Confidentiality: 
Not Confidential
Firstname: 
Bernard
Lastname: 
Llanos
Faculty: 
David Mould
Term: 
Fall
Year: 
2016
Honours Project Title: 
Image representation by region-aware abstraction and stippling
Abstract: 
An algorithm for producing stipple art from digital images, recently developed by Hua Li and David Mould, conveys image structure, detail, and tone, using only black dots as rendering primitives. Compared to image stylization methods based on colour smoothing, stippling retains sharpness and contrast, but poorly represents flat textures, or extremely light or dark regions. I attempt to combine the strengths of these complementary stylization paradigms. To add variety and appropriate emphasis to the result, I segment the image into superpixels using Simple Linear Iterative Clustering, and apply stipples only to superpixels with higher salience. To soften transitions between stippled regions and their surroundings, I both modify Li and Mould's stippling method, and develop my own simple smoothing method. In the output stylizations, background colours and foreground stipples blend into each other. I compute a variety of quantities as proxies for salience, including residuals from edge-preserving smoothing, and salience estimates from dedicated salience detection algorithms. Accurate salience measurements lead to pleasing stylizations, but no method is highly accurate on all images. I recommend further study of salience in the context of combined stippling and smoothing stylization, along with deeper consideration of alternatives to my stylization framework.
Cover Image: 
Upload description: 
All project content (report, code, and images) is contained in COMP4905A_Fall2016_bernardllanos_100793648.zip