Dissertation
Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device

Local non-uniform moment invariants described image stitching technology research

Author LiCaiHui
Tutor LuQiYong
School Fudan University
Course Circuits and Systems
Keywords Image Stitching Image Registration Image fusion Descriptor IHS Greedy method
CLC TP391.41
Type Master's thesis
Year 2011
Downloads 72
Quotes 1
Download Dissertation

Panoramic image of the application of computer technology development and people 's living standards in the pursuit of continuous improvement has become increasingly widespread. Panoramic image can be panoramic cameras and multi-camera images collected by stitching two methods . The former obtained image quality is better, but the panoramic camera costs are expensive , the latter costs are lower, but the image quality is often not ideal. Image Stitching technology research purpose is to image quality and cost to find a compromise . Image Stitching technology includes two key technologies: image registration and image fusion, existing image stitching technology research focus lies optimized separately these two algorithms. According image mosaic of registration , this paper selected image registration algorithm is based on local feature points matching algorithm , it is because it arithmetic precision and complexity of the algorithm is not very complicated. For registration in three steps - feature point detection , feature point description and feature point matching , description becomes more important , so this paper the study of the descriptor is proposed on the basis of a local non-uniform moment invariants describe sub - NUCFMs, and used and proved to have a better describe the image capability. Then this paper put this descriptor applied to image registration , because this descriptor can better distinguish each feature point , thus enabling registration accuracy and complexity of two aspects of performance are improved. Finally , in order to image fusion does not involve too complex calculations , this paper based on IHS space transform fusion algorithm a little improvement , that the process of integration involves only the I and S component , then use greedy method calculate an optimal boundaries. Experiments show that this combination of greed France and IHS space transform fusion method from color transitions and boundary smoothing two aspects to improve the integration of the results obtained fusion image is basically a seamless and no obvious difference in color images.

Related Dissertations
More Dissertations