An Automated Parallel Image Registration Technique Based on the Correlation of Wavelet Features
ABSTRACT With the increasing importance of multiple multiplatform remote sensing missions, fast and automatic integration of digital data from disparate sources has become critical to the success of these endeavors. Our work utilizes maxima of wavelet coefficients to form the basic features of a correlation-based automatic registration algorithm. Our wavelet-based registration algorithm is tested successfully with data from the National Oceanic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) and the Landsat Thematic Mapper (TM), which differ by translation and/or rotation. By the choice of high-frequency wavelet features, this method is similar to an edge-based correlation method, but by exploiting the multiresolution nature of a wavelet decomposition, our method achieves higher computational speeds for comparable accuracies. This algorithm has been implemented on a single-instruction multiple-data (SIMD) massively parallel computer, the MasPar MP-2, as well as on the CrayT3D, the Cray T3E, and a Beowulf cluster of Pentium workstations.
- SourceAvailable from: Jiayi Ma
IEEE Transactions on Geoscience and Remote Sensing 06/2015; DOI:10.1109/TGRS.2015.2441954 · 3.51 Impact Factor
- "The main idea of these methods is to compute the similarities of window pairs in two images, and consider the one with the largest similarity as a correspondence. In remote sensing applications, a correlation-like method utilizing maxima of wavelet coefficients has been developed for automatic registration . The correlation-like methods suffer from some drawbacks such as the flatness of the similarity measure in textureless regions and high computational complexity. "
[Show abstract] [Hide abstract]
- "Wavelet decomposition of the images was recommended for the pyramidal approach due to its inherent multi-resolution character. Le Moigne et al. (2002) utilized maxima of wavelet coefficients as the basic features and proposed a correlation-based automatic registration algorithm, which achieved higher computational speeds for comparable accuracies. Recently, several registration algorithms that combines wavelet-based pyramid with other similarity measures were proposed in Cole-Rhodes et al. (2003) and Zavorin and Le Moigne (2003, 2005). "
ABSTRACT: This paper proposes a multiscale registration technique using robust Scale Invariant Feature Transform (SIFT) features in Steerable-Domain, which can deal with the large variations of scale, rotation and illumination between images. First, a new robust SIFT descriptor is presented, which is invariant under affine transformation. Then, an adaptive similarity measure is developed according to the robust SIFT descriptor and the adaptive normalized cross correlation of feature point’s neighborhood. Finally, the corresponding feature points can be determined by the adaptive similarity measure in Steerable-Domain of the two input images, and the final refined transformation parameters determined by using gradual optimization are adopted to achieve the registration results. Quantitative comparisons of our algorithm with the related methods show a significant improvement in the presence of large scale, rotation changes, and illumination contrast. The effectiveness of the proposed method is demonstrated by the experimental results.World Pumps 12/2011; 14(2). DOI:10.1016/j.ejrs.2011.08.001
Journal of multimedia 06/2011; 6:236-243. DOI:10.4304/jmm.6.3.236-243
- "Hartkens et al  introduced features information into Voxel-based registration algorithms in order to incorporate higherlevel information together to produce consistent correspondences between images. Since feature-based control point extraction methods rely on the extraction and matching of prominent features to be the control points, these methods are not always capable of finding enough evenly distributed features as the control points for each local deformed region, especially for images that lack structures or patterns . In addition, feature extraction and matching are often difficult due to the dominance of ambiguous features. "