多源遥感图像局部细节分层匹配方法优化仿真

摘要

多源遥感图像局部细节分层匹配优劣对目标物体识别、检测和跟踪等众多领域应用具有重要参考价值.针对当前方法获得的多源遥感图像局部细节特征描述向量维数过高导致运算效率较低,匹配性能、抗亮度变化性能和抗旋转性能较差的问题.提出一种基于改进SIFT算法与遗传算法的匹配优化方法,建立了多源遥感图像的高斯差分金字塔,确定特征关键点的位置和尺度信息,对对比度较低和边缘响应点不稳定的关键点进行去除;利用关键点的梯度和方向分布特性,为关键点分配梯度和方向,同时对特征描述向量进行降维和归一化处理,生成新的特征描述向量,克服了当前方法获得的特征描述向量维度过高影响匹配结果的弊端;采用遗传算法对其进行自然编码、适应度函数选择、交叉、变异操作,同时设计了一个停机准则,使得遗传算法进行匹配操作时能够兼顾效率和精度.仿真结果证明,所提方法运算速度更快,具有较好的匹配性能,且具有较好的抗旋转性能和抗亮度变化性能,实现了优化.%This paper proposes a matching optimization method based on improved SIFT algorithm and genetic algorithm.The pyramid of difference of Gaussians of multi-source remote sensing image is established,and then the location and scale information of feature key point are determined.The key points with low contrast ratio and unstable edge response point were removed.Moreover,gradient and direction distribution of key point are used to allocate gradient and direction for key point.At the same time,dimeusionality is reduced and feature description vectors are normalized,so as to generate new feature description vectors.Finally,the disadvantage that dimensionality of feature descriptor dimension obtained by current method is too high to influence the matching result is overcome.The genetic algorithm is used for natural coding,selection of fitness function,crossover and mutation.At the same time,a stop criterion is designed.Thus,when performing match operation,the genetic algorithm strikes a balance between the efficiency and accuracy.Simulation results show that the proposed method has high operational speed and good matching performance,and it has good rotation-resistant performance and anti-brightness variation.