Stephen Welsh and co-workers in the UK have developed a computational imaging system that can produce full-colour, two-dimensional images of three-dimensional scenes containing multiple objects within a few seconds using only single-pixel photodetectors (Opt. Express 21, 23068–23074; 2013).

Instead of using a monochromatic laser, the researchers employed a digital light projector as a light source to provide spatially incoherent, binary-structured illumination when imaging the colour scene. The digital light projector contains a digital micromirror device and three coloured (namely, red, green and blue) light-emitting diodes. For each incident pattern, the reflected light from an object was directed onto a composite dichroic beamsplitter by a large collection lens. The dichroic beamsplitter was used for spectrally separating white light into three outputs (red, green and blue) and three unfiltered single-pixel photodiodes were used to measure the reflected intensity for each incident pattern generated by the digital light projector. The three photodetector signals were digitized using an analogue-to-signal converter and stored on a computer. They were subsequently used for image reconstruction by applying an appropriate algorithm.

In the work, models of dinosaurs with dimensions of 20 cm × 10 cm and positioned 1 m from the digital light projector were used as the colour scene. They were illuminated with approximately 1,300 different light patterns each second. To construct a full-colour image, these patterns were correlated with the backscattered light measured by the three single-pixel photodetectors. The frame rate was 60 Hz.

Credit: © 2013 OSA

To overcome the various latencies in the computer graphics pipeline, the team added a synchronization process during data acquisition. A flash bit-plane in the sequence of images was achieved by turning all mirrors to the on position and then to the off position. Additionally, every 50th frame was used as a synchronization frame. This allowed robust correlation of the frames to the measured signals and ensured that only data between two matching synchronization frames were used.

The researchers also employed the differential detection method to reduce the influence of noise (such as fluctuations in the ambient light) on the measurement. A compressive sensing technique was used to reduce the number of measurements required for faithful image reconstruction. Even for a highly ill-conditioned system of equations, it was possible to produce high-quality reconstructed images by exploiting the sparsity in natural images. The team found that the image quality was significantly improved when the number of measurements was below the Nyquist limit. They claim that the technique can be readily extended for imaging applications at non-visible wavebands.