An FFT-Based Technique and Best-first Search for Image Registration

The image registration is a fundamental task in image processing used to match two or more pictures taken, for example, at different time, from different sensors, or from different viewpoints. One of the major challenges related to image registration is the estimation of large motion, when input images contain small overlapped area. Common image registration using the search algorithm can accurately finds large motion but requires high computation cost due to large search space. Fourier-based technique is an alternative approach since it can rapidly achieve the registration results through its FFT algorithm. However, only Fourier-based technique cannot produce the correct results in the case of large translation. Thus, this paper presents a Fourier-based technique cooperated with best-first search algorithm to analyze the correct translation between two input images. The Fourier-based technique is used to estimate the candidate translations to decrease searching space while best-first search algorithm is used to further search for the correct translation. The proposed technique can estimate large translations, scalings, and rotations in images by an extension of well-known phase correlation technique. The experimental results using various image details show the accuracy of the proposed technique to detect large translations compared to the other techniques in frequency domain.