Higher order spatial transformations for motion analysis and modelling

In this paper we present two novel methods for obtaining the ST motion parameters of arbitrary shaped video objects. These methods use genetic algorithm (GA) optimisation to efficiently and robustly search the multidimensional space of possible solutions. An overview of ST and GA techniques, is provided and the proposed algorithms are described. Some results are given, comparing the performance of the proposed algorithms with a traditional multi-level search method and an MPEG-4 like shape-adaptive BMA. Finally some concluding remarks and proposals for further work, are given.