Sequential Image Registration for Astronomical Images

Abstract

Astronomical images are characterized by their smooth features, low level of Signal to Noise Ratio (SNR), and their extreme sensitivity to the motion of platform. Due to the low SNR, it is necessary to collect a large number of frames and consider the average. However, it is a common occurrence to have unregistered frames in the sequence. Frame registration using feature-based approach fails due to low contrast. Also, area-based approaches such as template matching and phase correlation methods, although accurate, suffer from computational inefficiency as a result of the large size and number of image frames in a sequence. This paper introduces a novel two-stage algorithm to accelerate the process of registration. The first stage projects the direction of movement as a cluster of parallel streaks and determines the angle of motion, using Linear Hough Transform. The next stage utilizes Normalized Cross Correlation only in the estimated direction to find the exact amount of displacement. Experimental results have been tabulated to illustrate superior computational efficiency of the proposed algorithm versus phase correlation, as well as robustness of the procedure in the presence of the noise.