Automated Photomontage Generation with Neural Style Transfer

Abstract

Applying Photoshop to an image to extract it’s style and generate a new image with different text is a tough task and everybody is not an expert to pull it off. Therefore, this report illustrates the practical approach of copying the style of an image and applying it to the content of another image by the method of neural style transfer. This method is generally used for applying artistic or abstract filter to the content image, till this date no one has tried to reciprocate this method for text images and hence we will use it for combining the text as content of one image to style of another image. This novel idea was implemented using Models like VGG-19 and RESNET which are already trained on Imagenet dataset, is demonstrated in the report via transfer learning. Future work includes reducing the style and content loss by creating new model just for extracting style and making a GUI Framework for user ease.