Abstract

Artistic image creation, facilitated by Neural Style Transfer (NST), represents a dynamic frontier in the realm of convolutional neural networks (CNNs). This project is dedicated to the exploration and implementation of NST, with a focus on evaluating outcomes through a dual lens of qualitative and quantitative analysis. The efficacy of the trained neural network is scrutinized by assessing resultant image styles and measuring performance through visual
analysis and loss calculations, given the absence of a direct accuracy metric for this intricate task. Exploring the complexities of picture-style transmission, this study examines six different approaches in the context of neural networks. model yields the highest Structural Similarity Index (SSIM) score out of all of them, also indicating an image that is closer to the content. which used CNN with five relu layers in the style and content images. These performers are distinguished for their exceptional versatility in capturing and translating a wide range of artistic styles. This thorough methodology offers insightful information about the efficiency and possible uses of neural style transfer methods.


KEYWORDS: dataset of images, Convolutional neural network(CNN)

Objective

The objective of neural style transfer using Convolutional Neural Networks (CNNs) is to combine the content of one image with the style of another image to generate a new image that preserves the semantic content of the original content image while adopting the visual style of the style reference image. Neural Style Transfer (NST) using CNNs can be applied to both images and video.