Apply online to license this technology
One approach for optical target recognition, usually for viewing by a person, fuses multiple target images with respect to a reference image of interest, often merely the first target of many target images. The goal is to create a fused image that is optimally “information rich” to the viewer and to do so at the lowest computational cost.
The reference image is usually an original optical image produced by a digital camera. Modern military systems may include a variety of additional sensor devices able to produce images based not only on multiple EM radiation frequencies, so-called multi-spectral images that include frequencies below and above the normal visual range but also based on other than EM radiation. A particular problem with these systems is computation cost, particularly computational speed. Further complicating this process is that an original reference image may be compromised by having too little or too much light, reduced contrast or size, and cluttered with other objects that confuse a user, as well as other possible image defects.
The Air Force is addressing this problem with an image registration and fusing solution that is less computationally intense. The process has two primary components. First, rotating, translating and scaling each target image to register (transforming different sets of data into one coordinate system) them against a reference image and, second, determining weighting factors for each registered target image to select those images which add value to a final target image. The first component determines optimal rotational and translation matrices by a least squares measure using singular value decomposition. The second component determines weighting factors using correlations and statistical signal analysis techniques.
- Determines the best alignment of target images with respect to a reference image
- Selects which candidate images add value to the final fused image
- US patent 8,768,101 available for license