follows:

[0106] Express the obvious connection relationship between the components in a role model from the perspective of general user cognition and build a basic view semantic tree model;

[0107] Express the connection relationship between the various components in a character model from the animator's cognitive perspective, determine the skeleton information and skin information of the semantic tree model, and build a skeletal view semantic tree model;

[0108] Perform consistent segmentation on a batch of role models with the same structure and construct a branch view semantic tree model;

[0109] Expand on the basic view semantic tree model, introduce an empty component symbol to describe the possible expansion area, and build the design view semantic tree model.

[0110] Semantic trees are often used to establish a relational expression of the research object, such as in natural semantic understanding or scene image annotation. In addition to describing hierarchical associations, semantic trees are more about the connection associations between component symbols. The multi-view model is a method of representing the same object from different angles and aspects, and there is a certain relationship between multiple views, which can be transformed into each other. It is an important display technology in the field of visualization.

[0111] Reference Figure 4 As shown in the figure, the data preparation and semantic tree expression preprocessing work for multiple types of 3D characters from different sources specifically includes:

[0112] Get at least one role model from the model database;

[0113] Based on the acquisition methods of various semantic tree models, various semantic tree models of different types of 3D characters are constructed;

[0114] At least one character model in the model database is input into various semantic tree models of different types of three-dimensional characters to obtain preprocessing models of the multiple types of three-dimensional characters.

[0115] Because the models vary greatly and have different defects, it is difficult to process them using a unified method. Therefore, most of the original data needs to be manually edited and adjusted to have no topological errors and standard postures. However, many character models still cannot form topologically closed meshes due to reasons such as hair and clothing. This poses a great challenge to the segmentation algorithm and will be specially addressed in subsequent operations.

[0116] Reference Figure 5 As shown in the figure, the segmentation and labeling based on the branch view, under the premise of known semantic constraints, is verified by searching for geometric features of end points and curvature to find the segmentation boundary. Specifically, the following steps are involved:

[0117] Using morphological simulation to perform corrosion operations, the objects in the preprocessing model are reduced in size and the parts smaller than the preset threshold are eliminated.

[0118] The output residual is the end point of the preprocessing model;

[0119] Get the parameter equations of the preprocessing model;

[0120] The arc length parameter of the preprocessing model is calculated using the arc length parameter formula;

[0121] Based on the arc length parameter of the preprocessing model, the unit tangent vector of any point on the preprocessing model is calculated using the unit tangent vector formula;

[0122] Calculate the curvature of any point on the preprocessed model using the curvature formula;

[0123] Based on the curvature of each point between the end points, the segmentation boundary is found and the pre-processed model is segmented;

[0124] The parametric equation is:

[0125]

[0126] Where, are the preprocessing model parameters, are the coordinates of each point in the preprocessing model;

[0127] The arc length parameter formula is:

[0128]

[0129] Where, is the arc length parameter of the preprocessing model, is the first derivative of x with respect to t;

[0130] The normalized tangent vector formula is:

[0131]

[0132] Where, is the unit tangent vector of any point on the preprocessed model;

[0133] The curvature formula is:

[0134]

[0135] Where, is the curvature at any point on the preprocessed model, for The derivative with respect to t.

[0136] Based on a full understanding of the composition of animated characters and a multi-angle semantic tree description, we use the existing prior knowledge of the characters and combine it with geometric methods to design an automatic segmentation and labeling method for character models from different sources. It requires very little human interaction and is adaptable to a variety of model styles.

[0137] Reference Figure 6 As shown in the figure, the skeletal view-based segmentation uses the skeleton and skin information of the animated character to pre-segment the parts. The semantic tree model subtree segmentation results are obtained through boundary optimization. Specifically, the results include:

[0138] Based on the hierarchy and connection relationship of the skeletal view semantic tree model, select the skeletal subtrees that need to be split. These subtrees represent different components in the character model.

[0139] Perform weight analysis on the skin vertices around the selected bone subtree to determine which vertices are most affected by the bones in the subtree;

[0140] Perform preliminary component segmentation based on the skeleton subtree and skin weight information;

[0141] For gaps or holes in the initial segmentation results, gap filling is performed by interpolating the median values ​​of adjacent points.

[0142] The two-dimensional Gaussian distribution value of each point on the segmentation boundary is calculated using the two-dimensional Gaussian distribution formula;

[0143] The original vertical coordinate value of each point is replaced by the average value of the two-dimensional Gaussian distribution values ​​of all points in the neighborhood of each point on the segmentation boundary, and Gaussian filtering is performed on the segmentation boundary to eliminate jagged edges or sharp corners, thereby obtaining the subtree segmentation result of the semantic tree model;

[0144] The two-dimensional Gaussian distribution formula is:

[0145]

[0146] Where, is the two-dimensional Gaussian distribution value of each point on the segmentation boundary, is an adjustable smoothing parameter.

[0147] The Gaussian function is a single-valued function. This means that the Gaussian filter replaces the pixel value at a point with the weighted mean of its neighborhood. The weight of each neighboring pixel increases and decreases monotonically with its distance from the center. This property is important because edges are local image features. If smoothing operations still have a significant effect on pixels far from the center of the operator, the smoothing operation will distort the image.

[0148] Reference Figure 7 As shown in the figure, the semantic tree model subtree is used as a genetic gene to perform non-allelic exchange, expand the solution space, and create a novel role model. Specifically, the following are included:

[0149] S101: Randomly generate at least two character models as the initial population;

[0150] S102: Evaluate all individuals in the population according to predetermined evaluation criteria to obtain a fitness value of each individual, wherein the evaluation criteria include aesthetics and functionality;

[0151] S103: Select excellent individuals as parents based on fitness values;

[0152] S104: performing a non-equivalent exchange operation between parent individuals, exchanging the positions and contents of subtrees of different semantic tree models, and generating new child individuals;

[0153] S105: Perform random mutation operations on offspring individuals and introduce new semantic tree model subtrees as genetic genes;

[0154] S106: Replace the non-excellent individuals in the population with the newly generated offspring individuals;

[0155] S107: Determine whether the maximum number of iterations has been reached. If so, end the iteration and output the current population as a novel character model. If not, return to step S102.

[0156] In genetic algorithms, genes are the basic units of genetic information, which are transmitted and changed through operations such as crossover and mutation. In character model generation based on semantic trees, we can select subtrees in the semantic tree as genetic genes. These subtrees can contain a specific part of the character (such as arms, legs) or a set of related attributes (such as skin color and texture).

[0157] Reference Figure 8 As shown in the figure, combining user selection with the probabilistic graph model as the fitness judgment standard in evolution, optimizing the solution space search path, and improving the acceptability of the results specifically include:

[0158] Users select and rate within a given range of role models based on their preferences and needs;

[0159] Based on the user's ratings of each role model, a rating probability graph model is constructed;

[0160] The scoring probability graph model replaces the predetermined evaluation criteria as the fitness judgment criterion in evolution to optimize the solution space search path.

[0161] Combining user selection with probabilistic graphical models as fitness judgment criteria in evolution and improving the acceptability of results by optimizing the search path in the solution space requires full consideration of user needs and preferences, rational use of the probabilistic dependencies of probabilistic graphical models, and effective implementation of the selection and operation mechanisms in evolutionary algorithms.

[0162] Furthermore, the present solution also proposes a computer-readable storage medium on which a computer-readable program is stored. When the computer-readable program is called, the above-mentioned image optimization method for game character and biological model design is executed.

[0163] It is understandable that the storage medium may be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; an optical medium, such as a DVD; or a semiconductor medium, such as a solid state disk (SSD).

[0164] To sum up, the advantages of the present invention are: starting from the basic idea of ​​the semantic tree model, expanding its extensibility and describing the structural characteristics of the three-dimensional character from multiple angles, so that it can support the synthesis of more novel and diverse characters of various categories, conducting in-depth exploration of the model segmentation technology that supports character creation, and establishing a special method for the special object of three-dimensional characters. For complex three-dimensional character models from multiple sources, more accurate and meaningful segmentation results can be obtained, which can assist and guide users to conduct combined searches and generate diverse new characters.

[0165] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate 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 by the present invention is defined by the appended claims and their equivalents.

Claims

1. An image optimization method for game character and biological model design, characterized in that: include: Classify and define components of character design from the perspective of biological populations; Construct different semantic tree models for different operations, and obtain the acquisition methods and model construction results of various semantic tree models; Through experiments with role models from different sources, analyze the applicable objects of each role model and their respective advantages; Perform data preparation and semantic tree expression preprocessing for multiple types of 3D characters from different sources; Based on the segmentation and labeling of branch views, the segmentation boundary is found by searching and verifying the geometric features of end points and curvature under the premise of known semantic constraints; The segmentation based on the skeleton view uses the skeleton and skin information of the animation character to pre-segment the parts, and obtains the subtree segmentation result of the semantic tree model through boundary optimization; Using semantic tree model subtrees as genetic genes, non-allelic exchange is performed to expand the solution space and create novel role models; Combining user selection with the probabilistic graph model as the fitness judgment criterion in evolution, optimizing the solution space search path and improving the acceptability of the results; The segmentation and labeling based on the branch view, under the premise of known semantic constraints, finds the segmentation boundary by searching and verifying the geometric features of the end points and curvatures, specifically including: Using morphological simulation to perform corrosion operations, the objects in the preprocessing model are reduced and the parts smaller than a preset threshold are eliminated; The output residual is the end point of the preprocessing model; Obtain parameter equations of the preprocessing model; The arc length parameter of the preprocessing model is calculated using the arc length parameter formula; Based on the arc length parameter of the preprocessing model, the unit tangent vector of any point on the preprocessing model is calculated using the unit tangent vector formula; The curvature at any point on the preprocessed model is calculated using the curvature formula; Based on the curvature of each point between the end points, the segmentation boundary is found and the preprocessed model is segmented; The parametric equation is: ， In the formula, is the preprocessing model parameter, is the coordinates of each point in the preprocessing model; The arc length parameter formula is: ， In the formula, is the arc length parameter of the preprocessing model, is the first-order derivative of x with respect to t; The normalized tangent vector formula is: ， In the formula, is the unit tangent vector of any point on the preprocessed model; The curvature formula is: ， In the formula, is the curvature at any point on the preprocessed model, for The derivative of t.

2. The image optimization method for game character and biological model design according to claim 1, characterized in that: The classification and component definition of character design from the perspective of biological populations specifically include: The character model parts are defined as a head, a torso, limbs, and appendages, wherein the appendages include wings, a tail, and tentacle structures; Based on the character model components, the character models are classified into two-legged upright type, two-legged head-up type, four-legged head-up type, multi-legged head-up type, footless horizontal type and four-legged horizontal type characters; Bipedal characters have complete upright legs and feet, with the feet touching the ground as support parts, and obvious upper limb arm structures, modeled with reference to human form; The two-legged head-up character has two slender legs, a long neck covered with feathers, and a pair of wings on the back of the torso. It is modeled based on the shape of birds. The four-legged head-raising character has strong front and back limbs, which are in contact with the ground as support parts. The torso is parallel to the ground and a distance away from the ground. There is an obvious head located above the front part of the torso and sticking out in front of the torso. It is modeled with reference to the morphology of four-legged mammals; The multi-legged head-raising character is close to large quadrupeds, while maintaining the characteristics of multi-legged limbs, with more accessories and modeling based on the morphology of arthropod insects; The legless horizontal character has a simple structure, consisting of only two parts: the torso and the head. It is modeled based on the morphology of fish and snake reptiles. The design style of the four-legged horizontal character is similar to that of the multi-legged head-raising character. The outline of the torso part is not clear, and the legs are relatively thick. The most notable feature is that the model is close to the ground and is modeled with reference to the morphology of large reptiles.

3. The image optimization method for game character and biological model design according to claim 2, characterized in that: The method of constructing different semantic tree models for different operations and obtaining various semantic tree models and the model construction results specifically include: Express the obvious connection relationship between the components in a role model from the perspective of general user cognition and build a basic view semantic tree model; Express the connection relationship between the components in a character model from the animator's cognitive perspective, determine the skeleton information and skinning information of the semantic tree model, and build a skeleton view semantic tree model; Perform consistent segmentation on a batch of role models with the same structure and build a branch view semantic tree model; Expand on the basic view semantic tree model, introduce an empty component symbol to describe the possible expansion area, and build the design view semantic tree model.

4. The image optimization method for game character and biological model design according to claim 3, characterized in that: The data preparation and semantic tree expression preprocessing work for multiple types of three-dimensional characters from different sources specifically includes: Obtain at least one role model from a model database; Based on the acquisition methods of various semantic tree models, various semantic tree models of different types of three-dimensional characters are constructed; At least one role model in the model database is input into various semantic tree models of different types of three-dimensional roles to obtain preprocessing models of the multiple types of three-dimensional roles.

5. The image optimization method for game character and biological model design according to claim 4, characterized in that: The segmentation based on the skeleton view uses the skeleton and skin information of the animation character to pre-segment the parts, and obtains the semantic tree model subtree segmentation results through boundary optimization, which specifically includes: Based on the hierarchy and connection relationship of the skeleton view semantic tree model, select the skeleton subtrees that need to be split. These subtrees represent different parts in the character model. Perform a weight analysis on the skin vertices around the selected bone subtree to determine which vertices are most affected by the bones in that subtree; Perform preliminary component segmentation based on the skeleton subtree and skin weight information; For the gaps or holes in the preliminary segmentation results, the gaps are filled by interpolating the middle values ​​of adjacent points. The two-dimensional Gaussian distribution value of each point on the segmentation boundary is calculated using the two-dimensional Gaussian distribution formula; The original vertical coordinate value of each point is replaced by the average value of the two-dimensional Gaussian distribution values ​​of all points in the neighborhood of each point on the segmentation boundary, and Gaussian filtering is performed on the segmentation boundary to eliminate jagged edges or sharp corners, thereby obtaining the subtree segmentation result of the semantic tree model; The two-dimensional Gaussian distribution formula is: ， In the formula, is the two-dimensional Gaussian distribution value of each point on the segmentation boundary, is an adjustable smoothing parameter.

6. The image optimization method for game character and biological model design according to claim 5, characterized in that: The method of using the semantic tree model subtree as a genetic gene, performing non-allelic exchange, expanding the solution space, and creating a novel role model specifically includes: S101: Randomly generate at least two role models as an initial population; S102: Evaluate all individuals in the population according to a predetermined evaluation standard to obtain a fitness value of each individual, wherein the evaluation standard includes aesthetics and functionality; S103: Select excellent individuals as parents according to fitness values; S104: performing a non-equivalent exchange operation between parent individuals, exchanging the positions and contents of subtrees of different semantic tree models, and generating new child individuals; S105: Perform random mutation operations on offspring individuals and introduce new semantic tree model subtrees as genetic genes; S106: Replace the non-excellent individuals in the population with the newly generated offspring individuals; S107: Determine whether the maximum number of iterations has been reached. If so, terminate the iteration and output the current population as a novel character model. If not, return to step S102.

7. The image optimization method for game character and biological model design according to claim 6, characterized in that: The method of combining user selection with the probabilistic graph model as a fitness judgment standard in evolution, optimizing the solution space search path, and improving the acceptability of the result specifically includes: Users select and rate within a given range of role models based on their preferences and needs; Based on the user's ratings of each role model, a rating probability graph model is constructed; The scoring probability graph model replaces the predetermined evaluation criteria as the fitness judgment criterion in evolution to optimize the solution space search path.

Images