Global motion estimation in frequency and spatial domain

We propose a fast and robust global motion estimation algorithm based on two-stage coarse-to-fine refinement strategy, which is capable of measuring large motions. Six-parameter affine motion model has been used. Coarse estimation is done in frequency domain using polar, log-polar or log-log sampling of Fourier magnitude spectrum of sub-sampled image. Fourier magnitude spectrum, as translation invariant domain, allows for determination of 4 parameters independent from translation. Sampling scheme is adaptively selected based on past motion pattern. Adaptive selection of sampling scheme insures best trade-off between accuracy and maximum range of motion measurements for current motion pattern. Refinement stage consists of RANSAC based model fitting to motion vectors of randomly selected high-activity blocks, and hence is robust to outliers. Motion vector of blocks is measured using phase correlation, which offers two advantages in this context: sub-pixel accuracy without significant computational overhead, and if a particular block consists of background as well as foreground pixels, both motions are simultaneously measured; as opposed to other methods like block matching which rely on SAD or SSD error metrics and hence fail in such situations. Due to its hardware-friedly nature proposed algorithm holds potential for real-time GME even for television images.