期刊名称：International Journal of Computer Trends and Technology
电子版ISSN：2231-2803
出版年度：2013
卷号：6
期号：3
出版社：Seventh Sense Research Group
摘要：Image registration based classification is among the important image processing procedures in medical imaging and remote sensing, it has been developed and studied for a long time. Complex image registration issue arising as the dependencies between intensities of images to remain registered are not spatially homogeneous. However, as yet, it is still rare to locate a reliable, robust, and automatic image registration method, and most existing image registration methods are intended for particular application. At present, high resolution remote sensing or medical images have used it more convenient for people to study, however, in addition they bring some challenges regarding the traditional research methods. In terms of image register, there are some problems with using existing image registration techniques for high quality images, namely (a) precisely locating control points isn't as simple as with moderate resolution images; (b) manually picking the large number of control points necessary for precise registration is tedious and time consuming; (c) high data volume will adversely influence the processing speed within the image registration; and (d) local geometric distortion can not be removed thoroughly using traditional image registration methods even with enough control points. In accordance to these reasons, the demand for an image registration approach that could resolve those issues is need to improve. We present Local Naive Bayes Nearest Neighbor, an improvement onto the NB image classification algorithm that increases classification accuracy and improves its ability to scale to high object classes. The important observation may be that only the classes represented in the local neighborhood regarding a descriptor contribute significantly and reliably on their posterior probability estimates. Alternatively to maintaining a separate search structure for each individual class, we merge all the reference data together into one search structure, providing fast recognition of a descriptor’s local neighborhood. This proposed approach improves classification accuracy when we ignore adjustments to the more distant classes and shows that the run time grows with the log of the number of classes instead of linearly in the number of classes as did the original. In association with classification technique a log polar registration module is introduced to improve arbitrary rotation angles and a wide range of scale changes. This serves to furnish a good early estimation for the optimizationbased affine registration stage. This proposed work will overcome the existing deficiencies in terms of accuracy and experimental results.
