Parameter optimization in digital media and 3D animation design and its visual performance research

Vertices are the basic geometric elements that form a 3D model, and their number and distribution directly determine the level of detail of the model. A larger number of vertices can accurately describe complex shapes, but it also increases the storage and computation burden. When constructing organic models, such as human body models, a reasonable increase in vertex density is needed in key areas such as faces and hands to capture delicate morphological changes.

Polygonal faces are connected by vertices, and their topology affects the surface quality and deformation effects of the model. A good topology should avoid narrow triangular faces and non-fluid geometries to ensure the smoothness and stability of the model during deformation animation. For example, when creating a character model, a quadrilateral-based polygonal surface layout is usually adopted and certain topological rules are followed, so that the model can deform naturally and smoothly when performing expression animation and body movement animation.

Subdivision is the process of increasing the level of detail in a model by adding more vertices and faces to the model, and is often used to create highly accurate models. For example, when creating movie-level character models or fine industrial work models, the base model will be subdivided several times to make the model surface smoother and richer in details. In some application scenarios with high real-time requirements, the model needs to be simplified in order to ensure the smooth operation of the system. The goal of simplification is to reduce the number of vertices and faces while maintaining the appearance characteristics of the model as much as possible.

The color parameter contains the object’s base color, highlight color, and ambient color. The base color determines the main color tone of the object under diffuse illumination, the highlight color affects the highlight color and glossiness of the object under strong light irradiation, and the ambient light color simulates the effect of the ambient light around the object on its color. The reasonable combination of these color parameters can make the object in the light environment to present rich color changes and layering.

Texture is an important means to provide details and realism to the surface of the object, and the 2D texture image is mapped to the surface of the 3D model through texture coordinates. The texture coordinates should be set to ensure that the texture is correctly laid on the model surface to avoid stretching, distortion, and other phenomena. Meanwhile, the resolution and quality of the texture also have an impact on the visual effect. Higher resolution textures can provide clearer details, but they also take up more memory resources.

The roughness parameter controls the microscopic roughness of an object’s surface. The higher the roughness, the more uniformly the light is scattered over the surface, and the object has a matte effect. The Warp Blocking parameter is used to describe the metallic properties of a physical surface. The higher the metallicity, the closer the object’s reflection and refraction behavior of light is to real metal. The refractive index parameter is responsible for determining how much light is refracted as it passes through the surface of an object. Different materials have different refractive indexes, and the correct setting of the refractive index can simulate the transparent and refractive effects of an object.

Common types of light sources include point light sources, parallel light sources, spot lights, etc. Different types of light sources have different light characteristics and distribution laws. The position of the light source determines the incident direction and angle of the light, which has an important impact on the sensitive relationship and shadow effects of the object.

Light intensity affects the overall brightness of the scene, while the color temperature determines the warm and cool tones of the light, and reasonable adjustment of light intensity and color temperature can create different atmospheres and emotional tones.

Key frames are the key time points of the animation, which record the position, rotation, scaling, and other state information of the object. Interpolation algorithms are used to generate smooth over-animation between key frames, such as linear interpolation, Bessel interpolation, etc. Different interpolation algorithms affect the smoothness of the animation and the naturalness of the motion trajectory.

By editing the animation curve, you can accurately control the motion speed, acceleration, and other dynamic characteristics of the object, making the animation more vivid and expressive.

Genetic coding chooses the real number coding method. From the elaboration in the previous section, it can be seen that the 3D animation design parameter optimization needs to solve the objective of four real values, corresponding to the model parameters, material parameters, lighting parameters, animation parameters, through the genetic algorithm will be optimized for the four parameters, it can be created through the 3D engine to create a three-dimensional animation design model and set up the correct parameters and the actual object size mapping one by one.

The encoding of the ist individual of the algorithm population is defined as: X_i = {w,l,h,b}, where w is a model parameter, l is a material parameter, h is a lighting parameter, and b is an animation parameter. In 3D animation design animation production for calibration of the vector is 8, 8 calibration vector and individual information asked the relationship between the following equation: (1) { V1={ −w2,l2,h−b }V5={ w2,−l2,h−b }V2={ w2,l2,h−b }V6={ −w2,−l2,h−b }V3={ −w2,l2,−b }V7={ w2,−l2,−b }V4={ w2,l2,−b }V8={ −w2,−l2,−b }

Thus after the transformation of the solution by Eq. (1), the 8 vectors obtained are the search space that can be handled by the genetic algorithm, which can perform operations such as crossover, mutation, etc., on these 8 vectors and match them with the calibrated vectors with the adapted values.

The specific steps are:

1) Generate P_i ∈ U(0,1)randomly, when P_i > P_i, then this chromosome does not satisfy the crossover probability, no crossover operation is performed, and jump to step 3) to end the crossover operation.

The specific steps are:

1) Randomly generate P_i ∈ U(0,1), when P_i > P_m, then this chromosome does not meet the mutation conditions, do not carry out the mutation operation operation, jump to step 5), end the mutation operation.

Genetic algorithms simulate the process of reproduction of realized populations through selection operations. Selection means meritocracy, so the selection operation is a filtering method based on the results of the calculation of fitness values, and the probability P_i of each individual being selected is calculated as follows: (2) Pi=fitness(i)∑j=1nfitness(j)

After unitizing the 8 vectors V_i of the calibration and the 8 vectors V_xi of the possible solutions, the conditions for comparison at the same scale are in place.

The next step is to calculate the vector difference of the 8 pairs of vectors and to find the value of the length of the vector difference and finally the value of the adaptation, which is calculated as follows: (3) DOi=(Vi⋅x−Vxi⋅x)2+(Vi⋅y−Vxi⋅y)2+(Vi⋅z−Vxi⋅z)2 (4) fitness=∑i=18[ (1−∑i=18DOi8)×(1−DOi)8 ]

Eq. (4) is a function that calculates the adaptation value, where the physical meaning of DO is the value of the deviation of the unit vector of object dimensions represented by the possible solutions from the unit vector of actual object dimensions. The first half of Eq. (4) represents the accumulation of eight DO deviation values as a combined deviation. In the second half of the equation, the smaller DO represents the smaller deviation of the two unit vectors, and the closer the object size represented by the possible solution is to the actual object size.

In this paper, we use Maya 3D animation production software to design a 3D scene containing multiple objects and complex lighting environments, including models with different geometries, such as character models, mechanical parts models, natural object models (number, stone, etc.), as well as objects with different materials (metal, plastic, wood, fabric, etc.). Several light sources, including point light sources, parallel light sources, spot light sources, and ambient light, are set in the scene to simulate different lighting conditions, such as indoor light, outdoor sunlight, and stage lighting.

In this paper, model, material, lighting and animation parameters such as the number of fixed points of the model, the number of polygonal surfaces subdivided into levels, the material roughness, metallicity, refractive index, the intensity of the light, the color temperature, the position of the light source, as well as the animation of the keyframe interval, interpolation algorithms, etc., are selected as variables, and a number of different levels are set for each parameter to generate a number of groups of experiments through the design of full factorial experiments or the design of part of the factorial experiments method. Parameter combinations are generated using either full factorial experimental design or partial factorial experimental design.

Subjective evaluation: Invite 30 professional 3D animation designers, digital media experts and general audience to form an evaluation team to subjectively evaluate the visual effects of 3D animation before and after parameter optimization. The evaluation indexes include the model’s detail clarity, realism, material texture, naturalness of lighting effect, smoothness and vividness of animation, etc. A scoring standard of 1-5 points is adopted, in which 1 point means the worst and 5 points means the best.

Objective assessment is quantitatively calculated using the visual performance evaluation method proposed in this paper. In order to further evaluate the impact of parameter optimization on the quality of stereo images generated by 3D animation design, this paper adopts the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) of the image as the most objective evaluation indexes of the image quality, and in order to facilitate the comparison of the final results, the study enlarges the SSIM 100 times In order to facilitate the comparison of the final results, the SSIM is expanded by 100 times, and the calculation formula is as follows: (17) PSNR=10lg(Max12/MSE)

Where, Max₁ denotes the maximum pixel value of the original image and MSE denotes the mean square error between the original image and the generated image after parameter optimization.

Where, μ_x and μ_y denote the mean values of the original image x and the parameter optimized image y respectively. σ_x and σ_y denote the x,y standard deviation, respectively. σ_xy denotes the covariance of x,y. c₁,c₂ is a constant to avoid a denominator of 0.

The performance of the encoding module and selection module is analyzed at the same time, and the comparison object is the genetic algorithm using binary encoding and general selection module for comparison, to ensure that the parameters of other modules except the encoding module and selection module are the same, the number of iterations is the same, and the optimization search is carried out on the same object by using the application of the improved genetic algorithm in this paper and the genetic algorithm using the general binary encoding and selection operation, respectively. The results are shown in Figure 1, where the horizontal coordinate is the number of iterations and the vertical coordinate is the optimal value obtained. From the figure, it can be seen that the improved real number coding and selection operation converge to the optimal value of 20 in 34 iterations, while the algorithm using binary coding and general selection operation needs to be iterated to 47 generations to converge to the optimal value, but the optimal value is far from 20.Therefore, the selection operation using the improved real number encoding in this paper converges with the least number of iterations than the genetic algorithm using binary encoding and ordinary selection operation, converges better and can obtain better optimal values.

Figure 1.

The optimization process of both methods

In this subsection, the crossover and mutation modules of the improved genetic of this paper are unified for comparative analysis, and the crossover and mutation modules in the genetic algorithm used for the comparison are the ordinary crossover and mutation operations before the improvement, which are not adaptive. The functions utilized are identical to the optimization objects employed in the previous section, and the outcomes are represented in Fig. 2. From the figure, it can be seen that the number of iterations required for convergence using the crossover and mutation method with improved genetic algorithm in this paper is about 61, while the number of iterations required for convergence using the pre-improvement one needs 114, and the convergence to the optimal value is much worse than the improved one. Therefore, the improved crossover and mutation modules in this paper require fewer iterations to converge compared to the ordinary genetic algorithm with crossover and mutation operations before improvement, and the final converged value is less different from the optimal value, and the convergence will be better.

Figure 2.

The Optimization pross before and after the improvement

The performance analysis of the simulated annealing of the improved genetic algorithm is carried out to compare the parameter optimization process before and after adding the simulated annealing module in the same parameter optimization process, and the results are shown in Fig. 3. The horizontal coordinate in the figure also indicates the number of iterations, and the vertical coordinate indicates the value of the objective function. From the figure, it can be seen that the genetic algorithm with simulated annealing module can quickly converge to the optimal objective function value in the 0-5th iteration, while the genetic algorithm without simulated annealing module slows down the convergence speed when the objective function value converges to 18.73 until the 121st iteration to complete the convergence and converge to the optimal objective function value. It can be concluded that the improved genetic algorithm using the simulated annealing module added in this paper will get the optimal value faster than the one before the improvement and can converge to the optimal value accurately.

Figure 3.

The optimization process before and after improvement GA

The subjective assessment results before and after the optimization of 3D animation design parameters are shown in Table 1. From the table, it can be seen that before the parameter optimization, the 30 experts’ scores for the visual performance of the 3D animation design are low, with the highest score being 3, the lowest score being 5, and the mean value of the scores being 1.73. After the parameter optimization using the improved genetic algorithm, the scores of the 30 experts’ scores for the visual performance of the 3D animation design are obviously improved, with the lowest score being 4, the highest score being 5, and the mean value of the scores being 4.47.The subjective assessment results show that after optimizing the key parameters in the 3D animation design using improved genetic algorithm, this paper significantly improves the visual performance effect of the 3D animation design model or image, and enhances the subjective visual attraction of the design results to the audience.