An adaptive motion estimation algorithm based on evolution strategies with correlated mutations

Based on evolution strategies (ESs) with correlated mutations, a novel algorithm - adaptively correlated ES motion estimation (ACESME) is presented. ESs consider the evolution progress on the phenotype level. In contrast, genetic algorithms focus on heredity genetic mechanism on the chromosomes level. The mutation operation in ESs accords with the normal distribution law. In the ACESME algorithm, the (μ, τ)-ES algorithm with correlated mutations is adopted to block motion estimation. In this algorithm, the motion direction factor participates in motion vector computing as a variable for the first time and affects the whole search process, neither just being an implicit factor nor a predictive measure. The adaptive schemes are advanced in the step length control and population sizing. Experimental results demonstrate that this algorithm has similar performance to that of the full-search (FS) algorithm. Furthermore, owing to the inherent parallelism and low complexity of ESs, ACESME is applicable for VLSI implementation.