Image Registration using Hampel and Modified Hampel M-estimators

摘要

Image registration (IR) is an image processing technique which is a basic process used to match two or more images of same scene captured with different sensors at different times from different viewpoints. This technique has a wide range of applications in medical imaging, computer vision, remote sensing, military automatic target recognition, and analyzing images and data from satellite images. Earlier, image processing researchers proposed variety of algorithms for IR. Sum of the squared differences (SSD), normalized cross-correlation coefficient (NCC), increment sign correlation coefficient (ISC) and selective correlation coefficient (SCC) are some of the well studied correlation-based methods for IR. It had been proved that these methods fail in presence of outliers. Outliers are nothing but, a perception that lies an anomalous separation from different qualities in an irregular example from a populace. To overcome outliers robust algorithmic methods are required. M-estimation correlation coefficient (MCC) method is such a robust image registration method. Hence, this paper presents the work on MCC method. In earlier, Huber M-estimator and Tukey M-estimator are used for IR. In this paper, Hampel M-estimator and Modified Hampel M-estimator are used for IR. To know the efficiency of these estimators for IR, comparison is done by computing correlation mask function (CMF) using Hampel and Modified Hampel's M-estimators. The CMF stifles the affect of outliers and robustify the IR algorithm. Efficacy of this process is presented in experimental results in terms of computational time and registration performance. Results showed that Hampel and Modified Hampel based MCC method works more efficiently than Huber and Tukey estimators based MCC.