Font Size: a A A

Research On The Key Algorithms Of Image Registration

Posted on:2016-08-11Degree:MasterType:Thesis
Country:ChinaCandidate:H L HuoFull Text:PDF
GTID:2298330452964894Subject:Electronic Science and Technology
Abstract/Summary:PDF Full Text Request
With the development of science and technology and emergence of new sensors,people’s ability to acquire image data has been improved continuously and the imageinformation has been increased. Before conducting the multi-source co work with thesemulti-source image like data fusion, target change detection and target recognition, imageregistration is the key step. This paper aimed at researching the multi-source imageregistration technology with scaling, rotation, translation and other complex transformexisted, based on the Fourier Merlin algorithm, combined with the image of the cornerfeature, edge feature, structure feature and other related extraction algorithm, registered thehomologous and heterologous image separately and evaluated the registration resultseffectively.In the homologous image registration, this paper used the frequency domain iterativeimage registration algorithm based on local region extraction, effectively used the fastcharacteristics of SURF feature points algorithm and Fourier Merlin algorithm, accuratelyestimated the translation rotation and scale parameters between images respectively. Duringthe registration period, it took a coarse to fine the two step iterative method, not onlyavoided too much calculation and the demand of image features, but also improved theaccuracy of image registration effectively.Aiming at the problem of multi sensor image registration, after further research ofheterologous image registration algorithm based on edge significantly, this paper presenteda method which combined phase congruency edge extraction and frequency domain FourierMerlin algorithm. First, conducted the image edge extraction using the feature that at theedge of image signal the Fourier component had maximum phase consistency. Thenconducted the parameter estimation according to the Fourier Merlin algorithm, ultimatelyachieved the registration parameters with high accuracy.In view of the above methods can not solve the problem of multi-sensor imagesregistration with scaling transformation parameters, this paper presented a method based onthe combination of image structure characteristics and frequency domain Fourier Merlin transform. First,extracted the closed connected region which was suitable for registration,according to the local structure of image. Then conducted the parameter estimationcombined with the Fourier Merlin algorithm to obtain the registration parameters. Thismethod is also mainly based on Fourier transform, using the local image structureinformation instead of all the information, reduced the calculation amount, shorten thecalculation time and fully embodied the advantage of speed.
Keywords/Search Tags:Image registration, SURF feature points, Fourier Merlin, Local regionextraction, Phase correlation, Edge feature, architectural feature
PDF Full Text Request
Related items