Computer Vision News - August 2022

17 Glass Imaging little bit. No one’s thinking, wait, is there a better design? ” Glass discovered that the same real lens architecture camera design had been used for the last 150 years. It plans to change how the camera works, using anamorphic lenses and wide aspect ratio sensors on the hardware side, which are more akin to costly Hollywood systems. No one had thought of using them on a smartphone before. “ We capture ten times more light than a traditional smartphone, which means we get a higher signal-to-noise ratio , ” Tom explains. “We’re using image restoration approaches, such as blind deconvolution and super-resolution, to recover the possible detail according to the trade-offs we make in the lenses. Another interesting area is the end-to-end optimization of the entire system. We can model the optical system using our deep learning approach in an integrated fashion, optimizing the lens parameters along with the neural network doing the image restoration .” Further benefits include single-shot HDR – with larger pixels and a high dynamic range; extreme low-light imaging, so night photography and freeze motion; and a natural depth of field. “ When we created the portrait mode, we wanted to emulate the look from SLR cameras where you can blur out the backgrounds, ” Tom recalls. “ Software has done a great job there, but you still get artifacts around the borders if you have someone’s hair against a complicated background. With our hardware optics, we can get a natural shallow depth of field. We can apply that to video as well. Apple released the cinematic mode not so long gives it and trying to improve it but also thinking about how the optics can be designed differently. An image may be distorted, degraded, or corrupted, but the AI algorithms work to get it perfect at the end of the pipeline. That whole system working together creates the end image, and the neural networks are trained with knowledge of the optical design to do that all in one go. How has Glass managed to get ahead of the crowd with many big players circling in this field? “ Other companies are trying todo things like this, but their teams are often disconnected across hardware and software, and it requires deep integration and a mindset of trying things that are quite different from what’s done currently, ” Tom responds. “ Innovation in the big companies is somewhat incremental – let’s take last year’s design and improve the lens, the sensor, the algorithm, and the processor a Render showing the difference between a Glass Imaging anamorphic camera module and a regular smartphone camera module; note the larger area and shape of the former.