Abstract: Traditional Niche Genetic Algorithm（NGA） converges too slowly and is easy to trap in local extrema，thus it is modified in three aspects：Firstly，solution space is partitioned into multi-layers，and distance factors are forced to vary with layers.Secondly，difference value coding method is adopted to guarantee that this proposed algorithm converges more easily.Finally，the traditional neighborhood model is meliorated into a repeat-neighborhood model.Simulations demonstrate that this Adaptive Niche Genetic Algorithm（ANGA） performs better than the popular GA and NGA.PSNR of ANGA is 0.2~0.3 dB higher than other algorithms while the consumed time of ANGA is only 40%~50% of other algorithms’.

Key words: image restoration, genetic algorithm, niche algorithm, neighborhood model

摘要： 传统的小生境遗传算法收敛太慢，且容易陷入局部最优，对小生境算法做出以下三点改进：一是将解空间划分为多层区域，每层使用不同的距离因子；二是采用差值编码方式，使得算法更易收敛；三是使用伪并行加速算法，改进经典的邻居模型为镜像邻居模型。实验表明改进算法的PSNR比常用的遗传算法以及小生境算法高0.2~0.3 dB，且运算时间仅有它们的40%~50%。

关键词: 图像复原, 遗传算法, 小生境算法, 邻居模型