Recursive displacement field estimation with application to image sequence processing.

摘要

In this dissertation the problem of estimating the displacement vector field (DVF) from a noisy dynamic image sequence is addressed. A recursive model-based estimator is developed which provides very accurate estimates of the true motion field. To model the DVF, a nonstationary vector field model called the Vector Coupled Gauss-Markov (VCGM) model is developed. The VCGM model consists of two levels, an upper level, which is made up of several submodels with various characteristics, and a lower level or line process which governs the transitions between the submodels. A detailed characterization of the line process is provided through the use of an energy function and Gibbs distribution. The VCGM model is well suited for estimating the DVF, since the resulting estimates preserve the boundaries between the differently moving areas in an image sequence. The resulting estimator has an architecture similar to a Kalman filter, followed by a decision criterion for choosing the appropriate line process. Several experiments using synthetic and real image sequences serve to highlight the superior performance of the proposed algorithm with respect to prediction error, interpolation error and robustness to noise.; The DVF estimation algorithm proposed in this dissertation is not dependent on any particular application. Due to the accuracy of the provided displacement estimates it is therefore suitable for any number of applications or analysis tasks. The approach of separating the motion estimation task from the subsequent image processing task is widely used today. In this dissertation, the foundation is laid for a new paradigm in image sequence processing, in which the estimation of motion and the particular processing task are performed simultaneously. Such an approach is developed for the image sequence processing tasks of image sequence restoration and image sequence compression. Specifically, under this new paradigm a recursive model-based MAP estimator is developed, that simultaneously estimates the displacement vector and intensity fields from a noisy-blurred image sequence. By simultaneously estimating these two fields, a link between the two estimators is established. Through this link the DVF estimates and its corresponding accuracy information are shared with the other intensity estimator, and vice versa. Also developed is a novel coding technique that makes use of the nonstationary characteristics of the DVF for encoding motion information. The proposed motion estimator is used as a means of producing displacement vectors which may be regenerated at the decoder with a coarsely quantized error term created in the encoder. In both cases, comparisons are made to other techniques that separately estimate the motion, treating it as a preprocessing step.