英文资料快速的基于相位的多式联运图像数据配准Word文件下载.docx

1、Received 19 July 2008; revised 14 October 2008;accepted 18 October 2008.Available online 5 November 2008. AbstractAn interesting problem in pattern recognition is that of image registration, which plays an important role in many vision-based recognition and motion analysis applications. Of particula

2、r interest among registration problems are multimodal registration problems, where the images exist in different feature spaces. State-of-the-art phased-based approaches to multimodal image registration methods have provided good accuracy but have high computational cost. This paper presents a fast

3、phase-based approach to registering multimodal images for the purpose of initial coarse-grained registration. This is accomplished by simultaneously performing both globally exhaustive dynamic phase sub-cloud matching and polynomial feature space transformation estimation in the frequency domain usi

4、ng the fast Fourier transform （FFT）. A multiscale phase-based feature extraction method is proposed that determines both the location and size of the dynamic sub-clouds being extracted. A simple outlier pruning based on resampling is used to remove false keypoint matches. The proposed phase-based ap

5、proach to registration can be performed very efficiently without the need for initial estimates or equivalent keypoints from both images. Experimental results show that the proposed method can provide accuracies comparable to the state-of-the-art phase-based image registration methods for the purpos

6、e of initial coarse-grained registration while being much faster to compute.Keywords: Image registration; Phase; Fast Fourier transform; Multimodal; Keypoints; Dynamic sub-cloudsArticle Outline1. Introduction 2. Multimodal registration problem 3. Previous work 4. Proposed registration algorithm 4.1.

7、 Keypoint detection and sub-cloud size estimation 4.2. Phase sub-cloud extraction 4.3. Simultaneous sub-cloud matching and feature space transformation estimation 4.4. Solving the simultaneous matching and feature space transformation estimation problem in the frequency domain 4.5. Outlier pruning t

8、hrough resampling 4.6. Algorithm outline5. Computational complexity analysis 6. Experimental results 7. Conclusions and future work Acknowledgements References1. IntroductionImage registration is the process of matching points in one image to their corresponding points in another image. The problem

9、of image registration plays a very important role in many visual and object recognition and motion analysis applications. Some of these applications include visual motion estimation 1 and 2, vision-based content-based retrieval 3 and 4, image registration 5, 6, 7 and 9, and biometric authentication

10、10. In the best case scenario, the images exist at the same scale, in the same orientation, as well as represented in the same feature space. However, this is not the case in most real-world applications. There are many situations where the images exist in different feature spaces. This particular p

11、roblem will be referred to as the multimodal registration problem and is a particularly difficult problem to solve. Examples of this problem in real-world situations include medical image registration and tracking of MRI/CT/PET data 11 and building modeling and visualization using LIDAR and optical

12、data 12 and 13. There are several important issues that make multimodal registration a difficult problem to solve. First, many registration algorithms require that equivalent keypoints be identified within each image. However, given the differences between feature spaces in which the images exist, i

13、t is often a very difficult task. The significant differences between feature spaces also make it impractical to perform direct intensity matching between the two images. In recent years, an effective approach to multimodal registration has been proposed that utilizes local phase 14 and 15. This sta

14、te-of-the-art approach evaluates the mutual information between the local phase of two images to determine the optimal alignment and has been shown to be very effective at matching multimodal medical image data, outperforming existing multimodal registration methods 14 and 15. However, this approach

15、 is computationally expensive （O（N6） for the mutual information evaluation process）. As such, a registration method that is able to take advantage of local phase information to determine point correspondences between images while being computationally efficient is highly desired for the purpose of initial coarse-grained registration.The main contribution of this paper is fast phase-based registration algorithm for aligning multimodal images. The proposed method is designed to provide a fast alternative to