“Three-Dimension Animation Character Design Based on Probability Genetic Algorithm”, 2024-07-26 (; similar):
The 3-Dimension (3D) animation design typically relies on front and back information, which is a critical component in many downstream image processing tasks such as activity recognition and motion tracking. However, there is currently a lack of automated approaches for generating anime characters in the 2D design, and 3D design is performed efficiently but computationally costly, with many similar characters.
This paper proposes a Probability Genetic Algorithm (PGA) method that performs efficiently by detecting the similarity of another animation character design in 3D animation. The PGA automates a time-consuming process, allowing rapid generation of diverse character designs and helping to explore a wider set of design-specific parameters and constraints to be incorporated into the algorithm.
Initial data were acquired from the Danbooru dataset and pre-processed using data augmentation techniques for rotation and flipping to find various directions of the animation character design. The proposed PGA method is evaluated using the Danbooru data, achieving a higher accuracy of 91.25%.
The existing techniques, Mask Region Convolutional Neural Network (RCNN) and Deep Neural Network (DNN), were also evaluated against the proposed method.
[Keywords: deep neural network, data augmentation, 3-dimensional character design, mask region convolutional neural network and probability genetic algorithm]