Image Signal Processor parameter tuning with surrogate-assisted Particle Swarm Optimization

Evolutionary algorithms (EA) are developed and compared based on well defined benchmark problems, but their application to real-world problems is still challenging. In image processing, EA have been used to tune a particular image filter or in the design of filters themselves. But nowadays in digital cameras, the image sensor captures a raw image that is then processed by an Image Signal Processor (ISP) where several transformations or filters are sequentially applied in order to enhance the final picture. Each of these steps have several parameters and their tuning require lot of resources that are usually performed by human experts based on metrics to assess the quality of the final image. This can be considered as an expensive black-box optimization problem with many parameters and many quality metrics. In this paper, we investigate the use of EA in the context of ISP parameter tuning with the aim of raw image enhancement.