Clients Case Studies Papers
Through still and video photography and sensors such as radar, infrared, sonar and sensors covering other parts of the electromagnetic spectrum, humans have extended their ability to "see" beyond that which is visible to the eye -- underwater, on the ground, from space and in space. The sensors are scanned to form two dimensional pictures, stereo images, and volumetric images. The processing of these sensor pictures and images to provide useful information free of noise is called image processing, image understanding or computer vision. Images are often very large, consisting of millions of bits and need to be compressed for storage and transmission.
LNK's experience in Visual Information Processing is in the areas of:
LNK has a long history of developing and applying innovative image technologies for commercial and governmental customers. Our work includes processing, analysis and compression of visible, infrared, multi-spectral and hyper-spectral data.
In the area of image processing and analysis we have developed image enhancement tools for the Army for use with infrared sensors to improve the visibility of man made objects in natural scenes. We have located and measured debris in jet engine oil systems using a 1000 frame per second processing system. We have detected and tracked individuals in real time video using very low cost hardware for a commercial application. Further work has been done in developing custom image processing libraries for special purpose DSP (TiC80) and parallel processors ( LNK_ImageLib for the CNAPS SIMD machine). LNK has worked on standards committees (VSIP) as well. LNK has also developed tools to analyze video to identify navigational landmarks for unmanned guided vehicles. We have also worked in the automatic classification of terrain for use in automated image database archives.
In the area of compression LNK has provided solutions in the compression of graphical grayscale images and we are presently working on hyperspectral image compression. Hyperspectral data sets are very large and are thus difficult to transmit and store. Our compression algorithm works from lossless ( ~2.7:1, AVIRIS dataset) to much higher compression ratios (lossy). A novel feature of our work is that the error introduced in the lossy modes can be constrained so that it is not concentrated at any location and the maximum absolute error can be guaranteed.
These tools have
been developed for workstations, PC's, DSP chips and parallel architectures
using Unix, NT, Windows and special purpose operating systems. Written in C/C++
with assembly(MMX) the applications are responsive and met critical timing needs.