AI-Driven Image Transformation: Adaptive Approach Optimization of Neural Style Transfer

Neural Style Transfer (NST) is a deep learning technique that takes two images, a content image and a style image and blends them, the output images are the mixture of content and style in the form of artistic style. Style images could be painting, oil painting, texture art etc. This study showcases an enhanced approach to NST by leveraging perceptual similarity metrics like SSIM and LPIPS to dynamically modify the ratio of style to content, the work involves VGG 19, a pre-trained model for style transfer. Compared to traditional methods that rely on fixed ratios, this dynamic system enhances the quality of stylized images. The efficacy of this method is evaluated with the Structural Similarity Index Measure (SSIM), Peak Signal-to-Noise Ratio (PSNR), texture similarity, Learned Perceptual Image Patch Similarity (LPIPS), as well as Content and Style Loss. By using the Dynamic approach, the overall visual quality improves because it provides better style integration and content preservation. Experimental results show that the dynamic SSIM score reaches up to 0.80 and PSNR up to 26.00, which is better than fixed ratio approaches. Additionally, this method achieves higher texture similarity, demonstrating better blending of artistic styles. A major advantage of this approach is that it reduces the need for manual tuning, making the workflow more automated, time-saving, and consistent. Its potential applications extend to digital art, cultural preservation, education, and augmented reality.