BICV in a Nutshell
What defines a biologically-inspired approach to algorithm design ? There are many facets (power consumption, architecture, algorithms), but our inspiration is largely fed by the way that biological neurons encode visual data. For example, two interesting and relevant phenomena of biological sensory processing are: a) complex visual neurons of V1 and b) population encoding. At a systems level, we have also used approaches that are similar to those now appearing in deep learning, such as denoising visual auto-encoders [Bharath & Ng, 2005]. The spin-out Cortexica Vision Systems used this technology to provide commercial visual search using cloud-based GPUs as early as 2010. You can find out more about the original technology here. Cortexica is now an independent company, and so you should contact them for access to this technology.
Some of our research, in collaboration with other academic research groups or companies, is based around parallel computation. For example, we have simulated populations of visual neurons using Graphical Processing Units (GPUs). We have also explored the use of analogue networks for processing visual data [Ip et al, 2010]. Recently we have run a series of highly popular CUDA taining workshops; a recent Deep Learning for Computer Vision workshop was heavily oversubscribed.
We apply our research to the analysis of visual data from a wide variety of sources, and with widely differing image statistics, including microscopy images [Peiffer et al, 2013], natural scenes and images of man-made objects and environments [Alexiou & Bharath, 2010], [Rivera-Rubio et al, 2015]. We have also been looking at images from historical collections and camera simulations from robotic environments.