期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2017
卷号：95
期号：14
页码：3146
出版社：Journal of Theoretical and Applied
摘要：Automatic SAR image alignment or registration is one of the fundamental preprocessing operations in several applications of computer vision, image processing, pattern recognition etc. Advances in feature extraction algorithms have led to the development of efficient alignment algorithms which offer invariance to deformations like scale, rotation, view angle etc. To further improve the performance of standard SAR image alignment, an iterative approach generating multiple synthetic views, known as view synthesis has been employed. State of art detectors like SIFT, SURF, Hessian Affine and MSER when used with view synthesis have been proved to be invariant to a wider range of deformations, however at the cost of additional memory and time to be spent on generation of views and feature extraction across all the views. Hence we studied if the characteristics of source and target SAR images to be aligned can be used to provide some knowledge and guide SAR image alignment approach. This paper focuses on identifying the possibility of aligning two SAR images and also predicts suitable alignment approach by building a predictive data mining model using classification based algorithms. The dataset of 540 Terra SAR X band images is tested and applied on various classification algorithms such as Naive Bayes, SVM, and J48 using attributes computed from Bag of Words model applied on feature detector and descriptor.
关键词：Image Alignment; Feature Detection; Feature Description; View Synthesis; Bag of Visual Words; SVM
