Automated Photomontage Generation with Neural Style Transfer

Applying relevant processing methods to an image to extract its style and generate a new image with target text is a complicated task and manually doing it is time-consuming. This paper illustrates the practical approach of extracting the style of an image and applying it to the content of another image by the method of neural style transfer (NST) in an automated way. This method is generally used for applying an artistic or abstract filter to the content image. Applying relevant processing methods to an image to extract its style and generate a new image with target text is a complicated task and manually doing it is time-consuming. This paper illustrates the practical approach of extracting the style of an image and applying it to the content of another image by the method of neural style transfer (NST) in an automated way. This method is generally used for applying an artistic or abstract filter to the content image. The proposed work reciprocates the NST method for text images and is used for combining the text as content of one image to the style of another image. The proposed approach is supplemented with experimentation carried out using VGG-19 pre-trained over the ImageNet dataset. The implementation incorporates the transfer learning approach for the neural style transfer (NST) method. Content and style loss have been also analyzed for some sample styles and contents.