Evolving image processing operations for an evolvable hardware environment

This paper describes the application of genetic algorithms to evolve new spatial masks for non-linear image processing operations, which are ultimately to be implemented on evolvable hardware. The development environment was custom-built to allow full control over the evolution process and enable the importance of the evolution strategy (including the representation scheme, parameters and fitness function) to be investigated and understood. Results of applying the evolved mask to threshold real-world images are provided and are shown to be an improvement on conventional image processing operations. The envisaged infrastructure for the evolvable hardware is also considered, and the implementation of the image processing operations discussed.