GIST develops retina-inspired camera sensor to sharpen noisy images

By performing part of the image processing directly in the camera sensor, this approach improves recognition accuracy.

“In-sensor image processing,” in which the sensor performs part of the image processing as it detects light, reduces the amount of data that must be sent to external processors and thus lowers power consumption and processing latency.

Using tungsten diselenide (WSe2), an ultrathin two-dimensional semiconductor, the team developed a transistor device that reproduces this retinal signal suppression function.

The team then used simulations to evaluate situations in which images become noisy due to rain, fog, and similar conditions.

However, when the developed image processing was first applied to the same noisy images before input, the accuracy rose to 97.3%.