Visual style conversion strategy for visual media based on MGADNN algorithm

Article type: Research Article

Authors: Qi, Ma

Affiliations: School of Advertising, Communication University of China, Beijing, China | E-mail: [email protected]

Correspondence: [*] Corresponding author: School of Advertising, Communication University of China, Beijing, China. E-mail: [email protected].

Abstract: An improved genetic algorithm is proposed to optimize the deep neural network algorithm for visual style conversion in visual media. It consists of two parts: optimizing the deep neural network algorithm design and designing a video style conversion model. The genetic algorithm selection strategy is enhanced to optimize the neural network structure. A non-recursive neural network is used to handle temporal inconsistency in a single frame. Experimental results on the Heart dataset show that the accuracy of the optimized deep neural network algorithm is 0.8913, outperforming other algorithms like the generative adversarial dual neural network (0.8696), ant colony optimization (0.8651), active network (0.8536), genetic algorithm (0.8566), and particle swarm algorithm (0.8558). Moreover, the optimized algorithm achieves high temporal stability and running speed in single and multi-style conversion networks. In conclusion, the proposed strategy using improved genetic algorithms to optimize deep neural network algorithms for visual style conversion offers effective solutions with high application value in terms of accuracy, temporal stability, and running speed.

Keywords: Genetic algorithm; deep learning; visual media; style conversion

DOI: 10.3233/JCM-247194

Journal: Journal of Computational Methods in Sciences and Engineering, vol. 24, no. 3, pp. 1571-1584, 2024