The aim of the project is to generate an image in the style of the image by a well-known artist. The experiment will use artificial neural networks to transfer the style of one image onto another. In Computer Vision context: capturing the content invariant that is the style of an image and applying it on the content of another image.
Initially captures the tensors that we need from the content and style image and then we pass the input image which will initially be an image with noise and our algorithm will try to minimize the loss between the input and content image and that between input and style image thus capturing the essence of both the images into one. The traditional method of style transfer generated image has an artistic effect that is the model successfully capture the style of the image but does not preserve the structural content of the image. The proposed method uses a segmented version of images to faithfully transfer the style to semantic similar content. Also, a regularizer term modified in loss function that helps in avoiding style spill over and have photographic results.
Electrical Engineering (MS)
Department, Program, or Center
Electrical Engineering (KGCOE)
Kyada, Krushal, "Structure Preserving regularizer for Neural Style Transfer" (2019). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus