产生纹理“Tuned”模板的遗传算法

GENETIC ALGORITHM FOR PRODUCING TEXTURE "TUNED" MASKS

本文对现有产生纹理"Tuned"模板的方法进行了研究与分析,指出这种方法的缺点:解的随机性,井提出了改进意见,在模板的调整中引入变异操作,在分析的基础上,提出产生"Tuned"模板的新方法—遗传算法,使寻找最佳模板的随机性减小。通过航空影像的试验,与现有方法作对比,表明新方法的纹理分类效果有明显改善。此外,还对"Tuned"模板对纹理方向的自适应性和模板元素的对称性作了实验分析,得到一些有实用价值的结果。

Our discussion and analysis on existing method of producing texture 'Tuned' masks show that the shortcoming of the method is the randomness of solution. In this this pape an improved approach of producing 'Tuned' masks based on Genetic Algorithm is proposed, which add mutation operation in masks adjusting to reduce the randomness of finding the best mask. The new approach and the existing method are used individually on aerial images. The experiments show the texture classification using the new approach has better result. Experiments and analysis on the texture direction adaptability and element's symmetry of 'tuned' masks are also presented in this paper.