Title: Rapid generation method of anime character images based on conditional depth convolutional network

Authors: Ran Zhang

Addresses: School of Fine Arts and Design, Hefei Normal University, Hefei, Anhui, China

Abstract: In order to solve the problems of lack of image details and high generation time cost in existing image generation methods, this paper proposes a fast generation method for anime character images based on conditional deep convolutional networks. Firstly, a conditional deep convolutional network model is constructed, and image features are extracted and fused using an automatic encoder. Then, the nearest neighbour interpolation algorithm is used to process the image, adjusting the weight and bias of the convolutional kernel. Finally, the Tanh activation function and LeakyReLU function are used to activate the internal recognition module of the network for discriminant output, so as to realise the generation of cartoon characters. The experiment shows that the proposed method generates anime character images that are most similar to the original image, and the final generation time of 500 anime character images is 54 seconds, which has good image generation efficiency and effectiveness.

Keywords: conditional deep convolutional network; anime character images; image generation; image features; activation function.

DOI: 10.1504/IJCAT.2024.144669

International Journal of Computer Applications in Technology, 2024 Vol.75 No.1, pp.1 - 9

Received: 17 May 2023
Accepted: 31 Jan 2024
Published online: 26 Feb 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article