Lead Inventor: Tianzhi Yang
Image Alignment Based on Algorithms Cannot Align Nonlinear Distortions
Image alignment or registration algorithms are used to match images or patterns in images. These algorithms are used in a wide range of applications, including the seamless stitching of partial images of a subject or scene, identifying and tracking objects in a video, and stabilizing videos obtained with jittery cameras. Constructing good image registration algorithms is challenging because of the difficulty of matching patterns or shapes in images with large deformations or discontinuities. Many existing algorithms can only align images with small rigid perturbations, cannot handle nonlinear distortions well, and sometimes can only identify suboptimal alignments.
Stochastic Technique Aligns Images with Large Nonlinear Distortions
The technology is a new method for accurately matching two sets of data points (or some subsets of those sets) based on their spatial locations. It uses a novel stochastic technique for constructing possible maps (i.e., potential matches) from one data set to another, a novel metric for assessing map quality, and simulated annealing to search for a globally optimal match. In comparison to less robust image alignment techniques, this technology is capable of aligning images that have undergone large nonlinear distortions. It can also align substructures of images with discontinuities, and can exploit various feature detection schemes to improve the matching of specific structures in images.
Applications:
This technology could be useful in applications that make use of image alignment, for example:
• Bioinformatics applications such as face recognition and fingerprint matching
• Security and military applications such as object tracking and automatic control of vehicles or aircraft
• Industrial applications such as the inspection of manufactured products
• Design applications such as 3D scene reconstruction
Advantages:
In contrast to many other data set alignment methods, this technology exhibits all of the following qualities:
• Accuracy: it can determine very close alignments between images.
• Robustness: it can align images with large deformations or distortions.
• Flexibility: it can recognize and align patterns that only constitute a portion of an imaged scene or subject.
• Adaptability: it can use a wide range of feature detectors to match images or patterns.
Patent Status: Copyright
Licensing Status: Available for Licensing and Sponsored Research Support