Home Services Image Processing

Image Processing

EM Photonics' image processing division explores cutting-edge algorithms for the improvement and enhancement of images and videos. In particular, we specialize in implementing advanced algorithms on a variety of platforms and systems. Our engineers develop hardware accelerated solutions capable of running complex image processing algorithms at real time speeds. These solutions are typically implemented on rugged, portable FPGA platforms or low-cost GPU accelerated desktop systems.

Among the algorithms we've accelerated are atmospheric compensation, super resolution, and compressive sampling. If you have a computationally intense image processing algorithm you'd like EM Photonics to investigate, do no hesitate to This e-mail address is being protected from spambots, you need JavaScript enabled to view it .

Atmospheric Compensation

When imaging over long distances (over 1 mile), atmospheric turbulence becomes the major limiting factor to image resolution, regardless of the quality of the optical systems being used. Researchers at the Lawrence Livermore National Laboratory have developed a bispectrum averaging speckle method to combat these undesirable effects.

EM Photonics engineers have accelerated the atmospheric compensation algorithm to run at real time speeds using FPGA and GPU systems. More information regarding our acceleration efforts and the bispectrum averaging algorithm be found at our ATCOM product information page.

Before Atmospheric Compensation :: Long range imaging before processing. Images courtesy of LLNL. After Atmospheric Compensation :: Long range imaging after processing. The once blurry faces and license plate are now distinguishable. Images courtesy of LLNL.

Super Resolution

Super resolution is an image processing technique that aims to accurately increase the resolution of an image. The algorithm works by analyzing a short sequence of low-resolution images to produce a single high-resolution image. If the frames in the sequence each offer a slightly different view of the objects being imaged, a large amount of information can be extracted about the objects. Using this information, advanced reconstruction algorithms can accurately produce a high resolution image of the objects being analyzed. Super resolution works best when recording a low-resolution video of a moving object. If the scene is still, no additional information can be extracted.

The mathematics behind super resolution are very complicated and require a large amount of processing power. EM Photonics researchers are currently investigating accelerated desktop and portable embedded solutions that would allow the super resolution algorithm to run at real time speeds.

Full Size Image :: A low-resolution image of a building on the University of Delaware campus. Nearest Neighbor Zoom :: Zooming in on a section using nearest neighbor enlargement methods. The bricks and sharp edges look very poor. Bicubic Zoom :: The same zoom, now done with the bicubic enlargement method. The edges look a bit smoother, but the small details of the plants and bricks are still poor. Super Resolution Zoom :: Super resolution methods have now produced a much clearer image enlargement. Specifically, the bricks in the building are much sharper.

Compressive Sampling

The number of samples required to reconstruct a signal is proportional to its bandwidth. Recently, it has been shown that acceptable reconstructions are possible from a reduced number of random samples, a process known as compressive sampling.