Embedded vision is about making machines see. It is about extracting information from visual inputs. “Embedded” calls for low power consumption, ruling out standard GPUs. “Vision” calls for very complex algorithms demanding 10B ops/s or more, ruling out standard CPUs. The three P’s (power, performance and programmability) call for the design of very specialized processors, tailored to the application. This white paper discusses how ASIPs (Application Specific Instruction-Set Processors) can help meet the three P’s and introduce a simple and efficient ASIP design methodology.