A wide range of algorithms have been developed over time to approach the non-trivial problem of enlarging bitmap images, each providing a different balance between aesthetic quality (the preservation of smooth gradients versus sharp edges) and processing efficiency. Aiming for aesthetics over speed, artificial neural networks have recently been used to precisely map the X and Y coordinates of an image's pixels to corresponding red, green and blue (RGB) colour channels. This time-consuming process converts a bitmap image into a complex mathematical function, or Scalable Function Graphic (SFG). SFG files are significantly larger than the original bitmap, however they act as a resolution independent representation of an image, capable of outputting colour vectors at X and Y coordinates that were not defined in the original (low-resolution) bitmap. By this process of generalisation, the mathematically modeled image can be rendered at any scale, with minimal impact on aesthetic quality.

The aim of this research is to build on recent SFG technology, by exploring how the efficiency of conversion and storage can be improved (using a database of pre-trained neural networks that can be concurrently fine-tuned), while preserving or improving aesthetic quality.

​

< Currently only a research review/proposal is available for this project. The actual thesis will be made available in September/October 2016