一种基于改进的遗传算法的块匹配运动估计方法

块匹配方法(Block Matching Algorithm,简称BMA)是目前广泛使用的运动估计方法,但该方法的最大缺点是容易陷于局部最优,这主要是由搜索模式决定的。而遗传算法(Genetic Algorithm,简称GA)是一种具有广泛适应性的全局最优的搜索算法。将块匹配方法的局域性搜索与遗传算法的全局性搜索结合起来,本文提出了一种基于改进的遗传算法的块匹配运动估计方法。实验证明,该方法的平均绝对误差(MAE)接近全搜索(FSS),优于三步法(TSS),而运算量相对较低,接近三步法。

A Modified Genetic Algorithm Based Block Matching Motion Estimation Method

The Block Matching Algorithm (BMA) is currently widely used in Motion Estimation, but it is suboptimum and susceptible to be trapped into local optimum due to its specific searching pattern. While, Genetic Algorithm (GA) is a global optimum searching method used in many fields which require global optimum from large data. This paper combines BMA definite local searching with GA elective global searching and proposes a block matching algorithm based on a modified GA. The simulations show that the Mean Absolute Error (MAE) performance of this new algorithm is similar to that of FSS, better than TSS, while the computation complexity of it is lower than that of FSS and similar to TSS.