|
|
|
“Segmentation and Motion Estimation of Noisy Image Sequences”
by Dimitrios Spiridon Kalivas
February 1990
Two of the most important problems in image sequence are the segmentation of image frames into frames into moving parts. In this dissertation, we describe a method of solving these two problems. Our algorithm has two parts: segmentation and 2-D motion estimation. These two parts are interactively connected in a mutually beneficial way.
The segmentation us performed by a statistical boundary estimation algorithm, which is based on the Minimum Mean Square Estimation criterion. The noise is assumed to be additive. The algorithm requires the a priori knowledge of the region where the boundary is expected to lie. This region is estimated for the rest of the frames. The algorithm does not use any a priori information about the shape of the objects.
The 2-D motion estimation is performed by a region matching algorithm. This algorithm estimates accurately the linear 2-D motion and gives a very good linear approximation in the case of a nonlinear 2-D motion. It requires segmented images which are provided by the segmentation part.
The performance of each part and the combined algorithm is examined using computer generated and real images corrupted by additive white Gaussian noise. The algorithm performs very well in a very noisy environment. The boundaries are estimated very accurately. The motion parameters estimates are approximate but robust.
Finally, the application of the combined algorithm in the image sequence enhancement problem is examined. A motion compensated, edge preserving image sequence enhancement algorithm is presented and its performance is examined using computer generated and real images. The results are very satisfactory and they verify the efficiency of the combined segmentation and motion estimation algorithm.