Image Processing: Applying Styles to Images

Existing Implementation

​

Since “A Neural Algorithm of Artistic Style” was published a few years ago, there exist many implementations of the code. Most of these implementations are done in python do to the libraries and functions provided. Since none of our group members were very familiar with python, and we wanted to write our own implementation, we decided to write our code in MATLAB. However, we found it very helpful to look at other implementations and learn from how they implemented certain things. We decided to download and run an existing implementation (https://github.com/anishathalye/neural-style.git). We also used this existing implementation to analyze what happens when changing the parameters. The images below were created using a content image of a ballerina and a style image of a sketch. One can see how the result with fewer iterations looks more like the original white noise image, while the higher iteration versions begin to look like a combination of the content and style images. 

​

                   50 Iterations                                            500 Iterations                                    1000 Iterations 

 

 

 

 

 

 

 

 

 

 

 

 

 

​

​

Combining a content photo of a cat and a style painting by Monet yielded an arguably more aesthetically pleasing result. The content, style, and output images are shown below. 

​

                  Content Image                                      Output Image                                         Style Image

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

​

Image Sources

​

Monet Painting: Public Domain

Ballerina: https://40.media.tumblr.com/781201f62d7caa198b15aa9b87e6bda5/tumblr_o0t4kdlBfP1qmz328o4_500.jpg

Sketch: http://images.metmuseum.org/CRDImages/ap/web-highlight/50.130.140m.jpg

Cat Image: http://superepus.com/search/cat/snow/