HIGH ACCURACY SUB-PIXEL BASED CORRESPONDENCE POINT MATCHING BETWEEN TWO IMAGES

Abstract

Correspondence Matching is an important and necessary step in the Image Registration. As we know image registration has wide applications in the field of medical imaging, stereo vision and 3D measurement, remote sensing, geographic information system, weather forecasting, environment monitoring etc. The most important factor regarding image registration is, accuracy of corresponding point measurement. This work is intended to estimate the corresponding point with high accuracy at sub pixel accuracy level. In this work, we are investigating the minimum shifting between two image in terms of sub pixels, whereas images are shifted in micrometer range. We are using Area Based Matching, so computational complexity is more. Another focus of this work is to reduce the computational complexity for calculating the corresponding point between two images. To reduce computational complexity we are using 1D BLPOC (l3and Limited Phase Only Correlation), in which we search for corresponding point along the epipolar line only. So this algorithm has lower computational complexity in comparison to other existing algorithm. Additionally, we are using the different kind of spectral weighting functions to reduce the effect of noise and enhance the estimation accuracy. Finally we compared this algorithm with two existing algorithm, Normalized Gross Correlation and 2D FOG considering two factors, computational time and estimation accuracy. And then we have discussed about seine al)plicatioll of this algorithin in different area where high accuracy estimation is necessary.