Effectiveness of Deep Convolutional Neural Network-Based Style Transfer Technique for Animal Image Datasets
Bella Dwi Mardiana, Darlis Herumurti · 2024
Image processing and style transfer techniques have evolved rapidly with advances in deep learning, especially in Convolutional Neural Network (CNN). This study explores the effectiveness of Deep Convolutional Neural Network (DCNN)-based style transfer techniques on animal image datasets. The popular DCNN architectures VGG16 and VGG19 were investigated and compared for their performance in transferring artistic styles to animal images. This study used three types of animal images as content images: wolf animals, dogs, and cats. This process includes image preprocessing, model architecture, and model optimization using an ascending gradient algorithm. Evaluation is performed quantitatively using Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) metrics to measure the similarity of the output image with the content and style images. Results showed performance variations between VGG16 and VGG19. For wolf and dog images, VGG19 showed higher PSNR and SSIM values. However, for cat images, VGG16 produced better PSNR and SSIM values. This difference indicates that the effectiveness of the model may vary depending on the type of animal image being processed. This research provides new insights into the effectiveness of DCNN architectures in performing style transfer on animal images and may contribute to the development of future digital art and creative design applications.