Maximum likelihood image motion estimation with application to image sequence filtering and restoration

This dissertation deals with the problem of image motion estimation from degraded image sequences, and applies motion estimation to image sequence filtering and restoration. Three approaches for image motion estimation using the maximum likelihood (L) principle are developed, presented and evaluated. The first two methods estimate image motion in noisy situations. The third method assumes that the image sequence is corrupted by both blur and noise. The first method processes several frames of a noisy image sequence jointly. The relative motions from each frame to the reference frame are estimated simultaneously. The reference frame is also filtered in the linear minimum mean square (LMMSE) sense during the process of motion estimation. The second method uses the Expectation-Maximization (EM) technique to estimate image motion iteratively from a pair of consecutive noisy frames. The results of motion estimation are used to filter image sequences. The third method estimates the motion from two consecutive frames in noisy and blurred environments. The estimated motion is applied to image sequence restoration. Simulation results show that all three methods provide reliable motion estimation from degraded image sequences. Applications of motion estimation to image sequence filtering and restoration show improvement visually and quantitatively.